The GeForce RTX 5080 is set to arrive in the next few days, and performance benchmark leaks have started appearing. A Redditor has unearthed Blender results for the GeForce RTX 5080, providing a glimpse of its capabilities.

According to benchmark results, the Nvidia GeForce RTX 5080 has a median score of 9077.3 points across two results, which is only a little over 700 points or 8% above the previous generation RTX 4080 Super. Compared to the RTX 4080, the RTX 5080 delivered up to 10% higher performance in Blender.

The numbers may look disappointing, especially as previously leaked benchmarks showed a 22% uplift over the RTX 4080. However, the results were from Geekbench, a questionable benchmark for evaluating graphics card performance. Many would argue that Blender is a more solid benchmark. The RTX 5090 performs about 25% better than the RTX 4090. It'll be interesting to see how the RTX 5080 stacks up to its predecessor, as the gap in actual gaming FPS you get with the RTX 5080 versus the RTX 4080 Super and RTX 4080 might even be smaller.

Still, we will have to wait until Nvidia lifts the embargo on the RTX 5080 on January 29 until we see real-world results for this GPU. We still have our fingers crossed that it will deliver a performance uplift that will make it a worthy upgrade over the previous generation GPU and a viable option for gamers who don’t want to spend $2,000 on a high-end graphics card.

If these leaked Blender benchmarks would track with the real-world FPS that the RTX 5080 would deliver, then it makes no sense to spend over a thousand dollars just to get between 8% to 10% better performance from your RTX 4080-series GPU. But if you’re coming from an RTX 3080 Ti, the near doubling of performance might make it a good upgrade.

You might have to be patient if you plan to upgrade at launch, though, as the RTX 5090 currently has limited stocks. Some scalpers are even taking advantage of this situation and selling the rights to buy the $2,000 GPU for over three times the MSRP. So, if Nvidia doesn’t fix the situation with the RTX 5080, you can expect the same, with the card selling for way over the MSRP.

Nvidia’s upcoming RTX 5090 and RTX 5090D graphics cards have appeared in the Blender benchmark database, demonstrating decently higher performance than the outgoing RTX 4090 (via VideoCardz).

The RTX 5090 was tested in Blender 3.6.0, while the RTX 5090D used version 4.3.0, the latest. The Blender benchmark database doesn’t offer any details about specific runs, but we know the exact scores since there’s only one result for each card. However, it would have been nice to know details about other components like the CPU and RAM; the database’s search refinement options indicate that the PCs were both running Windows.

As for why these cards are even in the database before launch, it is a mystery, but since Blender will automatically upload benchmark results if it has internet access, there’s a good chance this was an accident. We do at least know that these results were probably from two different computers since the Blender versions differ between the two cards; plus, the RTX 5090D is a China-exclusive GPU, while the RTX 5090 isn’t legal to export to the country, so it would be weird for someone to have both at this time.

Compared to their direct predecessors, the two graphics cards perform decently better. The RTX 5090 is roughly 36% faster than the RTX 4090, while the RTX 5090D beats the RTX 4090D by 40%. However, take these numbers with a grain of salt, not only because they’re leaked but also because the sample size for all these GPUs, except the 4090, is relatively small. Even if all the data is legit, having so few results means the numbers aren’t all that definitive.

Still, both results demonstrated the same uplift, even though they were almost certainly derived from different PCs. That lends a significant amount of credibility that the data is both legitimate and reflective of the cards’ actual performance since these scores happening by chance doesn’t seem very likely.

A ~40% performance improvement over the previous generation seems alright on paper, but it’s a far cry from what we saw with the 40 series compared to the 30 series. Across several versions of Blender, the RTX 4090 is roughly twice as fast as the 3090, according to the Blender benchmark database. The RTX 3090 was similarly twice as fast as the RTX 2080 Ti.

It’s hard to come to a firm conclusion based on just two benchmarks, but if the results are accurate, then the RTX 50 series' performance might not be awe-inspiring. That aligns with what we see on paper, as the RTX 5090 only has about 33% more CUDA cores than the RTX 4090, a significant factor in overall performance. It’s already clear that Nvidia is relying on AI to achieve its lofty performance claims for the RTX 5070, and the story may be the same for the RTX 5090.

A new entry at Blender Open Data sees AMD's upcoming Ryzen 7 9800X3D flex its muscles - beating the last generation and even its Ryzen 9000 non-X3D counterpart by a fair margin, per HXL on X. Evidently, the Ryzen 9000X3D CPUs will cater to more than gamers this time, as AMD's second-generation V-Cache technology promises higher clock speeds and better thermals owing to redesigned CPU packaging.

Blender Open Data aggregates the performance of different CPUs in the Blender V4.2.0 benchmark in a user-friendly web interface. While the benchmark does not list the test bench, the CPU's specifications, or even the operating conditions, it can be helpful to determine the processor's productivity potential. However, we must be cautious with results like these - it is a sample of one, and some pranksters like to fake this kind of data.

The Ryzen 7 9800X3D processor is expected to host eight cores, sixteen threads, and 104MB of total cache. According to rumors, the CPU features a boost clock of 5.2 GHz and DDR5-6000 memory with support for DDR5-8000+ through overclocking. Anyhow, Blender sees the new 9800X3D amass 323.76 points based on one entry —putting it a whopping 26% faster than the last generation.

Interestingly, the Ryzen 7 7700X was faster than the 7800X3D last generation, since AMD's X3D chips are limited by a combination of slow clocks and thermals. However, the 9800X3D easily outclasses the Ryzen 7 9700X this time by almost 11%. What's more bizarre is that the 9700X has a higher boost clock at 5.5 GHz. Gaming was always a strong suit of these X3D CPUs but It almost certainly feels like AMD's redesign allows the 9800X3D to keep up with and even beat its non-X3D counterparts in productivity, too.

Many enthusiasts consider AMD's V-Cache equipped chips to be a "One trick pony" - they shine solely in gaming but lackluster single-core clock speeds cripple performance in productivity-centric applications. Much of that seems to be resolved with Zen 5X3D but ultimately, it is a circular argument at the end of the day. Enthusiasts seeking the best of both worlds require chips with high-core counts and multi-CCD Ryzen CPUs with V-Cache will most likely be prone to scheduling issues much like the last generation.

AMD is set to announce the Ryzen 7 9800X3D on November 7. Retailers have put up early listings for these CPUs ranging between $484 and $525 - but the MSRP is subject to change. Expect an uptick in leaks and rumors as we inch closer to the launch.

Intel's upcoming Core Ultra 9 285K, the flagship of the Core Ultra 200S lineup, has been spotted in CPU-Z (via momomo_us) and Blender (via HXL). The initial performance is somewhat disappointing; however, that may be attributed to the test bench used and the operating conditions, at least in CPU-Z.

Ironically, CPU-Z indicates that this CPU is manufactured using a "7nm" process. However, in reality, the Compute Tile uses TSMC's N3B (3nm) node. Anyhow, here, the 285K is clocked at 5.5 GHz and 4.6 GHz across all P-cores and E-cores, respectively. The validation statistics show that the CPU is running at 100 degrees Celsius, so do not take this benchmark as an indication of the final performance.

Speaking of the frequencies, the Uncore or ring bus is clocked at 3.8 GHz, which is around 700 MHz lower than Raptor Lake. It explains the subpar gaming performance since the ring bus connects the CPU cores to the memory controller.

Moving over to the performance side, it is important to mention that the CPU is not running at its peak potential due to thermal throttling. The test bench features 2 x 16GB of CL32 DDR5-5600 memory, the ASRock Z890 Steel Legend WiFI motherboard, and an RTX 4080 Super.

In the single-core category, the Core Ultra 9 285K scores 909 points, struggling even against its predecessor. Surprisingly, in multi-core, the Core Ultra 9 285K outpaces the Core i9-14900K by around 12% despite being thermally throttled. It appears that Skymont's IPC uplifts have made up for removing Hyper-Threading.

In Blender v4.2, we are unaware of the test bench used, so these benchmarks do not reflect the final performance, which we'll see on October 24. Considering that, the Core Ultra 9 285K lands around 10% faster than the Core i9-14900KS but loses to AMD's Ryzen 9 9950X by a fair margin. The 20-core Core Ultra 7 265K also appeared in this test and is neck to neck against the AMD Ryzen 9 9900X.

With regard to these benchmarks, Arrow Lake's focal point will be efficiency, not performance. The average generation-on-generation uplift is in the ballpark of 10-15%, so Arrow Lake will need some serious efficiency gains to capture consumer interest. We'll know the exact numbers soon enough, as these CPUs are scheduled to hit shelves in just over a week.

Since our last AMD Ryzen 9 9950X ES leaked benchmarks story, AnandTech forum member Igor_Kavinsky has continued posting new Engineering Sample benchmarks in his original thread. The enthusiast has now undertaken 253W PPT and "Unlimited" Package Power Tracking (PPT) testing. This is in addition to the previously-covered 90W, 120W, 160W, and 230W PPT results we've covered. We've also included our own Ryzen 9 7950X benchmarking results for a quick comparison with the newer chip.  

In our last article on these benchmarks, we noted that the Ryzen 9 9950X seems to boast significant efficiency improvements over the Ryzen 9 7950X, not just higher performance in general. In particular, it was noted that the Ryzen 9 9950X seems capable of outperforming the Ryzen 9 7950X even when operating at a lower maximum wattage. It also remained fairly competitive with the 170W Ryzen 7950X at wattages as low as 120W.

AMD Ryzen 9 9950X ES Blender Benchmark Scores

*Author's Note: My prior article on this topic referred to these power targets as "TDP" instead of "PPT". These are...mostly the same thing, but whereas TDP (Thermal Design Power) refers to the CPU's power target, PPT (Package Power Tracking) refers to all power being directed to the CPU socket, and adjustments change the maximum wattage to the socket, and thus real TDP is lower. "Unlimited PPT" allows for as much wattage as the CPU and socket can support.

Isolating the two new benchmark results, one has to immediately note that only minor improvements have been gained by pushing the PPT power limits to their absolute maximum. The 253W result with a 5.5 GHz overclock still maintains impressive temperatures of 61C or less thanks to the liquid cooling setup used with this ES. However, fully removing the power limits kicks up temps to 80C under liquid cooling while achieving only the most marginal of performance improvements.

In other words, the most impressive results here... still start at around 170W, compared to the preceding CPU. It is fully within expectations for a successor to outperform its predecessor at the same or higher power targets, but the efficiency gains remain the most impressive aspect of this story.

That said, it's still nice that the Ryzen 9 9950X could be pushed this far with (apparently) a standard liquid cooling setup, though we don't know if it was done with an AIO or a custom loop. Apparently, no CPU delidding was needed to achieve these results, and in fact doing so would have likely upset AMD, since this is an Engineering Sample that must eventually be returned, per Igor_Kavinski's secondhand reports. (Note that while Igor posts these benchmarks, an unnamed source is actually running them and sending them to him.)

Starting on July 7, AnandTech forum member Igor_Kavinski began posting Ryzen 9 9950X engineering sample Blender benchmark results courtesy of an unnamed source — starting at a super-slim 60W TDP. Over the course of the following week, 90W TDP, 120W TDP, 160W TDP, and finally, max-capacity 230W TDP results were also posted. The results give us a comprehensive idea of how power efficiency will improve with next-gen Zen 5 AMD CPUs.

Before proceeding, it's evident that the newer Ryzen 9 9950X would outperform the older chip when given a more generous power budget. We didn't test our Ryzen 9 7950X at 230W TDP, but reports from other users in the thread point toward a ~20% performance improvement still present in that scenario. The interesting results here start at 170W and below.

AMD Ryzen 9 9950X vs Ryzen 9 7950X Blender Benchmarks

 *Note: All benchmark results listed above use AMD's Precision Boost Overdrive for a small performance boost. Additionally, the Ryzen 9 9950X ES is liquid-cooled.

At 170W, the Ryzen 9 7950X achieves a cumulative Blender score of 599.2. The Ryzen 9 9950X scores 678 at 160W, which outperforms its predecessor by about ~11% when both operate at more standard CPU TDPs.

The performance differentials between Ryzen 9 9950X and Ryzen 9 7950X start to narrow down when the newer chip is put at 120W. It is still within about ~5% of its predecessor's performance despite running with a 50W deficit in comparison.

These engineering sample benchmarks aren't the only insight we've received into AMD's upcoming Ryzen 9000 Series of CPUs. Earlier this week, Ryzen 9 9900X Geekbench results appeared that seem to have the new architecture pinned to take the crown in single-core performance, far outstripping the last-gen Ryzen 9 7950X3D and even the Intel Core i9-14900K.

Overall, we have to say that these emerging benchmarks are looking quite favorable for the future of AMD desktop platform users. However, some salt is required with pre-release benchmarks like these. Beyond raw performance gains, the power efficiency gains here also bode well for the eventual arrival of Zen 5 laptop chips, and should generally be nice for anyone trying to limit their power consumption. Even the 60W TDP results make this CPU look pretty usable since those scores align with an Intel Core i9-10980XE, per Blender's benchmark database.

A Blender benchmark for an alleged AMD Strix Point CPU engineering sample has surfaced. The results aren't breathtaking, but many unknown factors make it hard to say if the score reflects the chip's true capabilities.

The benchmark was spotted by @9550Pro (HXL) and posted to X (formerly Twitter). HXL suggests the CPU is a Zen 5 Strix Point engineering sample with four performance and eight efficiency cores. Strix Point is the codename for AMD's upcoming Zen 5-based Ryzen mobile processors, which will succeed Zen 4 processors like the Ryzen 7 Pro 7840U.

In the Blender benchmark, the CPU demonstrated a median score of just 270.92. This is slightly better than the 8-core AMD Ryzen 7 7700X, which scored 269.02. The engineering sample also outperformed the Ryzen 9 3900X 12-core processor (267.89) and Intel Core i7-13700HX (255.58), to name just two examples.

On the other hand, the AMD Ryzen 9 3900XT 12-core processor outperformed the Strix Point engineering sample. However, that is a desktop CPU, not a low-power mobile processor.

In a possibly more apples-to-apples comparison, the Ryzen 7 Pro 7840U scored 216.09 on the Blender benchmark.

Blender’s benchmarks don’t mention power, an important factor in CPU performance. If this Strix Point sample shows a 65W Zen 5 CPU faring better than a 105W Zen 4 CPU, the results are encouraging.

We've recently spotted other benchmarks, including Geekbench 6 results, that were underwhelming at first glance.

Blender benchmarks scale almost perfectly with more cores and higher clock speeds. So, if this is, in fact, a 12-core Zen 5-based mobile CPU, it’s scoring right about where we’d expect it to.

The unknown factors mean we have to take these results with a grain of salt. We don’t really know the CPU configuration, and we have no idea how much power is supplied to the CPU. None of that is really important, though, if you’re interested in seeing the progress (in the way of benchmarks) of the new processor family in the interim before we can get our hands on the chips ourselves.

A week ago, news about AMD's Inception vulnerability broke, and the first deep dive into the performance impact of mitigations has been published. Linux-centric Phoronix has just uploaded eight pages of test results. Using an AMD Epyc 7763 (Zen 3) based system, running Linux (of course), the site tested a plethora of apps and tabulated before and after-patching results. Depending on workload, you might not see much difference — however, some tasks were up to 54% slower on a patched system.

Data from Phoronix benchmarking

For the above table we looked at some of the worst results, as well as some of the test results from more-familiar apps like 7zip, Blender, and Firefox. Those three familiar apps don't suffer too much from the AMD Inception mitigations. Of the three, compression app 7zip seems to be the most affected — but how long do you spend de-compressing files in an average day?

Much more serious performance consequences are observed in applications that work on databases, code compilation, engineering, and image processing. The worst result we saw, with MariaDB, shows database operations were severely impacted on a patched Epyc system.

If you head over to Phoronix for a closer look at the data and a wider selection of results you will see that the results sometimes show more than just the AMD Inception mitigation being 'off' or 'on'. There will be up to three levels of patching with different configurations — some with purely kernel-based mitigations, others with the newest microcode, and another with the most secure Indirect Branch Prediction Barrier (IBPB) mitigation. Please note that IBPB was frequently (but not always) shown to be the worst performer of all mitigations. The default AMD Linux mitigation is 'safe RET mode'.

AMD announced that it has released the HIP SDK for Windows intending to democratize GPU computing. You no longer must choose between Team CUDA or Team HIP, as the HIP SDK will help developers make CUDA applications run on AMD hardware. Notably, this new SDK will run on a select number of consumer Radeon GPUs.

There has always been a significant divide between developers that work with GPU-accelerated applications. Some prefer Nvidia's proprietary CUDA API, while others opt for the open-source HIP API. The HIP SDK, part of AMD's ROCm platform, wants to bridge that gap, allowing developers to convert CUDA applications into C++ code that will work on Nvidia and AMD graphics cards. ROCm targets HPC and AI applications, whereas HIP is for typical desktop applications.

AMD asserts that porting a CUDA application to HIP SDK isn't challenging since CUDA and HIP are based on C++. Furthermore, the HIP SDK provides tools to help developers speed up the process, such as the HIPIFY toolset that will convert CUDA code into portable HIP C++. The HIP SDK doesn't work miracles, such as optimizing code. That's still a manual task that you have to do by yourself.

The HIP SDK works on 32-bit and 64-bit Windows operating systems, including Windows 10 (22H2), Windows 11 (22H2), and Windows Server 2022. According to AMD, the list of compatible graphics cards extends from workstation-grade to mobile gaming. AMD even brags about APUs being on the list. Of course, support also depends on the developer. The chipmaker cites an example of Blender HIP embracing AMD Radeon graphics cards going back to the Vega days.

AMD is still updating the compatibility list, but only ten Radeon graphics cards, between RDNA 3 and RDNA 2, are officially supported thus far. The Radeon Pro W7900, W7800, and W6800 hail from the Radeon Pro lineup. On the consumer end, the Radeon RX 7900 XTX, RX 7900 XT, RX 7600, RX 6950 XT, RX 6900 XT, RX 6800 XT, and RX 6800 support the HIP SDK.

Offering HIP SDK on Windows is a milestone for AMD. Nonetheless, the chipmaker will continue to make HIP SDK better by adding new features in the future and making an effort to deliver updates on par with AMD Software: Pro Edition graphics driver.

Recently, we've seen AI models that produce detailed text-to-video or use run a chatbot on your phone. Now, OpenAI, the company behind ChatGPT, has introduced Shap-E, a model that generates 3D objects you can open in Microsoft Paint 3D or even convert into an STL file you can output on one of the best 3D printers.

The Shap-E model is available for free on GitHub and it runs locally on your PC. Once all of the files and models are downloaded, it doesn't need to ping the Internet. And best of all, it doesn't require an OpenAI API key so you won't be charged for using it.

It is a huge challenge actually getting Shap-E to run. OpenAI provides almost no instructions, just telling you to use the Python pip command to install it. However, the company fails to mention the dependencies you need to make it work and that many of the latest versions of them just won't work. I spent more than 8 hours getting this running and I'll share what worked for me below.

Once I finally got Shap-E installed, I found that the default way to access it is via Jupyter Notebook, which lets you view and execute the sample code in small chunks to see what it does. There are three sample notebooks which demonstrate "text-to-3d" (using a text prompt), "image-to-3d" (turning a 2D image into a 3D object) and "encode_model" which takes an existing 3D model and uses Blender (which you need installed) to transform it into something else and re-render it. I tested the first two of these as the third (using Blender with existing 3D objects) was beyond my skillset.

How Shap-E Text-to-3D Looks

Like so many AI models we test these days, Shap-E is full of potential but the current output is so-so at best. I tried the text-to-video with a few different prompts. In most cases, I got the objects that I asked for but they were low res and missing key details.

When I used the sample_text_to_3d notebook, I got two kinds of output: color animated GIFs which displayed in my browser and monochrome PLY files I could open later in a program like Paint 3D. The animated GIFs always looked a lot better than the PLY files.

The default prompt of "a shark," looked decent as an animated GIF, but when I opened the PLY in Paint 3D, it seemed lacking. By default, the notebook gives you four animated GIFs that are 64 x 64, but I changed the code to up the resolution to 256 x 256 outputted as a single GIF (since all four GIFs looks the same). 

When I asked for something that OpenAI had as one of its examples, "an airplane that looks like a banana," I got a pretty good GIF, particularly when I upped the resolution to 256. The PLY, file, though, exposed a lot of holes in the wings. 

When I asked for a Minecraft creeper, I got something that a GIF that was correctly colored green and black and a PLY that was in the basic shape of a creeper. However, real Minecraft fans would not be satisfied with this and it was too messy of a shape to 3D print (if I had converted it to an STL).

Shap-E Image to 3D Object

I also tried the image-to-3d script which can take an existing 2D image file and turn it into a 3D PLY file object. A sample illustration of a corgi dog became a decent, low-res object that it outputted as a rotating, animated GIF which had less detail. Below, the original image is on the left and the GIF is on the right. You can see that the eyes seem to be missing. 

By modifying the code, I was also able to get it output a PLY 3D file that I could open in Paint 3D. This is what it looked like.

I also tried feeding the image-to-3d script some of my own images, including a photo of an SSD, which came out looking broken and a transparent PNG of the Tom's Hardware logo, which didn't look much better.

However, it's likely that if I had a 2D PNG that looked a bit more 3D-ish (the way the corgi does), I'd get better results. 

Performance of Shap-E

Whether I was doing text or image to 3D processing, Shap-E required a ton of system resources. On my home desktop, with an RTX 3080 GPU and a Ryzen 9 5900X CPU, it took about five minutes to complete a render. On an Asus ROG Strix Scar 18 with an RTX 4090 laptop GPU and an Intel Core i9-13980HX, it took two to three minutes. 

However, when I tried doing text-to-3D on my old laptop, with has an Intel 8th Gen U series CPU and integrated graphics, it had only finished 3 percent of a render after an hour. In short, if you are going to use Shap-E, make sure you have an Nvidia GPU (Shap-E doesn't support other brands of GPUs. The options are CUDA and CPU.). Otherwise, it will just take too long.

I should note that the first time you run any of the scripts, it will need to download the models, which are 2 to 3 GB and could take several minutes to transfer.

How to Install and Run Shap-E on a PC

OpenAI has posted a Shap-E repository to GitHub, along with some instructions on how to run it. I attempted to install and run the software in Windows, using Miniconda to create a dedicated Python environment. However, I kept running into problems, especially because I could not get Pytorch3D, a required library, to install.

However, when I decided to use WSL2 (Windows Subsytem for Linux), I was able to get it up and running with few hassles. So the instructions below will work either in native Linux or in WSL2 under Windows. I tested them in WSL2.

1. Install Miniconda or Anaconda in Linux if you don't already have it. You can find a download and instructions on the Conda site.

2. Create a Conda environment called shap-e with Python 3.9 installed (other versions of Python may work).

conda create -n shap-e python=3.9

3. Activate the shap-e environment.

conda activate shap-e

4. Install Pytorch. If you have an Nvidia graphics card, Use this command.

conda install pytorch=1.13.0 torchvision pytorch-cuda=11.6 -c pytorch -c nvidia

If you don't have an Nvidia card, you'll need to do a CPU-based install. The install is speedy but processing the actual 3D generation with the CPU was extremely slow in my experience. 

conda install pytorch torchvision torchaudio cpuonly -c pytorch

5. Build Pytorch. This is the area where it took me hours and hours to find a combination that worked. 

pip install "git+https://github.com/facebookresearch/pytorch3d.git"

If you get a cuda error, try running sudo apt install nvidia-cuda-dev and then repeating the process.

6. Install Jupyter Notebook using Conda.

conda install -c anaconda jupyter

7. Clone the shap-e repo.

git clone https://github.com/openai/shap-e

Git will create a shap-e folder underneath the one you cloned it from.

8. Enter the shap-e folder and run the install using pip.

cd shap-epip install -e .

9. Launch a Jupyter Notebook. 

jupyter notebook

10. Navigate to the localhost URL the software shows you. It will be http://localhost:8888?token= and a token. You'll see a directory of folders and files.

11. Browse to shap-e/examples and double-click on sample_text_to_3d.ipynb.

A notebook will open up with different sections of code.

12. Highlight each section and click the Run button, waiting for it to complete before moving onto the next section.

This process will take a while the first time you go through it, because it will download several large models to your local drive. When everything is done, you should see four 3D models of a shark in your browser. There will also be four .ply files in the examples folder and you will be able to open those in 3D imaging programs such as Paint 3D. You can also convert them to STL files using an online converter.

If you want to change the prompt and try again. Refresh your browser and change "a shark" to something else in the prompt section. Also, if you change size from 64 to a higher number, you get a higher resolution image. 

13. Double-click on sample_image_to_3d.ipynb in the examples folder again so you can try the image-to-3d script.

14. Highlight each section and click Run.

You'll end up, by default, with four small images of corgis. 

However, I recommend adding the following code to the last notebook section so that it will output PLY files as well as animated GIFs.

from shap_e.util.notebooks import decode_latent_meshfor i, latent in enumerate(latents):    with open(f'example_mesh_{i}.ply', 'wb') as f:        decode_latent_mesh(xm, latent).tri_mesh().write_ply(f)

15. Modify the image location in section 3 to change the image. Also, I recommend changing the batch_size to 1 so it only makes one image. Changing the size to 128 or 256 will give you a higher resolution image.

16. Create the following python script and save it as text-to-3d.py or another name. It will allow you to generate PLY files based on text prompts at the command line.

import torchfrom shap_e.diffusion.sample import sample_latentsfrom shap_e.diffusion.gaussian_diffusion import diffusion_from_configfrom shap_e.models.download import load_model, load_configfrom shap_e.util.notebooks import create_pan_cameras, decode_latent_images, gif_widgetdevice = torch.device('cuda' if torch.cuda.is_available() else 'cpu')xm = load_model('transmitter', device=device)model = load_model('text300M', device=device)diffusion = diffusion_from_config(load_config('diffusion'))batch_size = 1guidance_scale = 15.0prompt = input("Enter prompt: ")filename = prompt.replace(" ","_")latents = sample_latents(    batch_size=batch_size,    model=model,    diffusion=diffusion,    guidance_scale=guidance_scale,    model_kwargs=dict(texts=[prompt] * batch_size),    progress=True,    clip_denoised=True,    use_fp16=True,    use_karras=True,    karras_steps=64,    sigma_min=1e-3,    sigma_max=160,    s_churn=0,)render_mode = 'nerf' # you can change this to 'stf'size = 64 # this is the size of the renders; higher values take longer to render.from shap_e.util.notebooks import decode_latent_meshfor i, latent in enumerate(latents):    with open(f'{filename}_{i}.ply', 'wb') as f:        decode_latent_mesh(xm, latent).tri_mesh().write_ply(f)

17. Run python text-to-3d.py and enter your prompt when the program asks for it.

That will give you a PLY output, but not a GIF. If you know Python, you can modify the script to do more with it.

According to a review by Phoronix, AMD's new Ryzen 7 7800X3D (one of the best CPUs for gaming) is 7% faster on average in Linux Ubuntu 23.04 compared to Windows 11 Pro. The Linux OS also outperformed its Windows counterpart 72.5% of the time, in a suite of 80 applications Phoronix tested. While 7% isn't massive, it's nice to know that Linux users won't suffer a performance penalty compared to Microsoft's more mainstream Windows operating system.

Testing included a plethora of applications including OpenJDK Java, Image encoding, chess benchmarks, LuxCore, Video Encoding, Intel oneAPI, ASTC encoding, blender, Indigo Renderer, Appleseed, V-Ray, Geekbench, and Google Chrome browser benchmarks.

A few noteworthy wins for Linux include a 50% performance advantage in DaCapo Benchmark 9.12-MR1, 21% in Blender 3.5, 22% in OSPRay, and 32% in JPEG XL libjxl 0.7. Meanwhile, some noteworthy wins for Windows 11 Pro include: 22% faster performance in Blender 3.5's BMW27 benchmark, 17.3% in Blender 3.5's Barbershop benchmark, and 30% in Selenium PSPDFKit WASM benchmark.

But to re-iterate, the 7800X3D was faster in 72.5% of the tests overall with Linux compared to Windows 11, averaging  7% greater performance. (So most of the tests were almost neck and neck.)

Phoronix doesn't explain exactly how or why Linux is outperforming Windows 11, but it's no secret that AMD is actively adding CPU optimizations to Linux, in the form of the AMD P-State EPP driver. This driver, which was implemented in the Linux Kernel 6.0 not too long ago, adds additional CPU optimizations to improve the power consumption and performance of Zen 2, Zen 3, and Zen 4 chips. The new driver allows Ryzen CPUs to boost further than the vanilla ACPI CPUFreq driver, by scheduling tasks to the correct cores (i.e. tasking appropriate workloads to the cores with the highest clock speed potential).

This is probably at least partially the for AMD's 7% average performance advantage with Ubuntu. Phoronix has already seen a 6% performance improvement with this same driver on the Ryzen 9 7950X, so it wouldn't be unreasonable to assume the same would apply to the Ryzen 7 7800X3D.

As the official launch of AMD's Ryzen 7000X3D family of CPUs looms, the new Ryzen 9 7950X3D flagship processor's performance has been revealed in various popular benchmarks. On Friday, the new chip appeared in the Blender (via HXL) and Geekbench 5 (via @Benchleaks) benchmark suites. However, we cannot say that we are impressed with the performance of the new 3D V-Cache-enabled CPU compared to its counterpart without the extra L3 in these tests.

Blender

Blender is a tool for creating animations and visual effects. Like other rendering and video editing programs, it takes advantage of multi-thread performance and single-thread performance as well as application-specific optimizations.  

The 16-core Ryzen 9 7950X3D with 144MB of L2+L3 cache appears to be 5.4% slower than the 16-core Ryzen 9 7950X with 80MB of L2+L3 cache in the Blender benchmark.  Meanwhile, the model 7950X3D is a tad faster than Intel's Core i9-13900K (8P+16E, 32-threads) and substantially faster than Apple's 12-core M2 Max. 

The upcoming Ryzen 9 7950X3D features 16 high-performance cores supporting simultaneous multi-threading (i.e., processing 32 threads simultaneously) operating at up to 5.70 GHz. So, it is not surprising that it beats Apple's 12-core M2 Max (3.67 GHz) and is a bit faster than Intel's flagship CPU, which can process up to 32 threads simultaneously. It should be noted that the Core i9-13900K cannot clock its energy-efficient cores higher than 4.30 GHz. 

Meanwhile, the Ryzen 7950X3D's core chiplet die (CCD) with the additional 64MB of 3D V-Cache apparently features lower clocks than Ryzen 9 7950X, which diminishes the advantages of a larger cache for single-threaded performance.

Geekbench 5

Geekbench 5 is a synthetic benchmark that measures single-thread and multi-thread performance in different workloads. This program is not considered the most optimal benchmark for CPUs and GPUs for many reasons, but it still gives an idea about the performance of different hardware in similar workloads.

AMD's Ryzen 9 7950X3D with 3D V-Cache appears to be a tad slower than the Ryzen 9 7950X without extra cache in single-thread and multi-thread workloads. 

Regarding frequencies, the tested AMD Ryzen 9 7950X3D sample on the Asus ROG Crosshair X670E Hero worked at around 5480 MHz most of the time with a balanced power plan. Meanwhile, an AMD Ryzen 9 7950X on the same motherboard worked at a bit higher frequencies with the same balanced power plan, which is perhaps why the CPU without 3D V-Cache is faster in Geekbench 5.

Some Thoughts

Of course, we cannot draw accurate conclusions based on test results from one or two pre-production systems. But we can still share some thoughts about the results at hand. 

The main advantage of AMD's Ryzen 9 7950X3D CPU over its Ryzen 9 7950X counterpart is a massively larger L3 cache. Meanwhile, the former processor boasts higher base frequency and higher boost clocks across all cores; at least, it looks so based on results obtained in Blender and Geekbench 5. In many programs, clocks matter more than cache sizes, which is why the model 7950X3D is behind the model 7950X in the said benchmarks. 

It is reasonable to assume that in applications that depend primarily on single-thread performance and memory bandwidth, the Zen 4-based processors with 3D V-Cache will demonstrate significant advantages over their counterparts without the extra L3 cache. 

Keeping in mind that AMD's Ryzen 7000X3D CPUs are a couple of weeks away, it makes sense to wait for comprehensive reviews and find out all the details about the advantages and disadvantages of the new processors compared to regular Ryzen 7000-series units. 

The next Ada Lovelace graphics card that's due to be served up by Nvidia is the GeForce RTX 4080. Over the last few hours, there have been some interesting performance-indicative leaks claimed to feature this model, from usually reputable sources. The leaked benchmark scores show the RTX 4080 between 25% and 37% slower than the formidable RTX 4090. We must take these results with a pinch of salt, but they are worthy of closer examination.

Twitter user @Zed_Wang published screenshots apparently taken from 3DMark benchmark runs. 3DMark benchmarks provide an indicative gaming performance score, as they run through a fixed path within UL Benchmark’s tuned graphics engine.  We have both the Time Spy DX12 benchmark and Time Spy Extreme 4K DX12 benchmark scores to ponder over and compare. The RTX 4090 appears to be respectively 25% faster and 37% faster than the 4080 in these comparisons. If accurate, these results indicate that the RTX 4090’s advantages over the RTX 4080 are most apparent in higher-resolution gaming situations, at 4K or better.

For another perspective on RTX 4080 performance, seasoned leaker @Tum Apisak has shared what is purported to be a Blender benchmark run score. Blender is a free and open source 3D modelling application, so GPUs are useful in this situation for accelerating scene previews and rendered stills or animations. Additionally it supports GPU ray tracing hardware, so it helps to weight the relative potency of this aspect of a GPU. According to the Blender data, the upcoming RTX 4080 is going to be approx 28% slower than the flagship RTX 4090.

For a broader comparison, we have put the leaked benchmarking scores in the above table, alongside known performers like the RTX 3090 Ti and RTX 3080, as well as AMD’s Radeon RX 6950 XT for some red-on-green fun.

We have the official specs for both the GeForce RTX 4090 and RTX 4080 from the September launch event. At launch, there were two RTX 4080 models announced, but Nvidia has since trimmed it down to one. You can check out an extensive specs comparison in the linked story. But in essence, the RTX 4090 and RTX 4080 compare as per the table below.

Nvidia plans to launch the GeForce RTX 4080 on November 16, at $1,199. Expect a full and extensive review of the RTX 4080 for you to digest around the launch date.

As tipped by Twitter user HXL, Intel's 13th Generation Raptor Lake CPU lineup is putting up serious numbers against AMD's new Ryzen 7000 series processors in the Blender Benchmark. Intel's Core i7-13700K is competitive against the Ryzen 9 7900X, while the Core i5-13600K leaves the Ryzen 7 7600X and Ryzen 7 7700X in the dust.

For the uninitiated, Blender is an open-source content creation tool that can do several different tasks, including 3D rendering, video editing, modeling, animation VFX, and more. As a result, the program can be CPU and GPU-heavy, making it an excellent tool for benchmarking.

At the top of the list is AMD's 16-core behemoth, the Ryzen 9 7950X, spitting out a score of 607.53 points. Underneath the Ryzen 9 7950X sits the 24-core Core i9-13900K at 557.66 points. Then you have the Ryzen 9 7900X at 462.39 points, Core i7-13700K with 429.7 points, Core i5-13600K with 358.18 points, Ryzen 7 7700X with 305.51 points, and Ryzen 5 5600X with 234.65 points.

This Blender benchmark run is exciting to see and shows how potent Intel's hybrid microarchitecture can be right now in the form of Raptor Lake. Based on these numbers alone, we can see that Intel had clawed back much of that multi-core performance it lost to AMD in 2017 when Ryzen first launched.

AMD's Ryzen 9 parts are the only two chips that can beat Intel's Raptor Lake CPUs in Blender's heavy multi-core workload. The Ryzen 9 7950X is 9% faster than the Core i9-13900K, and the Ryzen 9 7900X delivers 7.6% more performance than the Core i7-13700K.

But, even here, the results aren't all that great, considering Intel's pricing strategy. For example, the Core i9-13900K is $50 cheaper than the 7950X, and the Core i7-13700K is a whopping $100 more affordable than the 7900X.

(Note: Technically, Intel's Core i7-13700K results were using an older version of Blender - version 3.2.1, instead of version 3.3 all the rest of the chips are using. But, according to a report by Phoronix, version 3.2.1 features a slower benchmark vs. 3.3, meaning the Core i7-13700K would probably be hair quicker in version 3.3.)

Once we get into the Ryzen 7 and Ryzen 5 parts, Intel completely dominates both parts, with both its Core i7-13700K and Core i5-13600K. The i5-13600K beats the Ryzen 7 7700X and Ryzen 5 7600X by a landslide, being 17% and 51% faster, respectively.

It'll be interesting to see how Raptor Lake behaves once we review the architecture in the next several weeks. But it appears like Intel has the potential to win the multi-threaded battle with its latest rendition of its hybrid CPU architecture.

Intel’s Arc graphics cards aren’t just for gamers, it seems, as the previously CPU-exclusive company has taken the lid off a new line of professional GPUs to complement the existing Arc line — well, existing in China, maybe. The new cards are called Arc Pro, and target those who use their graphics cards for more than shooting bad guys. Maybe they won't be among the best graphics cards for gaming, but the AV1 encoding at least might get some takers.

Intel today unveiled one mobile professional GPU, the A30M, and two desktop models: the single-slot A40 and double-slot A50. Both desktop cards are described as being for small form-factor machines, which makes us suspect Intel may have some much larger cards up its sleeve.

All the newly announced GPUs feature built-in ray tracing hardware, machine learning capabilities and industry-first AV1 hardware encoding acceleration. Google’s royalty-free, open source alternative to HEVC, AV1 hasn’t gained a lot of traction on the web so far despite promises from Netflix and YouTube, with its main use being in Google’s Duo video calling despite beating HEVC for compression quality. It’s always been very slow to encode, however, so a good hardware accelerator and Intel’s backing could see it take off.

Elsewhere, the Intel Arc Pro A-series graphics processors are certified ready with leading professional software applications within the architecture, engineering and construction, and design and manufacturing industries. And you can also use them to run the likes of Blender and the open source libraries in the Intel oneAPI Rendering Toolkit, which “are widely adopted and integrated in industry-leading rendering tools,” according to Intel.

We do have to wonder about the drivers situation with Arc Pro. There have been numerous documented problems with the Arc A380 cards, and Intel has already released several driver updates — which can ultimately get Windows into a state where the drivers won't even install, according to Gamers' Nexus. Other aspects of the drivers, like Smooth Sync, are broken. But perhaps that's because Intel has been focusing its driver efforts on the professional side of things? We hope that's the case, because as bad as the consumer drivers are, professionals will be far less willing to deal with broken support.

SIGGRAPH attendees in Vancouver this year will be able to see Arc Pro demos at Intel’s booth, while the rest of us can wait for the chips to be released “later this year.” We look forward to pitting the Intel Arc Pro cards against Radeon Pro and RTX A-series offerings from AMD and Nvidia.

Peru-based tech site XanxoGaming has put what it claims to be a retail version of the AMD Ryzen 7 5800X3D through a series of benchmarks (via VideoCardz). Sadly, for those eyeing the potential of the 5800X3D for gaming, no PC games were tested with the newly built system - just rendering and sys-info tool type standards.  

At the official unveiling of the 5800X3D in mid-March, AMD gave us the specs, pricing, and release date. Additionally, to build anticipation, it highlighted the PC gaming prowess of the upcoming 3D V-Cache infused processor. In a selection of modern games and older favorites, AMD's charts showed the new Ryzen 7 5800X3D was 15% better than the 5900X, wresting the upper hand from the Intel Core i9-12900K. 

So, it isn't easy to fathom why Xanxo should rush to release a review featuring the following; Cinebench, Geekbench, CPU-Z, and Blender. Perhaps it was just the quickest and easiest option. However, the site promises to add gaming benchmarks to its review shortly (it is working on them now).

Above is a selection of the benchmarking screenshots from the source, showing that the upcoming 3D V-Cache packing CPU is only about as fast as the Ryzen 7 5700X in many single- and multi-core tests. We reckon this is due to the lower clocks of the upcoming chip. However, when the V-cache comes into play in certain apps, like the rendering tasks in Blender, we see the Ryzen 7 5800X3D outpace the standard Ryzen 7 5800X by up to 11%.

Key confirmed specs of the AMD Ryzen 7 5800X3D are available on the official AMD product pages. In brief, it is a Zen3 architecture 8C/16T CPU with a base/boost of 3.4 / 4.5 GHz, and a hefty L3 cache of 96MB. This 7nm processor has a TDP of 105W. Since the product launch we have learned that it is not overclockable.

AMD explained that the 3D V-Cache is sensitive to higher voltages, and has thus roped off core voltage tweaking and user CPU frequency adjustments. Memory and fabric overclocking remain enabled. All the motherboard vendors are preparing BIOS updates for full support of the new chip and some, like Gigabyte, seem to suggest they have implemented changes for AMD's new "V-Cache optimizer."

The official release of the AMD Ryzen 7 5800X3D is scheduled for April 20, a little under a fortnight from today. AMD says the new AM4 CPU has an MSRP of $449. 

Intel Graphics has shared a video featuring the 3D rendering app Blender accelerated by an Intel Arc Alchemist GPU. Senior Software Evangelist & Developer Affinity Programs Manager at Intel, Bob Duffy, shows us how slickly the real-time rendering preview works with a bit of Arc desktop muscle applied to it.

After a short and sweet intro, Duffy takes us through setting up Blender for using the unreleased Intel GPU. Next, he does some snappy scene manipulation and quickly renders at excellent preview quality in the viewport.

Duffy doesn't mention any technical details of the Intel Arc GPU used in the demo. Moreover, the dialog boxes we see in setup and subsequent setting adjustments have the Intel Arc GPU technical details blurred out in the video. It spoils our fun somewhat, but thankfully someone in the Intel Graphics team must have given the following text overlay the green light (about 1min 10sec into the video): "Demonstration conducted on Intel Arc pre-production desktop discrete graphics card hardware. Results may vary."

Explaining how the upcoming Intel Arc Alchemist desktop GPU works in Blender, Duffy says that in this demo, the GPU specifically denoises the fast rendered scene and makes it a lot more pleasant and finished looking. Specifically, Duffy says Blender uses AI to leverage 'Intel Open Image Denoise.'

Duffy also shows how quickly Blender can update the high-quality preview when changing the properties of materials in the scene. To demonstrate this, he adjusts the refractive index of the glass object, center scene, several times.

Intel's Software Evangelist also shows some depth of field manipulation. Most of you will be familiar with the depth of field, but in brief, it is a photographic effect where things out of the chosen focus plane get blurred. It is an established and popular technique in portraiture, and now every new smartphone can do it or replicate it thanks to software/AI.

Duffy performed a final render of the tabletop scene to wrap up the video. It was completed in 7.3 seconds and looked snappy; however, he didn't provide any scene render comparison times using known GPUs from any vendor.

This updated Blender with Intel Arc acceleration features baked in will arrive "sometime in Q2," which is also the timescale in which we should see the discrete desktop GPUs. Remember, the Intel Arc laptop GPUs launch on Wednesday.

Intel seems keen for its Arc GPUs to succeed, so joining the Blender video demo earlier today in its social media feeds was a teaser video bearing the date 3/30/22.

The teaser video is quite opaque and probably ranks among the worst teasers, including from Apple, Samsung, and Asus. However, we thought it wouldn't hurt to tack it onto the end of our Blender rendering story. So if you see something in Intel's short, dark and blurry video that we have missed, please let us know in the comments.

Courtesy of Phoronix; It appears that the latest build of Windows 10 is the most optimal operating system to use for Intel's new Core i9-11900K Rocket Lake CPU. Tests show the i9 wining more benchmarks in a Windows 10 environment compared to Linux Ubuntu.

For the test bench, Phoronix ran a core i9-119000K with 32GB of 3200MHz RAM, with 1TB of SSD storage. on a Maximus XIII Hero.

As for the operating systems, Phoronix used the latest build of Windows 10 Pro, version 19042, and the latest version of Ubuntu, version 20.10, and version 5.12 of Linux.

The benchmarks posted above are just a few of the tests Phoronix conducted on both Windows 10 (see how to get Windows 10 for free) and Ubuntu. Overall, however, comparing all of Phoronix's tests shows that Windows 10 Pro wins 61.5% of the overall tests compared to Ubuntu which netted a score of just 38.5%.

Phoronix also tested the 11900K's integrated Xe graphics on both operating systems, and Windows 10 came out with an even higher win rate. In the eight graphics tests conducted, Ubuntu Linux managed only a single win, though in either case the integrated GPU it's nowhere close to matching the best graphics cards. 

This is unusual behavior coming from Intel's processors; due to Linux's superior resource management, we normally see Linux operating systems take the win compared to Windows 10. But with Rocket Lake, it appears the opposite is now true.

We don't know why the tests came out this way, but presumably, Microsoft has added some extra optimizations to Windows 10 we don't know about. We will have to do our own research into the matter to see what is really going on.

In our tests, the Core i9-11900K is faster for gaming than most of the best CPUs, but is outpaced by the AMD Ryzen 9 5900X. When we compared the AMD Ryzen 9 5900X vs the Core i9-11900K in a seven-round face-off, the Ryzen took five rounds.

Huawei, through its HiSilicon subsidiary, has a line of promising 7nm ARM v8-based Kunpeng processors that stretch up to 64 cores for the data center and support leading tech, like PCIe 4.0. Now at least one model of the chip is being used for desktop systems, too. Chinese YouTube channel 二斤自制 purchased and tested a Huawei-powered desktop PC that features both the company's eight-core eight-thread 7nm Kunpeng 920 ARM v8 processor and the Huawei D920S10 desktop motherboard in a third-party system, giving us the first glimpse of the new products enabled by Huawei's recent entrance into the market as a supplier to OEMs that produce desktop PCs. 

The development could help further China's targeted strategy to reduce its reliance on western semiconductor technology. Still, in many ways, the system highlights the difficulties the country has encountered, particularly in terms of software support. In fact, that's the primary focus of the video. The video doesn't give us much in the way of broadly-comparable benchmarks (though there are a few tidbits), but we do learn some specs that we'll cover below.

The narrator spends much of the video covering the problems she encountered with running meaningful software applications. Due to Kunpeng's ARM architecture, the system is limited to running the China-produced 64-bit UOS operating system that is largely a modified flavor of Linux. The narrator commented that the UOS operating system runs smooth and has an intuitive interface, and it even supports a 4K resolution at 60Hz via a Yeston RX550 graphics card. Still, she had to pay an extra 800 Yuan (~$115) to gain access to the app store. Moreover, the store had a woeful selection of applications, lacking such staples as Adobe and other apps. That's exacerbated by the system's lack of support for 32-bit software.

The system did run a Blender BMW test render, but it completed in 11 minutes and 47 seconds, which is woefully slower than most modern chips. The system did well streaming 4K video, but 'choked' during local playback due to poor encoding performance. The narrator says the system is obviously best for light office work only. 

The channel purchased the system for 7,500 Yuan (roughly $1,060 USD), and it comes with an eight-core eight-thread 2.6 GHz Kunpeng 920 2249K processor soldered to the motherboard. We can't find specs for this processor online, but the video lists it with 128K of L1 memory (64K I$, 64K D$), 512K of L2, and 32MB of L3. 

The Huawei D920S10 motherboard has four DIMM slots, but the system only has 16GB of Kingston DDR4-2666 memory spread across two DIMMs. Despite the chip's seeming support for PCIe 4.0, there are only three PCIe 3.0 slots available (x16, x4, x1), and the motherboard's connectivity options are also pretty mundane (6 SATA III ports, two M.2 slots, two USB 2.0 and USB 3.0 ports, and a VGA connection). The board also has a Gigabit ethernet port, or you have the option to use an optical connection of an unspecified speed. 

A 256GB SATA hard drive, 200W power supply, and a Yeston RX550 graphics card round out the other accommodations. The system also comes with an optical drive.

The difficulties the channel encountered with software availability, and performance, highlights that even the best chips in the world aren't very effective without a robust software and developer ecosystem. That's yet another facet of the challenge that China faces as it looks to reduce the amount of silicon it procures from external vendors, and the Huawei-powered systems could be designed to help foster a developer ecosystem for the ARM architecture and the UOS operating system. 

According to IC Insights, China-native vendors only produce 6.1% of the country's total silicon consumption, and a recent report indicates the country will fall far short of its 2025 goals for 70% semiconductor self-sufficiency, instead only hitting one-third of its original target. 

The country has invested heavily in multiple native semiconductor producers and their projects, with its multi-pronged efforts including chips like the x86 Zhaoxin KaiXian processor we recently tested. The EPYC-based Hygon Dryhana x86 processors were also developed under a joint venture with AMD that was later scuttled by the US government, and we've also seen signs that Huawei is testing a desktop PC design with AMD's Ryzen 4000 series processors. The latter shows that Huawei might also be pursuing an x86 route into the desktop PC market, albeit with non-native processors.

Raspberry Pi projects are more about what you can do with a Pi, not what you should do. Earlier this week, Shane with the Game Dev Academy YouTube channel pushed the Raspberry Pi to its limits by using it for 3D modeling with Blender. 

Don't get too excited. A Raspberry Pi is far from being the ideal 3D modeling computer. You'd still be much better off running Blender on a more capable desktop PC. But this developer has shown that Blender on Pi as at least possible. 

This project was inspired out of necessity (like most Pi projects). Shane said he was working on a new tutorial when his graphics card failed. Instead of rushing out and buying a new graphics card, he looked to see if a Pi could get the job done.

Blender doesn't require much in the way of hardware. Shane decided to use version 2.79, which requires 4GB of RAM and a dual-core CPU with a clock speed of at least 2 GHz. However, it's recommended that you use 16GB of RAM and a quad-core CPU. Shane chose to use a 2GB Raspberry Pi 4, which features a 1.5-GHz processor with four cores. 

Setting up Blender on Raspberry Pi 

Getting Blender to run on the Raspberry Pi looked easy enough, based on what the Game Dev Academy's video showed us. 

You'll need a Raspberry Pi with Raspbian installed. Shane searched for Blender in the Add / Remove Software window to get the Blender 2.79 package. 

Once installed, Blender launched, and Shane demonstrated its capabilities with a time-lapse video. The biggest drawback came with rendering, which took much longer to complete on a Raspberry Pi than a desktop. 

 If you want to see more of Shane's work, check out Game Dev Academy on YouTube for more projects and 3D modeling tutorials. 

When we first went looking for budget X570 motherboards to review, MSI didn’t sample us any. So we decided to buy the company's sub-$200 MPG X570 Plus motherboard to see what it was capable of. We already knew what readers were saying about the board, and confirming those findings should have generated an excellent article on how it failed. But when the board didn’t fail our tests, we went ahead and wrote the review in a more-positive light than others.

After publishing our MPG X570 review, we got a lot of feedback from readers and our colleagues in the press, with legitimate concerns about our test methodology and equipment. If our colleagues, and even MSI's own testers, were seeing VRM overheating, why weren't we? 

We appreciate this kind of feedback and take it very seriously, because testing is so important to what we do. Our senior editorial team reviewed the feedback and we decided to change our test methods: using new workloads, new temperature sensors and a more-demanding CPU. 

The Original Testbed

Our original testbed was a mid-tower with a 2x120mm radiator and was based on the Cooler Master HAF XB.  We upgraded it last year with an Eisblock XPX CPU water block, fed by an Eisbecher D5 150mm Pump/Reservoir through a NexXxoS UT60 X-Flow 240mm radiator, all from Alphacool. 

The goal of that upgrade was to support new HEDT processors without significantly increasing airflow over the voltage regulator compared to our previous Fractal Design S24 unit, but the thicker radiator required increased fanspeed to retain similar airflow. Since the new radiator was too thick to fit inside the case, we mounted it on the outside with the new EK Vardar fans in a less-efficient “pull” configuration. The result of this airflow restriction was that the new configuration had similar case airflow to the previous version.

Stepping Up To Ryzen 9

We began our X570 review series with a Ryzen 7, due to shortages of Ryzen 9 in the days leading up to the platform launch. Retaining it allowed our data to remain consistent, but understandably caused concerns about the lighter power requirement compared to Ryzen 9. Our first action was to buy a retail-boxed 3900X processor. To avoid any issue of voltage regulator thermal throttling, we set its limit to 125° in the motherboard’s firmware. This would allow us to see peak values beyond the default 102° to 105° throttling range.

While our original 3700X overclock reached 4.20 GHz at 1.375V with a peak voltage regulator temperature of approximately 90°, that peak dropped by one degree in our retest. The same voltage regulator runs almost one degree warmer when running the non-overclocked Ryzen 9 3900X. The ~89° temperature we measured isn’t a bad finishing point for the 3700X overclock, but it’s also not a great starting point for the 3900X at full load.

Stepping “Up” to Thermocouples

While we had researched the applicability of HWiNFO64 to our test board, many readers were concerned that the sensor it was reading might not be close enough to the hottest MOSFET to get an accurate number. For added perspective, here are the pertinent locations:

Though Amazon offers a variety of logging thermometers at lower prices, we opted for the Extech model SDL200, complete with four factory-calibrated Type-K thermocouple cables, for assured accuracy. As we only had four thermocouples to test eight phases, we attempted to address the hottest MOSFET of each phase pair: According to our IR meter, the most ideally placed thermistor is the one at the bottom.

Though our thermocouple occasionally read a higher temperature than the motherboard’s onboard thermistor, the overall impact was that the results from our first thermocouple reasonably tracked the onboard sensor. Yet there were still questions regarding the applicability of Prime95 to this test, so we moved on.

MORE: Best Motherboards

MORE: How To Choose A Motherboard

MORE: All Motherboard Content

Concerns about using Prime95 small-FFTs are addressed through the addition of Cinebench and Blender. The hottest default test for Blender 2.82 was its “pavillon_barcelona” scene producing the greatest amount of heat, so we downloaded the CLI (command-line interface) version under the organization’s “other platforms” dropdown and wrote a batch file to repeat the test eleven times. Meanwhile, Cinebench allowed us to simply set a minimum runtime and repeated the test until the runtime had lapsed.

Cinebench warms up faster, but to a slightly lower temperature than the other two tests. Perhaps if the benchmark length were a bit longer—requiring fewer cycles to complete this one-hour test—it might even have been the best metric. But, it’s close enough to the other power tests that we wouldn’t call anyone out for using it in this configuration.

Retesting At 40% PWM

Concerns that our fans might be performing beyond those of the typical system are easily addressed by dropping PWM cycles to 40%. The ~ 945 RPM result should be slow enough to represent a wider range of systems with front-mounted dual-fan radiators.

The worst reading of 102 degrees at 40% fans in Prime95 small-FFTs also happens to be the point at which the motherboard was supposed to have throttled the CPU, had we not configured it to a higher throttle point. If any of us owned this 3900X machine, we’d start looking for ways to get more airflow (such as fans that spin faster than 945 RPM).

Retesting In A Closed Build

The Phantom 410 is designed to support a 2x140mm radiator internally with two intake fans between the frame and top cover, but we flipped that arrangement with our 2x120mm unit to match traditional orientation: Both fans are under the radiator, blowing upward. We even tossed a pair of 3.5-inch drives in the lower bay to keep the build as realistic as possible.

We’ll speed this part up since you’re probably getting tired of redundant data: Putting the MPG X570 Gaming Plus into a top-radiator case resulted in dramatically lower temperatures compared to our front-cooled test platform. But, what about overclocking?

MORE: Best Motherboards

MORE: How To Choose A Motherboard

MORE: All Motherboard Content

The potential for increased VR MOS temperature could stem from the use of AMD PBO overclocking, which is built into every modern AMD motherboard we’ve tested. It’s supposed to push the processor frequency and voltage to its thermal limit, assuring increased VR MOS temperature.

Our closed system didn’t reach its 102° default throttle point at 100% fans, and dropping the case’s onboard controller to “medium” and the system’s PWM fans to 40% got it only a few degrees warmer. Another thing that happened was that our Prime95 test effectively dropped out.

Under Prime95, PBO would spike the CPU multiplier and core voltage (up to 1.5V VID!) and hold it at its throttle point before engaging a cooldown. To eliminate the motherboard, firmware, and CPU as culprits, we applied the same load to a Gigabyte TRX40 Aorus Xtreme and a Ryzen Threadripper 3970X that was already sitting on our original testbed.

Note that in order to get this result, we had to use PBO with the 3970X in a 2x2 Downcore (16C/32T) configuration: With PBO enabled at its default 32C/64T, a more traditional throttling curve appeared.

Manual Overclocking

Given that PBO settings aren’t necessarily optimized for the lowest-stable per-clock voltage or the highest per-voltage frequency specific to a single CPU sample, we spent some time to figure out that the highest setting we could get was 4.20 GHz at 1.2625V. Anything greater caused voltage regulator thermal throttling, despite the cooling advantages that this closed case has over our original testbed.

We successfully reached 115 degrees by manually overclocking the CPU, and this can be done at even lower settings: An earlier test on our original testbed yielded a lower 4.10 GHz limit at 1.175V, due to the voltage regulator running hotter on our semi-open system. Thus, while the 3900X works on this board, it’s not an ideal pairing when put into a less-than-ideal environment.

Conclusion

In our original conclusion, we noted that the MPG X570 Gaming Plus could be a bargain buy for those who simply wanted to pair a PCIe 4.0 SSD with a Ryzen 5 or 7 CPU. That narrow value advantage has since disappeared, as the cooler-running Asus Tuf Gaming X570 Plus has dropped to $165 at Newegg while the MPG X570 Gaming Plus remains $160 at Amazon. Even a small decrease in temperature would be enough to bump a board away from the margins that the 3900X-equipped MPG X570 Gaming Plus hit with our testbed cranked down to 40% fans.

While none of our tests validated the extremes we had come to expect, it’s nice to know that the thermal performance of our original testbed falls directly between that of our own closed build and MSI’s worst-case scenario. MSI has acknowledged that the VRMs on the X570 Gaming Plus can run hot under some operating conditions if the board is paired with a 3900X (or beyond) and put under heavy load. The company says to look to its forthcoming X570 Tomahawk motherboard as a similarly-priced alternative to the MPG X570 Gaming Plus. However, the company hasn’t specified a release date.

In short, it's certain that the MPG X570 can run hot depending on the CPU, settings, cooling system and workload that exceed our original test conditions. Responding to the valuable feedback we received has provided a great opportunity for us to refine our motherboard testing going forward so we can better reflect the range of components someone might use in the real world.

MORE: Best Motherboards

MORE: How To Choose A Motherboard

MORE: All Motherboard Content

Update 4/2/19, 7:45 a.m. PT: Last week Nvidia revealed more information about the first Creator Ready Driver release. The company said this first driver "comes optimized for a number of exciting app updates" that accelerate performance on its RTX GPUs, including Autodesk Arnold 5.3, Unreal Engine 4.22, REDCINE-X PRO 51, Adobe Photoshop Lightroom CC and Substance Designer by Adobe.

Nvidia also claimed performance improvements of up to 8-12 percent in Blender Cycles, Cinema 4D, Adobe Photoshop CC and Adobe Premiere Pro CC. Although, as always, the improvements will vary based on the rest of the system. The first Creator Ready Driver is available now via the GeForce Experience software.

Original article, 3/21/19, 8 a.m. PT:

People are used to Nvidia releasing driver updates that improve support for the latest in gaming via the Game Ready Driver program. Yesterday the company announced a similar program for creative tools called, fittingly enough, the Creator Ready Driver program.

"To achieve the highest level of reliability," Nvidia said in its announcement, "Creator Ready Drivers undergo extensive testing against multiple revisions of the top creative applications." That should improve performance in old and new versions of those apps.

The company isn't testing these apps in a vacuum. (Mostly because they can't run PhotoShop.) Nvidia said it will "conduct exhaustive multi-app testing for each type of creative workflow, evaluating driver quality in the same manner that creators work day-to-day."

Nvidia offered an example of such a workflow: using Adobe Premiere Pro CC to cut a video, sending it to After Effects CC for post-production and then kicking it back to Premiere Pro for rendering. Creator Ready Drivers focus on that process, as well as its parts.

The company was careful to note that its Creator Ready Drivers are supplementing its Game Ready Drivers, not replacing them. It said:

"Both Game Ready Drivers and Creator Ready Drivers will include the full Nvidia feature set and application support for games and creative apps, so users can continue to use either driver they prefer. But creators now have an option to receive designated driver releases with more in-depth testing to meet the stringent demands of their work."

Nvidia has released the first Creator Ready Driver; it's available via its website and the GeForce Experience app.(find it by opening the menu in the top-right corner of the window). Future releases will be "timed to key creative application updates."

Creator Ready Drivers will include support for Turing-based GeForce RTX, GTX and Titan GPUs; the Volta-based TITAN V, Pascal-based GeForce GTX and Titan GPUs and "all modern Quadro GPUs." Nvidia said they are optimized "for all the top creative applications."

AMD Radeon VII 16GB Review

A lopsided graphics market is no good for enthusiasts. Nvidia’s Turing-based cards proved this by serving up solid performance, but simultaneously turning many gamers off with steep prices. It was only when the company worked its way down to GeForce RTX 2060 and had to compete against Radeon RX Vega 56/64 that it got serious about telling a more compelling value story.

And there was nothing we could do. After all, GeForce RTX 2080 Ti was (and still is) the best card out there for 4K gaming. GeForce RTX 2080 couldn’t be touched by anything in AMD’s line-up, either. Even GeForce RTX 2070 flaunted an advantage over Radeon RX Vega 64 across our benchmark suite.

For all we know, AMD was just as surprised by Nvidia’s hubris as we were. By launching faster cards at higher price points, Nvidia failed to give its customers more for less. Sure, we have real-time ray tracing and DLSS to look forward to. But in today’s games, high-end Turing challenges you to pay a similar price for familiar performance.

Originally, it sounded like AMD would use its 7nm Vega 20 GPUs to shore up favor with customers able to benefit from the architecture’s formidable compute potential. But someone at the company must have smelled blood in the water, deciding that a Vega 20 processor with some of its on-die resources disabled could still be fast enough to counter GeForce RTX 2080 at a similar price point.

Enter Radeon VII, a card that looks a lot like the Radeon Instinct MI50 with some strategic tweaks to prevent cannibalization of the datacenter-oriented model. It does, in fact, take aim at GeForce RTX 2080, and AMD similarly seeks to sell Radeon VII for $700. But Vega 20 lacks the functionality to match Nvidia’s RT and Tensor cores, the Turing architecture’s future-looking features that enable real-time ray tracing and Deep Learning Super Sampling. Instead, AMD is pushing this card’s 16GB of HBM2 and massive 1 TB/s of memory bandwidth as the keys to smooth performance at high resolutions.

Meet Vega 20: Borrowed from the Data Center, Tuned for Enthusiasts

Under the hood, AMD’s Vega 20 graphics processor looks a lot like the Vega 10 powering Radeon RX Vega 64. But a shift from 14nm manufacturing at GlobalFoundries to TSMC’s 7nm node makes it possible for AMD to operate Vega 20 at much higher clock rates than its predecessor.

At 331 mm², Vega 20 is much smaller than the 495 mm² Vega 10. This helped free up room on AMD's interposer to add two more stacks of HBM2. At the same time, Vega 20 sports 13.2 billion transistors (compared to Vega 10’s 12.5 billion). AMD says the extra 700 million transistors go to optimizing for higher clock rates, improving the GPU’s video encode support at 4K/60 Hz, and enhancing the capabilities of its compute engine.

We’re guessing that the compute engine augmentations AMD refers to are half-rate FP64 and support for new INT8 and INT4 instructions. But whereas the Radeon Instinct MI50 and MI60 cards based on the same processor leverage all of its features, Radeon VII’s double-precision throughput is artificially limited to one-quarter of Radeon VII's single-precision rate. That's significantly better than the 1/16th-rate AMD originally claimed. But after we presented our original benchmark data to the company, AMD revised the card's specifications to acknowledge its pre-release VBIOS and drivers were already set to the higher 1/4 FP32 rate. As a result, Radeon VII offers higher peak FP64 performance than any other card in AMD's consumer line-up, and in fact only really trails Nvidia's pricier Titan V.

PCI Express 4.0 transfer rates, supported on Radeon Instinct MI50 and MI60, are still disabled though, as are the Instinct cards’ Infinity Fabric links. According to AMD, Radeon VII’s compute capabilities size up to Radeon RX Vega 64 and Radeon Instinct MI60 as follows:

Otherwise, a similar layout means a complete Vega 20 sports four shader engines, each with its own geometry processor and rasterizer. There are 16 Compute Units per Shader Engine, with 64 Stream processors and four texture units per CU. All told, that’s 4,096 Stream processors and 256 texture units across the GPU.

But Radeon VII doesn’t utilize a complete Vega 20. Rather, AMD disables four of the chip’s CUs, yielding 3,840 Stream processors and 240 texture units. The company compensates for its resource deficit compared to Radeon RX Vega 64 by operating Radeon VII at much higher clock rates. The former’s 1,274 MHz base frequency increases to 1,400 MHz on Radeon VII, while Vega 64’s 1,546 MHz boost clock rises to 1,750 MHz. AMD also specifies an 1,800 MHz peak engine clock for lighter workloads.

Each of Vega 20's Shader Engines sports four render back-ends capable of 16 pixels per clock cycle, yielding 64 ROPs. These render back-ends become clients of the L2, as we already know. That L2 is 4MB, similar to Vega 10 and twice the size of Fiji’s 2MB of L2.

Although a large L2 cache helps keep frequently-used data close to the GPU, Vega 20 shouldn’t be as reliant on it as Vega 10 thanks to a better-balanced memory subsystem. You see, Radeon RX Vega 64 served up significantly more theoretical shading performance than its predecessor. However, its narrower 2,048-bit memory bus actually limited bandwidth to 484 GB/s versus Fury X and its 512 GB/s. Radeon VII pops the cork with two more 4-hi stacks of HBM2 flanking a much smaller Vega 20 GPU, resulting in an aggregate 4,096-bit pathway, taking us back to the good old days. Only this time around, a 2 Gb/s data rate enables a staggering 1 TB/s of bandwidth.

Gone is Radeon RX Vega 64/56’s dual BIOS with multiple power/performance profiles. As a result, we’re looking at a single 300W total board power specification for Radeon VII, up 5W from Radeon RX Vega 64. It’d be hard to ignore the obvious performance/watt comparisons between a 300W Radeon VII going up against Nvidia’s 215W GeForce RTX 2080 at a similar price point. Needless to say, if efficiency is a variable in your purchasing decision, AMD is at a fairly severe disadvantage.

More troubling, we believe, is Radeon VII’s acoustic situation. Following in Nvidia’s footsteps, AMD retired its blower-style cooler in favor of one with axial fans. But rather than creating a quieter thermal solution able to keep Vega 20 cooler, Radeon VII is easily just as loud as the reference Radeon RX Vega 64 due to a fan curve that ramps up to 2,900 RPM under load. We approached AMD about Radeon VII’s noise because, frankly, it’s disappointing. The company explained that the card’s shipping configuration is tuned for enthusiasts, and that it’s working on other options that’d conceivably trade performance for better acoustics.

Changing the Way Thermals are Monitored

AMD also says Radeon VII reflects an effort to get more performance from its GPU by enhanced thermal monitoring. Rather than reading temperatures around its processor using 32 sensors as Vega 10 did, it now pulls data from 64 sensors strategically placed around the Vega 20 die. And whereas the junction temperature derived from those sensors previously controlled thermal shutdown protection, it’s now being used for throttling and fan control as well.

So, even if your hardware monitoring software reports a GPU temperature of 75 or 80˚C (referred to as edge temperature), the actual junction temperature might be 95˚ or more. As it happens, 95˚ is where Radeon VII’s fans kick up to 2,900 RPM.

The point of all of this is that AMD is able to get more aggressive about keeping Vega 20 at high clock rates, given more data from a larger network of sensors to know with reasonable levels of confidence that other areas of the GPU aren’t running at temperatures beyond their safe limits. Outwardly, gamers enjoy higher frame rates in thermally-limited scenarios—typically demanding titles played for long periods of time. The caveat is an assertive fan circuit that spends a lot of time at maximum speed in order to keep Vega 20 within acceptable bounds.

Again, the company says it’s looking to tweak Radeon VII via drivers to give enthusiasts options for quieter operation, but chose to focus on performance at launch given this board’s high-end pedigree. We certainly appreciate the emphasis on power users. And in a world without GeForce RTX 2080 to compare against, Radeon VII might not seem as noisy. In fact, without GeForce RTX 2080, AMD would have found it less necessary to flog Vega 20 for competitive benchmark results.

The Look and Feel of Radeon VII

Despite our critiques of Radeon VII’s balance between performance, power, and noise, we can still admire an attractive and well-conceived layout. Low-profile components leave plenty of room for a dual-slot cooler that effectively dissipates 300W of heat—no small feat.

AMD went with a more sophisticated power supply this time around. You can’t simply count its coils to know how the phases are assigned. A number of partial voltages are generated, and those all require their own circuits.

In order to generate VDDCR_GFX for Vega 20, five phases are controlled by an International Rectifier IR35217 on the PCB’s back side. With the help of one IR3599 multiplier per phase, Radeon VII ends up with 10 voltage converter circuits altogether. For this, 10 Infineon TDA21472 integrated power stages are used, each with a synchronous buck gate driver, control and synchronous MOSFETs, and a Schottky diode.

Since the IR35217 is a true multi-phase controller that can provide up to 6+2 phases, its second part is used for provisioning partial voltage for VDDCI_Mem. The voltage converter used is an IR35401M, a somewhat simpler PowIRstage.

The memory gets a little trickier. On the front side of the board, there is another IR35217 controller that generates two real (not doubled) phases for the HBM2’s four stacks, referred to as VDDCR_HBM. Two TDA21472s are directly addressed by the IR35217. This controller also provides the VDDCI_SoC (blue). Only one phase is generated for that, though, which is then doubled by an IR3599 into two phase-shifted voltage converters, each with its own TDA21472.

Other 1.8V, 0.85V, and 0.75V signals are generated separately by simpler buck controllers. It's interesting that AMD finishes all 12V rails with proper LC elements, which should help smooth out spikes.

Radeon VII: Inside and Out

At 26.8cm long, 11.5cm tall (from the upper edge of the PCIe slot to the top of the cooler), and 3.5cm thick, Radeon VII corresponds almost exactly to the dimensions of Nvidia's GeForce RTX 2080 Founders Edition. And at 1.282kg, it's also close to the same weight.

Without the PCB, AMD's heat sink, shroud, and axial fans weigh more than 1kg. AMD utilizes an oversized vapor chamber that makes direct contact with Vega 20 through special graphite pads, similar to Radeon Pro WX8200. This stuff outperforms normal thermal paste. It took an ultra-thin layer of Innovation Cooling Diamond compound to restore the card’s performance after re-assembly.

Vertically-oriented fins stretch across the cooler, drawing heat away from the vapor chamber on one side of the sink and five flat heat pipes on the other. This whole assembly is cooled by air flow from three 7.5cm fans, each of which populates an 8.2cm opening.

The GPU, memory, and SoC voltage regulation circuitry is cooled by thick heat conducting pads applied to the black mounting frame. The cooler sits on top of this frame. In turn, the frame helps stabilize the PCB underneath by sandwiching it with a cast aluminum backplate.

While the backplate certainly looks nice, it doesn’t do anything to cool Radeon VII. Fortunately, AMD cuts slits into the metal to allow some airflow and help prevent heat from building up back there.

Around front, similar-looking slits decorate the expansion bracket. Of course, because the heat sink's fins are oriented vertically, these do nothing for ventilation since hot air exhausts out the card's top and bottom. Nevertheless, AMD limits Radeon VII's display outputs to three full-sized DisplayPort connectors and one HDMI interface.

MORE: Best Graphics Cards

MORE: Desktop GPU Performance Hierarchy Table

MORE: All Graphics Content

How We Tested and Gaming at 2560 x 1440

How We Tested Radeon VII

AMD's latest and greatest is positioned primarily as a high-end gaming card, though the company also extols the virtues of its 16GB HBM2 for content creation workloads. To test Radeon VII across our gaming suite, we dropped it into our graphics workstation with an MSI Z170 Gaming M7 motherboard and an Intel Core i7-7700K CPU at 4.2 GHz. The processor is complemented by G.Skill’s F4-3000C15Q-16GRR memory kit. Crucial’s MX200 SSD remains, joined by a 1.6TB Intel DC P3700 loaded down with games.

As far as competition goes, the Radeon VII is meant to usurp GeForce RTX 2080. We also include Nvidia's GeForce RTX 2080 Ti, GeForce RTX 2070, GeForce GTX 1080 Ti, GeForce GTX 1080, GeForce GTX 1070 Ti, GeForce GTX 1070, and Titan V. The competition from AMD comes from Radeon RX Vega 64 and Radeon RX Vega 56. All cards are either Founders Edition or reference models.

Our benchmark selection now includes Ashes of the Singularity: Escalation, Battlefield V, Destiny 2, Far Cry 5, Grand Theft Auto V, Metro: Last Light Redux, Rise of the Tomb Raider, Tom Clancy’s The Division, Tom Clancy’s Ghost Recon Wildlands, and The Witcher 3.

We also ran a handful of benchmarks using Radeon VII in a more general-purpose role. For those tests, we set up a Core i7-8700K (6C/12T) CPU on an MSI Z370 Gaming Pro Carbon AC motherboard. All four of its memory slots were populated with 16GB Corsair Vengeance LPX modules at DDR4-2400, giving us 64GB total. Then, we loaded Windows 10 Professional onto a 1.2TB Intel SSD 750-series drive. Incidentally, this is the same system running our Powenetics software for power consumption measurement. On that machine, Radeon VII is compared to Titan RTX, Titan V, Titan Xp, GeForce RTX 2080 Ti, GeForce RTX 2080, and Radeon RX Vega 64.

The testing methodology we're using comes from PresentMon: Performance In DirectX, OpenGL, And Vulkan. In short, these games are evaluated using a combination of OCAT and our own in-house GUI for PresentMon, with logging via GPU-Z.

Performance Results: Gaming at 2560 x 1440

In challenging Nvidia’s GeForce RTX 2080, AMD’s Radeon VII finds itself offering plenty of performance for smooth frame rates at 2560 x 1440 with quality settings cranked all the way up across our benchmark suite (oftentimes with anti-aliasing added).

If we take the geometric mean of each game’s average frame rate, Radeon VII is 26% faster than Radeon RX Vega 64. That’s a truly impressive gain, given that we’re comparing the same architecture (with six of Radeon VII’s Compute Units disabled, even).

However, the same equation puts Radeon VII at just over 92% of GeForce RTX 2080’s average frame rate in these 11 games. It’s obviously possible to reshuffle the deck using different titles that might be better-optimized for AMD’s GCN architecture. But even parity between the two boards would be less than stellar from a card launching months later at the same price point, particularly when AMD’s biggest selling point—16GB of HBM2—isn’t as big of a factor at QHD.

Ashes of the Singularity: Escalation (DX12)

Battlefield V (DX12)

Destiny 2 (DX11)

Far Cry 5 (DX11)

Forza Motorsport 7 (DX12)

Forza Motorsport 7, a regular component of our benchmark suite, is conspicuously missing from today's review. One of the game's recent patches changed its frame time behavior, making some of our older results incompatible with newer data. More than likely, we'll replace FM7 with another race sim in the near future.

Grand Theft Auto V (DX11)

Metro: Last Light Redux (DX11)

Rise of the Tomb Raider (DX12)

Tom Clancy’s The Division (DX12)

Tom Clancy’s Ghost Recon (DX11)

The Witcher 3 (DX11)

Wolfenstein II: The New Colossus (Vulkan)

MORE: Best Graphics Cards

MORE: Desktop GPU Performance Hierarchy Table

MORE: All Graphics Content

Performance Results: Gaming at 3840 x 2160

By strategically turning down quality settings or disabling anti-aliasing, it’s possible to keep Radeon VII churning out smooth performance at 3840 x 2160.

Radeon VII’s 16GB of HBM2 and massive memory bandwidth increase the new card’s advantage over Radeon RX Vega 64. The geometric mean of its average frame rates is now almost 33% higher, exceeding 63 FPS compared to Vega 64’s 48 FPS. AMD even comes a little closer to GeForce RTX 2080 by achieving 93% of its average frame rate across our suite.

Of course, average frame rates aren’t everything. AMD knows that its big advantage is Radeon VII’s beefy memory subsystem, so the company is beating that drum hard with data showing where 8GB and 11GB cards may be overrun. We think this topic deserves further exploration, and we plan to run additional data using multiple products from both companies to isolate frame time aberrations.

Ashes of the Singularity: Escalation (DX12)

Battlefield V (DX12)

Destiny 2 (DX11)

Far Cry 5 (DX11)

Grand Theft Auto V (DX11)

Metro: Last Light Redux (DX11)

Rise of the Tomb Raider (DX12)

Tom Clancy’s The Division (DX12)

Tom Clancy’s Ghost Recon (DX11)

The Witcher 3 (DX11)

Wolfenstein II: The New Colossus (Vulkan)

MORE: Best Graphics Cards

MORE: Desktop GPU Performance Hierarchy Table

MORE: All Graphics Content

Performance Results: LuxMark, SPECviewperf, Cinema4D and Blender

LuxMark v.3.1

The latest version of LuxMark is based on an updated LuxRender 1.5 render engine, which specifically incorporates OpenCL optimizations that invalidate comparisons to previous versions of the benchmark.

We tested all three scenes available in the 64-bit benchmark: LuxBall HDR (with 217,000 triangles), Neumann TLM-102 SE (with 1,769,000 triangles), and Hotel Lobby, with 4,973,000 triangles).

Radeon VII is the first-place finisher in LuxMark's LuxBall HDR workload, beating the Turing-based Titan RTX and Volta-based Titan V thanks to a substantial memory bandwidth advantage. It doesn't fare quite as well in the subsequent tests, which are more compute-intensive.

With that said, Radeon VII also scores better than GeForce RTX 2080, its primary competition, in the Neumann TLM-201 SE and Hotel Lobby tests.

SPECviewperf 13

The most recent version of SPECviewperf employs traces from Autodesk 3ds Max, Dassault Systemes Catia, PTC Creo, Autodesk Maya, Autodesk Showcase, Siemens NX, and Dassault Systemes SolidWorks. Two additional tests, Energy and Medical, aren’t based on a specific application, but rather on datasets typical of those industries.

Oil and gas workloads tend to involve very large datasets, which play into Radeon VII's 16GB of HBM2 at 1 TB/s. The same goes for a certain medical applications. And in those tests, AMD's flagship is faster than GeForce RTX 2080.

Catia and NX, specifically, respond well to the professional driver optimizations that benefit Nvidia's Titan cards. AMD's Radeons are quite a bit slower in both benchmarks. However, the Radeon VII and Radeon RX Vega 64 make easy work of GeForce RTX 2080.

Cinema4D

ProRender is a physically-based GPU render engine. Unlike Arion Render, which we tested in our Titan RTX review, it utilizes OpenCL instead of CUDA.

Blender

Per AMD's recommendation, we tested Radeon VII and Radeon RX Vega 64 using Blender v.2.79b. In order to get CUDA acceleration from GeForce RTX 2080, we had to use v.2.80.

Rendering the bmw27_GPU test file using our Core i7-8700K took 5:16, regardless of the graphics card we had installed. Switching over to GPU acceleration through OpenCL or CUDA brought those times down significantly. Although Radeon VII trails GeForce RTX 2080, it definitely improves upon Radeon RX Vega 64's performance.

MORE: Best Graphics Cards

MORE: Desktop GPU Performance Hierarchy Table

MORE: All Graphics Content

Power Consumption

Slowly but surely, we’re spinning up multiple Tom’s Hardware labs with Cybenetics’ Powenetics hardware/software solution for accurately measuring power consumption.

Powenetics, In Depth

For a closer look at our U.S. lab’s power consumption measurement platform, check out Powenetics: A Better Way To Measure Power Draw for CPUs, GPUs & Storage.

In brief, Powenetics utilizes Tinkerforge Master Bricks, to which Voltage/Current bricklets are attached. The bricklets are installed between the load and power supply, and they monitor consumption through each of the modified PSU’s auxiliary power connectors and through the PCIe slot by way of a PCIe riser. Custom software logs the readings, allowing us to dial in a sampling rate, pull that data into Excel, and very accurately chart everything from average power across a benchmark run to instantaneous spikes.

The software is set up to log the power consumption of graphics cards, storage devices, and CPUs. However, we’re only using the bricklets relevant to graphics card testing. AMD's Radeon VII gets all of its power from the PCIe slot and a pair of eight-pin PCIe connectors. Should third-party Vega 20-based board materialize at some point in the future with three auxiliary power connectors, we can support them, too.

Gaming: Metro: Last Light

Three runs of the Metro: Last Light benchmark give us consistent, repeatable results, which makes it easier to compare the power consumption of graphics cards.

AMD extracts as much performance out of Radeon VII's power budget as possible. Through our three-run recording, the card averages almost 298W with spikes that approach 322W.

Very little power is delivered over the PCI Express slot. Rather, it's fairly evenly balanced between both eight-pin auxiliary connectors.

The blue overall power consumption line, representing the sum of all other lines, mostly obeys AMD's 300W limit.

Less impressive is the fact that Radeon VII does battle against a card rated for 75W less, manufactured on a 12nm node. Even GeForce RTX 2080 Ti, which is significantly faster, uses quite a bit less power.

At least AMD isn't in unprecedented territory. Despite a supposed 295W power limit, its Radeon RX Vega 64 demonstrated a similar power profile as Radeon VII.

Current over the PCIe slot stays just over 2A. Clearly, those eight-pin connectors do most of the heavy lifting here.

FurMark

FurMark is a steadier workload, resulting in less variation across our test run. Average power does rise slightly to 309W with spikes as high as 330W.

The much more consistent workload makes it easier to compare draw over each rail. Again, AMD achieves good balance between its two eight-pin auxiliary connectors, while the PCIe slot averages 30W.

GeForce RTX 2080 and 2080 Ti are well-behaved. They operate within a tight power range and generally obey Nvidia's limits.

Radeon RX Vega 64 has a harder time keeping up with the demands of FurMark. It starts off strong, quickly heats up, and then oscillates within a ~15W range to avoid violating one of AMD's upper bounds.

Radeon VII doesn't have the same issue. Its power consumption line chart isn't as tightly grouped. But the card still maintains its performance under that full-length heat sink and trio of axial fans.

Current draw over the PCIe slot hovers between 2A and 3A, leaving lots of headroom under the 5.5A ceiling.

MORE: Best Graphics Cards

MORE: Desktop GPU Performance Hierarchy Table

MORE: All Graphics Content

Fan Speed and Noise

AMD does not implement a semi-passive mode, which we're fine with. At idle, the three axial fans spin between 800 and 820 RPM, operating quietly. We did, however, measure some variation in their rotational speed using a laser tachometer, despite receiving the same PWM signal. At higher speeds, they give off a slight oscillating noise since their differences fall into an audible frequency.

That floating noise isn't the worst of Radeon VII's acoustic issues, though. A peek at the fan curve makes it clear how aggressive AMD is being with Vega 20's performance. In less than three seconds, the card goes from its idle fan speed straight up to its maximum. As you can imagine, this makes for a jarring experience compared to competing solutions that ramp up slowly in response to load. In both our U.S. and German labs, test notes indicate initial alarm that something was malfunctioning.

As mentioned on the first page, we approached AMD about our findings and were told that this behavior is intended. Although Radeon VII exhibits similar maximum-load acoustic performance as the reference Radeon RX Vega 64, that quick ramp from idle to 2,900 RPM explains why the new card seems so much louder. Again, the proposed solution is a manual adjustment in WattMan. Unfortunately, any attempt to relax Radeon VII's ability to cool itself off is going to cost you sustainable clock rates, negatively affecting performance.

AMD maintains that enthusiasts care more about their frame rates than noise, and we'd tend to agree. In this case, however, Nvidia's GeForce RTX 2080 Founders Edition card doesn't require the same sort of trade-off. Bear in mind that Nvidia's reference design does cost $100 more than Radeon VII. How do the two cards size up? Let's have a look at the measured values for both products, side by side.

Spectrum Analysis

The 49.2 dB(A) result is based on the 2,947 RPM fan speed observed in a closed case. We applied our gaming load with Radeon VII on an open test bench in our measurement chamber and fixed the card's fans to 2,950 RPM, replicating the same conditions.

With GeForce RTX 2080 Founders Edition subjected to similar methodology, AMD's flagship ended up a full 10 dB(A) higher. What AMD asks of its cooler in order to match Nvidia's performance is not acceptable at the same price point.

Under load, Radeon VII is simply too loud. You can tune it for quieter operation if you want. But if you ran Nvidia's GeForce RTX 2080 at the same fan speeds that AMD is using, its TU104 GPU would achieve higher GPU Boost frequencies, improving its relative position. A more fair comparison would involve benchmarking both cards at the same noise levels. After all, Radeon VII is tuned for performance, ignoring power and noise, while GeForce RTX 2080 offers a better balance between all three variables.

MORE: Best Graphics Cards

MORE: Desktop GPU Performance Hierarchy Table

MORE: All Graphics Content

Temperatures and Clock Rates

Remember that there are two representations of temperature in play here: edge, or GPU temperature, and junction temperature. In the measurements below, reported as GPU temperature, Vega 20 appears to operate coolly, even though we know the junction temperature AMD uses to control thermal throttling and fan control is quite a bit higher.

Temperatures do rise faster in the closed case and settle about 2°C higher than an open test bench. This isn't too painful at first. After all, AMD's throttling mechanism works differently from Nvidia's, where whole GPU Boost steps are lost as temperature increases.

As our test progresses and Vega 20 warms up, we see that Radeon VII starts switching back and forth between ~1500 MHz and 1740 MHz on the open bench. The jumps progressively become larger and more frequent as the temperature rises. In a closed case, the changes in frequency also seem to be dependent on the rate at which temperature changes. The card even sporadically drops to 1367 MHz!

Since fan speeds and temperatures are held within a very strict limit, clock rates are carefully balanced every step of the way. This behavior is very different from what we've seen from AMD's Radeon RX Vega and Nvidia's GeForce cards. In fact, we haven't seen such severe clock rate fluctuations from any other graphics card.

From what we can tell, Radeon VII doesn't really have a fixed boost frequency. So, you have to distinguish between peak, minimum, and average clock rates. Programs like WattMan and 3DMark reflect maximum values. But the practical relevance of those numbers is debatable. In order to make reliable observations about Radeon VII's boost clock, you need to record several seconds and calculate an average from them.

Let's again draw a comparison between Radeon VII and GeForce RTX 2080 Founders Edition using our benchmark results:

Board Analysis: Infrared Images

We brushed the board with a special lacquer before collecting measurements, and calibrated the system with thermal sensors at four unique reference points. The homogeneous coating gives us a known emissivity of 0.975, and we take readings at an angle of 90 degrees.

The following infrared images, captured during our gaming loop and stress test in a closed case and on an open test bench, convey meaningful real-world information. Compared to WattMan, which made logging data problematic at best, our infrared measurements are far more reliable.

Differences in the gaming loop between our open bench table and closed case are visible, but within the expected range of two to three degrees.

Our stress test imposes up to 20W-higher power consumption, which is dissipated as waste heat. Interestingly, the temperature we read below the GPU package was higher than the value that WattMan reported as GPU temperature.

In a closed case, everything gets hotter. What's reported as GPU temperature nearly hits 80°C, and according to AMD, junction temperature should be even higher.

Unfortunately, we couldn't log the results of our stress test with WattMan because the program returned nonsensical data. In several measurements, the software simply stopped recording after only a few lines. At least we have values from our camera.

MORE: Best Graphics Cards

MORE: Desktop GPU Performance Hierarchy Table

MORE: All Graphics Content

Conclusion

AMD’s Radeon VII announcement was a bit of a surprise to everyone. But regardless of whether you’re a gamer, a content creator, or a reviewer eager for some competition in the high-end graphics market, we can all agree that even an unexpected entry is a breath of fresh air.

The Radeon VII doesn’t quite aspire for tier-one billing. Rather, it goes up against Nvidia’s GeForce RTX 2080—a card that only recently usurped Nvidia’s former flagship, the GeForce GTX 1080 Ti. As a result, Radeon VII is ideal for gaming at 2560 x 1440 with details cranked up or 3840 x 2160 with some quality compromises to keep frame rates fluid. To sweeten the deal, AMD is offering a three-game bundle with Radeon VII that includes Resident Evil 2, Devil May Cry 5, and The Division 2—one of the better bundles we’ve seen.

Of course, bonus games are a great differentiator, all else being equal. But first, Radeon VII has to prove itself against capable competition. If we take the geometric mean of each game in our benchmark suite’s average frame rate at QHD, Radeon VII is 26% faster than Radeon RX Vega 64. That’s a truly impressive gain. However, the same equation puts Radeon VII at just over 92% of GeForce RTX 2080’s average frame rate in our 11 games. It’s obviously possible to reshuffle the deck using different titles that might be better-optimized for AMD’s GCN architecture. Granted, our benchmarks were run on Nvidia's Founders Edition card, which enjoys a 5%-higher GPU Boost rating than the least-expensive 2080s out there. But even parity between the two boards would be less than stellar from a card launching months later, particularly when AMD’s biggest selling point—16GB of HBM2—isn’t as big of a factor at 2560 x 1440.

Radeon VII’s heavy-hitting memory subsystem improves the new card’s advantage over Radeon RX Vega 64 at 3840 x 2160. The geometric mean of its average frame rates grows to almost 33% higher, exceeding 63 FPS compared to Vega 64’s 48 FPS. AMD even comes a little closer to GeForce RTX 2080 Founders Edition by achieving 93% of its average frame rate across our suite. But of course, average frame rates aren’t everything. AMD knows that its big strength is Radeon VII’s 16GB of HBM2 and 1 TB/s of bandwidth, so the company is beating that drum hard with data showing where 8GB and 11GB cards may be overrun, resulting in ugly frame time spikes that manifest as stuttering. We think this topic deserves further exploration, and we plan to run additional tests using multiple products from both companies to isolate frame time aberrations using practical examples, rather than hand-picked benchmarks conceived to make 16GB look necessary at a time when it’s especially convenient.

At least for gaming, then, we’d stop short of spending $700 on Radeon VII. The $800 GeForce RTX 2080 Founders Edition we tested tends to be a bit faster, it uses a lot less power, and it’s significantly quieter. The many 2080s available between $700 and $800 exhibit very similar attributes as Nvidia's reference design. Since two of the three games in AMD’s bundle (Devil May Cry 5 and The Division 2) aren’t available yet, there's plenty of time for additional testing before losing out on those extras. And AMD says it’s working on a potentially more elegant way to handle cooling, rather than spinning its fans up from idle to maximum speed in a matter of seconds.

The company says it controls throttling and fan control on Vega 20 through a network of 64 sensors that create a junction temperature, rather than the edge temperature reported by a single sensor that was used previously. As a result of this more informed reading, the company can be extra aggressive about extracting maximum performance from Vega 20 without exceeding the GPU’s limits. In thermally-constrained scenarios, Radeon VII should run a couple of percent faster with its clock rates modulated according to junction temperature. But because AMD is pushing this GPU hard at the expense of power and heat, there’s no real way to tame the unrefined cooler without giving up peak performance. Any attempt to make Radeon VII quieter is going to cost cooling, leading to lower clock rates. As a result, AMD is fighting an uphill battle against third-party GeForce RTX 2080s selling for $700. Even with a three-game bundle, we think it needs to be less expensive than its primary competition.

But again, that’s for gaming. Content creation is another matter entirely. We don’t have many rendering or encoding workloads in our suite. However, as we recently saw in our Nvidia Titan RTX review, extra memory makes a big difference in workloads able to utilize it. In fact, it can be the difference between a successful run and a crash. Bandwidth-intensive metrics like LuxBall HDR show that Radeon VII is capable of beating monsters like Titan RTX in the right situations. AMD also puts the hurt on GeForce RTX 2080 in the SPECviewperf 13 energy and medical viewsets, both of which presumably benefit from lots of fast on-board memory. Catia, NX, and SolidWorks go AMD’s way, too.

Our Cinema4D and Blender tests favor GeForce RTX 2080. However, once you start getting into more specialized workloads that incorporate specific optimizations for either AMD or Nvidia, recommendations become much harder to generalize. If you already know that your 8K Adobe Premiere projects utilize more memory than what you get from 8GB or 11GB cards, Radeon VII at $700 becomes an almost affordable alternative to the $2,500 Titan RTX with 24GB of GDDR6 (and a lot less memory bandwidth).

In the end, then, Radeon VII looks like the right card for the right kind of customer. By arming it with two times the HBM2 as Radeon RX Vega 64 and moving moving up to 1 TB/s, AMD uncorked its Vega 20 processor to serve up significantly more performance. Some workloads reward Radeon VII with a win over GeForce RTX 2080. Others show Nvidia retaining its lead. But AMD is clear that average frame rates don’t tell the whole story. Under certain conditions, 16GB of memory may also be the key to lower frame times and smoother performance. AMD didn’t give us enough time with Radeon VII to dig deep into those claims. We certainly plan to, though. And perhaps by the time we’re done, the company will have fixed the hardware monitoring issues we encountered and smoothed out Radeon VII’s meteoric fan curve.

MORE: Best Graphics Cards

MORE: Desktop GPU Performance Hierarchy Table

MORE: All Graphics Content

The Value-Oriented WX-Series Option

AMD's second-gen Ryzen Threadripper family was introduced to the world in the form of a 32-core, 64-thread 2990WX model priced at $1800. It set new performance records across workloads able to exploit the chip's copious resources. However, the flagship Threadripper chip's unique architecture also causes odd results in more common desktop applications. Consequently, we only recommend the 2990WX to professionals running certain workstation-class software.

The $1300 Ryzen Threadripper 2970WX includes 24 cores and 48 threads. It bears the same WX suffix meant to signal an affinity for heavy multitasking and professional workloads. Moreover, the 2970WX boasts more on-die resources than Intel's $2000 Core i9-7980XE, which offers 18 Hyper-Threaded cores.

Similar to Ryzen Threadripper 2990WX, two of the 2970WX's dies aren't connected directly to main memory. So, the CPU delivers great performance in threaded workloads that aren't sensitive to memory throughput, but less impressive results in bandwidth-hungry applications that don't scale well with extra cores. AMD introduced Dynamic mode to its Ryzen Master software in an effort to minimize the architecture's compromises, but it isn't always effective.

Given its similarities to the 2990WX, it's no surprise that Threadripper 2970WX demonstrates a lot of the same behaviors in our benchmark suite. You still need a particular type of workload to maximize its potential. Fortunately, if you have the right software, Ryzen Threadripper 2970WX offers a much less expensive route to 2990WX-like performance.

Ryzen Threadripper 2970WX

Earlier this year, AMD retooled its mainstream Ryzen line-up with new Zen+ optimizations that included 12nm manufacturing, improved memory and cache latency, higher clock rates, and enhanced multi-core Precision Boost frequencies. Those changes carry over to the company's newest Threadripper models, too.

AMD also split its Threadripper portfolio into the WX and X families. The two WX models are geared toward intense multitasking workloads, 3D rendering, media encoding, and cinema mastering. That makes them attractive to software developers, video/audio engineers, and content creators.

Ryzen Threadripper 2970WX is AMD's second quad-die processor for high-end desktops. Again, it sports 24 cores and 48 threads. A 3 GHz base frequency stretches as high as 4.2 GHz via AMD's XFR (eXtended Frequency Range) algorithms. The processor also features an improved Precision Boost 2 technology for achieving more aggressive multi-core turbo clock rates compared to the first-gen models.

Each of the WX CPU's four dies boast eight physical cores and 16MB of L3 cache. Thus, Threadripper 2990WX and 2970WX are both armed with 64MB of L3 cache. That's generous on AMD’s part, since Intel typically disables cache as it turns off cores to create lower-end models. Of course, AMD does carve out two cores per die to create the 2970WX's 24-core configuration, though. And like the 2990WX, Ryzen Threadripper 2970X is rated at 250W.

The dual-die X-series Threadrippers are better suited to enthusiasts and gamers. AMD launched its Ryzen Threadripper 2950X in September, but now there's a 12C/24T Threadripper 2920X available as well. It includes six cores per die and the same 32MB of L3 cache as the 16C/32T 2950X. Both X-series models are rated at 180W.

AMD ships all Threadripper CPUs with an Asetek bracket that provides partial coverage of the expansive heat spreader using compatible closed-loop liquid coolers. According to AMD, this partial coverage is fine for stock operation. But we found that full-coverage coolers work better. AMD also collaborated with Cooler Master to develop the Wraith Ripper heat sink/fan combo for its Socket TR4 interface. It's sold separately, though.

Of course, AMD uses Indium solder between its dies and heat spreader to improve thermal transfer. In contrast, Intel employs thermal grease and recommends liquid cooling for its Skylake-X processors. AMD says that's not necessary for Threadripper. Intel recently added Indium solder to its Core i9 series, so we may see this feature work its way up into the HEDT segment before long.

All of the second-gen Threadripper processors are backward-compatible with existing X399 motherboards. But older Socket TR4-equipped boards may struggle under the power requirements of AMD's 250W Threadripper WX series chips, particularly if you try to overclock. Consider shopping for a new X399-based platform if tuning is on the menu.

Familiar AMD value-adds abound on the 2970WX: you get an unlocked ratio multiplier for overclocking, the new Precision Boost Overdrive automated overclocking feature, Ryzen Master software, and 60 lanes of third-gen PCI Express (plus four lanes attached to the supporting chipset). Copious connectivity could come in handy for multiple add-in graphics cards, but it's also useful for high-performance storage and networking.

Threadripper CPUs feature independent dual-channel memory controllers located on two dies, which combine to provide quad-channel support with varying data transfer rates based upon your configuration. With the second-gen Threadripper processors, AMD bumps its maximum specification to DDR4-2933 (up from DDR4-2666). The platform supports ECC memory and up to 256GB of capacity, but it can accommodate up to 2TB as density increases.

MORE: Best CPUs

MORE: Intel & AMD Processor Benchmark Hierarchy

MORE: All CPUs Content

Dynamic Mode, Overclocking and Test Setup

AMD’s Threadripper processors employ a unique Multi-Chip Module (MCM) architecture that enables impressive modularity, but also hurts performance in certain workloads. The company masks much of this on the dual-die X-series chips. However, its WX series' four dies present new challenges. We previously covered the design's specifics. In short, though, two of the four dies are only used for their x86 cores, while the other two have active memory and PCIe controllers.

Windows' round robin thread scheduling mechanism tends to push important threads off of the I/O dies, requiring memory-hungry applications to access another die during execution, thereby hurting performance. AMD originally created a couple of operating modes to let its customers tailor the way they wanted Threadripper processors to behave. This did help side-step some of those compromises. But switching between the two modes required rebooting. Moreover, they didn't completely solve AMD's performance issues.

A new Dynamic Local Mode, which is strictly for Ryzen Threadripper 2990WX and 2970WX processors, runs as a background service inside the operating system and automatically detects memory-starved application threads (the top 13 to 16). It dynamically assigns them to dies with local memory controllers. Or, it can detect threads that aren't as sensitive to memory latency and move them to dies without memory controllers, thus optimizing the processor’s execution resources. 

This new implementation is transparent, and can be switched on without rebooting. AMD doesn't quantify the overhead of this service. However, we observed ~0.5% processor and 1MB memory utilization during normal use with the 2970WX.

For now, the service is enabled in AMD's Ryzen Master software. But the company plans to bake this functionality in to its chipset at some point in the future. The program works best with "mid-threaded" applications (as opposed to lightly-threaded ones). It also ignores apps that run on all cores and threads.

The performance measurements in the above chart were generated by AMD. We have our own tests on the following pages.

Overclocking

We tested several configurations, but stuck with Precision Boost Overdrive (PBO) for all of our tuned Threadripper WX-series configurations. This automated feature overclocks the processor to to its fullest based upon available current, power, and thermal headroom. Due to cooling and power delivery constraints, we ran through our full test suite at stock settings and with PBO activated, rather than using an all-core overclock. Our PBO-enabled configurations did benefit from higher memory transfer rates, as detailed in the table below. As with any overclocking feature, using PBO voids your warranty. 

Comparison Products

Test Setup

We tested the second-gen Threadripper models with MSI's MEG X399 Creation motherboard.

MORE: Best CPUs

MORE: Intel & AMD Processor Hierarchy

MORE: All CPUs Content

VRMark, 3DMark, AotS: Escalation and Dawn of War III

Test Notes

Unlike AMD's previous-gen Threadripper models, the WX-series CPUs include a Game Mode preset in the Ryzen Master software that disables three of four dies. Company representatives tell us this facilitates optimal performance in games. But AMD also provides toggles that allow experimentation with two- and four-die configurations. For this review's gaming benchmarks, we tested the Threadripper processors with Game Mode enabled. Ryzen Threadripper 2970WX consequently becomes a 6C/12T CPU for those tests.

Dynamic Local Mode is a new feature that debuts with the 2970WX, and it only applies to the WX-series models. We include a second bar chart for each game to quantify performance with this feature enabled or disabled at stock settings (labeled Creator/DM).

Gaming performance is measured at 1920x1080, minimizing graphics bottlenecks. Of course, as you step up to 2560x1440 or 3840x2160, the differences between processors shrink.

We have application test results with an overclocked Ryzen Threadripper 2920X, but weren't able to run that chip in its tuned state through our game benchmarks before it stopped working. Once we're able to get it back up and running, we'll update the gaming charts.

VRMark, 3DMark

We aren't big fans of using synthetic benchmarks to measure performance, but 3DMark's DX11 and DX12 CPU tests provide useful insight into the amount of horsepower available to game engines.

Ryzen Threadripper 2970WX's Game Mode drops the 24-core chip's count down to six active cores, but also prevents bandwidth-starved execution resources from handicapping performance during the DX11 and DX12 tests. Nevertheless, the 2970WX drops to the bottom of our chart. A dual-die 2920X easily beats the 2970WX, so it's clear that the quad-die CPU's topography is the issue.

UL's VRMark test lets you gauge your system's suitability for use with the HTC Vive or Oculus Rift, even if you don't currently own an HMD. UL defines a passing score as anything above 109 FPS. Precision Boost Overdrive yields a 13.6 FPS gain for the Threadripper 2970WX.

Ashes of the Singularity: Escalation

Ashes of the Singularity: Escalation is a computationally intense title that normally scales well with thread count.

Again, Ryzen Threadripper 2920X facilitates better performance than the 2970WX at stock settings. If there's a silver lining here, it's that PBO pushes today's subject beyond a stock 2990WX.

The second slide shows us that Dynamic Local Mode, which we used in tandem with Creator mode, imposes a slightly lower frame rate. Dynamic Local Mode's background process avoids shuffling threads that fully utilize the CPU's resources. Ashes of the Singularity does this, so the result isn't entirely surprising.

It is clear, however, that the 2970WX doesn’t scale linearly as cores and threads are added in Creator mode. You'll get similar (or better) performance in Game mode with just six cores active.

Warhammer 40,000: Dawn of War III

Threadripper processors perform well in this title. The 2970WX lands close to the top with PBO enabled, though the much cheaper Ryzen 7 2700 isn’t far behind.

The 2970WX in Game mode is faster than the same chip in Creator mode, even after we enable AMD's new Dynamic Local Mode. The company touted that switch's ability to improve performance "up to 49%,” but remember that those gains happen in Creator mode. You probably don't want to game in Creator mode.

MORE: Best CPUs

MORE: Intel & AMD Processor Hierarchy

MORE: All CPUs Content

Far Cry 5, GTA: V and Hitman

Far Cry 5

Intel's Core i9-9900K is fastest in Far Cry 5. But Ryzen Threadripper 2970WX is also impressive after a bit of overclocking. Really, though, there aren't big differences between most of our test field.

While Far Cry 5 is one of the most repeatable metrics in our suite with a variance of less than 0.5 FPS between runs, we did notice some oddities after enabling Dynamic Local Mode. For instance, the results alternated between 60 and 65 FPS. This repeatable phenomenon persisted after several retests. We also recorded much lower 99^th percentile frame rates and uneven gameplay with the feature active. AMD tells us that Dynamic Local Mode is still being optimized for a broader selection of applications and games, so our observation will likely by rectified in the future. For now, Game mode is the go-to choice, as evidenced by the test results.  

Grand Theft Auto V

Grand Theft Auto V favors Intel architectures and, more generally, multi-core designs with high clock rates.

An overclocked Ryzen Threadripper 2970WX slides past the Core i9-7900X, while the 2990WX doesn’t benefit as much from tuning. Once again, Game mode proves to be the obvious choice for gaming (despite a larger gain from Dynamic Local Mode this time around).

Hitman

Our Hitman benchmark was rendered almost useless by a patch that imposed a 90 FPS performance cap. A subsequent update restored our test to its prior glory.

Hitman responds well to high core counts and clock rates, so it isn’t surprising to find the overclocked Core i9-7960X in first place. The Core i9-9900K is impressive even in stock form, and the Ryzen 7 2700X proves to offer great bang for the buck.

MORE: Best CPUs

MORE: Intel & AMD Processor Hierarchy

MORE: All CPUs Content

Shadow Of War and Project CARS 2

Middle-earth: Shadow Of War

Shadow of War leans heavier on graphics resources than host processing, so we don't see large deltas between the fastest and slowest CPUs. The same observation applies to our experiments with Creator mode, Dynamic Local Mode, and Game mode. 

Project CARS 2

Although Project CARS 2 is purportedly optimized for threading, clock rates obviously affect frame rates. If you’re gaming in Creator mode, the Dynamic Local feature serves up a big speed-up. But the standard Game mode still offers the best performance.

MORE: Best CPUs

MORE: Intel & AMD Processor Hierarchy

MORE: All CPUs Content

Office and Productivity

Test Notes

We tested all Threadripper processors in Creator mode during our application benchmarks.

Web Browser

The Krakken suite evaluates JavaScript performance using several workloads, including audio, imaging, and cryptography. Like most Web browser workloads, single-threaded performance reigns supreme.

AMD's second-gen Threadripper line-up goes a long way to improve the performance of lightly-threaded workloads, but Intel still leads in these tests.

Ryzen Threadripper 2970WX lags the 2990WX in our Krakken benchmark, but automated overclocking functionality does deliver a solid improvement. Meanwhile, there's little to no improvement in the lightly-threaded Krakken from enabling Dynamic Local Mode. This feature does seem to help in MotionMark and WebXPRT. 

Productivity

The application start-up metric measures load time snappiness in word processors, GIMP, and Web browsers under warm- and cold-start conditions. Other platform-level considerations affect this test as well, including the storage subsystem.

Ryzen Threadripper 2970WX benefits from Dynamic Local Mode at stock clock rates, but still trails the 2990WX right out of its box. Interestingly, the 2920X and 2950X yield the best performance from any Threadripper CPU. Other desktop processors like the Ryzen 7 2700X and Core i9-9900K fare well, too.

Our video conferencing suite measures performance in single- and multi-user applications that utilize the Windows Media Foundation for playback and encoding. It also performs facial detection to model real-world usage. Again, mainstream processors offer the best value in these types of applications.

The photo editing benchmark measures performance with Futuremark's binaries using the ImageMagick library. Common photo processing workloads also tend to be parallelized. Unfortunately, the Dynamic Local Mode doesn't benefit all applications. It doesn't necessarily hurt performance, either. The 2% delta between our stock 2970WX falls within UL's 3% threshold for run-to-run variability with PCMark 10.

MORE: Best CPUs

MORE: Intel & AMD Processor Hierarchy

MORE: All CPUs Content

Rendering, Encoding and Compression

Rendering

Many of these workloads stress the memory subsystem, diminishing Ryzen Threadripper 2990WX's core count advantage due to accesses from the remote memory controllers. The 2970WX benefits from higher clock rates to surpass the 2990WX in our single-threaded POV-Ray and Cinebench tests. Both are closely matched after we active PBO, though.

If you're looking for single-threaded supremacy, Intel's ninth-gen chips cannot be beaten, as evidenced by Core i9-9900K's dominance.

Threaded workloads are an ideal match to Threadripper's high core counts. But the Ryzen Threadripper 2990WX doesn't scale linearly in these types of workloads. That means the 2970WX's $500-cheaper price is attractive for high-end desktop PCs.

Encoding & Compression

Our compression and decompression metrics work directly from system memory, removing storage throughput from the equation. This workload should benefit from threading. But either memory throughput or poor software scaling holds the 2990WX back from realizing its potential in the compression test. The same issue affects Ryzen Threadripper 2970WX, though Dynamic Local Mode provides a slight performance uptick. But neither the 2970WX nor the 2990WX are a match for the 2920X and 2950X with their dual-die architectures. Conversely, the WX chips dominate our decompression tests, illustrating the performance trade-offs AMD's highest-end CPUs force you to make.

y-cruncher, a single- and multi-threaded program that computes pi using AVX instructions, is a great test to measure Threadripper’s AVX performance. Intel’s Core i9 employs two 256-bit AVX FMA units per core that operate in parallel, whereas Ryzen's Zen architecture divides 256-bit AVX operations across two FMA units per core. Intel's AVX instruction support shines during the single-threaded benchmark. However, spreading the workload across Threadripper's many cores helps improve its standing. Despite an eight-core advantage, the 2990WX offers little benefit over the 2970WX in the threaded y-cruncher test. Clearly, their cores suffer from a memory bandwidth bottleneck. 

MORE: Best CPUs

MORE: Intel & AMD Processor Hierarchy

MORE: All CPUs Content

Workstation Compute and Graphics

Workstation Compute

Many workstation applications scale very well with additional cores in certain workloads or with special plugins, but the result is always the sum of many factors and tasks in which the pure computing power of all the cores is important, but even so also not crucial. Often enough, the parallelizable tasks do not scale beyond a certain number of cores / threads, so IPC will co-decide. And that's not the advantage of AMD.

The dynamic mode of the Ryzen TR 2970WX is another such thing in its own right, because between the individual iterations of a benchmark (between 3 and 5, depending on the application), it sometimes comes to very clear differences. We can only explain it again with the missing memory controller, since many AVX- and SSE-optimized codes (but not only those) depend on memory bandwidth. And when a software solution such as Dynamic Mode intervenes, the well-intentioned can sometimes turn into the opposite.

Workstation Graphics

While workstation graphics are a niche for most readers, some might consider using Threadripper 2970WX's twelve cores and 24 threads for professional tasks. Really, though, there aren't many threaded applications for real-time graphics output. These benchmarks mostly benefit from high IPC and frequency, which isn’t one of Threadripper’s inherent strengths. The results are not bad, but also not outstanding.

Nevertheless, there are also applications that have to calculate in parallel and are grateful for every additional thread. AutoCAD is just an example of the clock dependence of fewer threads when it comes to pure 2D drafting or real-time 3D graphics output. The graphic performance is very reminiscent of the general result in gaming, it doesn’t matter if you use DirectX, OpenGL, or just the Windows GDI.

MORE: Best CPUs

MORE: Intel & AMD Processor Hierarchy

MORE: All CPUs Content

Power Consumption

Interestingly, AMD reduced the idle power consumption of its second-gen Threadripper CPUs a bit. This is certainly motherboard-dependent, so be sure you're using the latest BIOS on your X399-based motherboard.

Just be ready for Windows 10 to bounce you back to 25-40W as background processes kick on and off (and particularly with PBO enabled for additional performance).

Ryzen Threadripper 2920X with PBO enabled hits clock rates as high as 4.3 GHz in our CAD workload, so power consumption spikes as well. The 2970WX is similar in that its high boost frequency results in a >64W measurement.

Our Witcher 3 benchmark is great for GPU power consumption comparisons. But it's less enlightening when we run it across different host processors because the underlying engine doesn't use enough cores. Assassin's Creed Odyssey is better about utilization, causing Threadripper's power to spike in excess of 100W. But that game's averages are far from reproducible, forcing us to keep it on the shelf.

Prime95 certainly isn't representative of an everyday scenario, but it's great for measuring peak theoretical power consumption. Some real-world applications pull even more power. For instance, a real Blender workload drives the 2970WX up to 230W, while we only measured 210W with Prime95! The same applies when we switch on PBO: Blender coaxes 447W from the Ryzen Threadripper chip compared to 416W in Prime95.

The smaller Ryzen TR 2920X is slightly different. It pulls 160W during the Blender workload, whereas Prime95 shows a power consumption of 180W. Activating PBO increases the delta: we measure 200W during the Blender test and 249W in Prime95. We can't explain why both CPUs react differently, but their results are reproducible.

MORE: Best CPUs

MORE: Intel & AMD Processor Hierarchy

MORE: All CPUs Content

Temperatures

If you really needed to, you could equip the Ryzen Threadripper 2970WX with a potent air cooler, so long as you don't overclock manually or activate Precision Boost Overdrive. However, a large Blender workload would completely overwhelm it. Conventional heat sinks and fans just aren't up to the task.

Ryzen Threadripper 2970WX can hit 450W during everyday operation. That amount of waste heat requires more than air or a compact all-in-one liquid cooler. You need a much more capable thermal solution if you want to unlock the CPU's maximum performance potential.

AMD deserves credit for this processor's finely-tuned protection mechanisms. Even with PBO active, you can easily push Threadripper 2970WX to the limits of a weaker cooler without damaging it. But you have to give up proper performance in return.

XFR2 and PBO both work well on a platform with ample cooling. The chip adjusts its voltages based on telemetry data, and PBO is actually preferable to manual overclocking. While we're not fans of hidden mechanisms, PBO does exactly what you expect.

AMD prioritizes package temperature: all measurements and information are based solely on this T_die reading. For compatibility reasons, the 27°C-higher T_ctl value is used for fan control. AMD sets the upper limit for Ryzen Threadripper 2970WX at 68°C, which translates to a T_ctl value of 95°C.

At idle, our cooler keeps Ryzen Threadripper 2920X below 25°C. Under a real-world Blender workload, we average about 43°C with the CPU keeping all cores at 4 GHz. With PBO active, the 2920X accelerates to 4.15 GHz across all of its cores. The average T_die rises to just over 49°C. This gives us a maximum delta  of less than 30°C for our potent cooling solution.

During a normal Blender workload without PBO, Ryzen Threadripper 2970WX reaches 3.55 GHz on all cores and averages 48°C. With PBO turned on, the all-core clock rate jumps to 4.025 GHz at an average temperature of 62°C.

The CPU does peak at 68°C though, meaning our sample is at its limit for full performance. Any higher and it would need to throttle back a bit.

MORE: Best CPUs

MORE: Intel & AMD Processor Hierarchy

MORE: All CPUs Content

Final Analysis

Ryzen Threadripper 2970WX marches onto the HEDT scene with 24 cores and 48 threads. That's more cores than any competing Intel processor. But not all cores are created equal. Intel still gets more work done per clock cycle with its ninth-gen Core CPUs than AMD does with second-gen Ryzen. What's more, the 2970WX exhibits the same idiosyncrasies as the flagship Ryzen Threadripper 2900WX. It's an impressive performer in heavily-threaded workloads that aren't memory-bound, but it struggles in some applications that don't scale well based on core count (particularly if they're sensitive to available memory bandwidth).

More mainstream Ryzen and Core CPUs offer better value to gamers. Even the X-series Threadripper processors are smarter purchases than the halo WX models if you're an enthusiast. AMD's less expensive Threadrippers promise a better experience in the lightly-threaded apps that Intel continues to dominate.

But maybe Ryzen Threadripper 2970WX serves as a less expensive entry point for professionals able to exploit its copious core count in workstation-class software. The $1800 Threadripper 2990WX doesn’t always scale well, particularly in AVX-heavy tasks like HandBrake. If you're going to have to make compromises like that, you might as well save some money on the $1300 2970WX and get similar performance in the apps able to utilize its quad-die design effectively. 

The X399 platform is expensive, and while drop-in compatibility with existing motherboards is a big advantage for AMD, you need a board with robust power circuitry. You also want a power supply with two EPS connectors. Cooling is a little easier on AMD's HEDT CPUs than Intel's competing Skylake-X chips, largely due to the Indium solder that AMD uses between its dies and heat spreader. But a beefy water cooler is still almost mandatory if you plan on overclocking.

AMD’s Dynamic Local Mode feature attempts to circumvent the performance issues endemic to its quad-die topology. Unfortunately, the mode is not as effective as we hoped it'd be. Some games do register big benefits. However, they still mostly trail the performance you get by flipping the CPU into Game mode via Ryzen Master, disabling 75% of the 2970WX's cores. AMD claims Dynamic Local Mode's background service will improve over time as the company characterizes more applications. Still, we see this as a bandage for the inconvenience of having to change modes and reboot your PC.

The competition isn't sitting still. Intel continues to get more competitive as it tries winning back the hearts and minds of enthusiasts. Its Basin Falls/Skylake-X Refresh processors should arrive next month. They'll still top out at 18 cores and 36 threads, and undoubtedly bear the company's notoriously high prices. But they are also rumored to employ Indium solder for improved thermal dissipation. That could make the new CPUs more attractive to tuners. It'd also be nice to see higher multi-core Turbo Boost bins that bolster performance in lightly-threaded workloads. Intel is also eschewing the practice of disabling PCIe lanes on less expensive HEDT processors, which is obviously a response to AMD’s practice of exposing all 60 PCIe lanes on every Threadripper model.

For now, AMD's Ryzen Threadripper 2950X, and as an extension the 2920X, offer the best value for high-end desktop PCs. Much like the Threadripper 2990WX, the 2970WX we tested today is a niche product for professionals seeking very specific capabilities. The competitive landscape is changing though, so we'd recommend waiting this one out.

MORE: Best CPUs

MORE: Intel & AMD Processor Hierarchy

MORE: All CPUs Content

WorldViz today revealed the 6^th edition of its Vizard Python-based, scientific-grade virtual reality (VR) development platform. The latest version offers support for Microsoft’s Windows Mixed Reality platform, HTC’s Vive Tracker system and a handful of new peripherals.

WorldViz provides immersive technology solutions for major companies and high-profile universities. The company offers projector-based CAVE systems for simulation, and in recent years it has embraced VR for enterprise use. Two years ago, WorldViz integrated support for the HTC Vive and Oculus Rift platforms into its offerings, and now the company is expanding its hardware support to reflect the current state of the industry. 

New Device Support

In addition to support for Windows Mixed Reality headsets and HTC's Vive Tracker system with support for hand-and-foot tracking for motion capture, WorldViz Vizard has also added support for a handful of new peripherals and VR-related technologies, such as Manus VR Gloves for better manual manipulation of the virtual world and Tobii Eye-tracking sensors to enable gaze-based interactions. WorldViz Vizard also works with biophysical electroencephalography (EEG, for recording electrical brain activity), electrocardiography (EKG, for recording electrical heart activity) and galvanic skin response (GSR senors, for monitoring skin's electrical resistance).

Better Graphics

WorldViz Vizard 6 also features improved graphics over previous iterations. The software features a robust graphics solution, but it doesn’t rely on existing game engines. WorldViz built Vizard in Python because most research scientists are familiar with the open-source language. The company said that Vizard 6 features support for the new GLTF 3D model format, which “improves the graphics rendering.” The new format also simplifies the process of bringing 3D files into Vizard from third-party modeling applications. The software supports imported files from Revit, Solidworks, Maya, Blender, SketchUp, Substance Painter, Modo and other 3D applications. Additionally, you can import models from the Sketchfab library.

Custom Avatars

WorldViz Vizard 6 also features support for developing collaborative, multi-user experiences. Earlier iterations of the platform offered a handful of generic avatars, but now you can customize your virtual appearance, which makes it easier to identify who you’re interactive with in the virtual world. Vizard 6 still includes the generic avatars, but you can now create character models with Adobe Fuse CC and import them into your virtual experiences. 

Available Now

WorldViz Vizard 6 is available now. The company has a free version for evaluation purposes, as well as a development edition, which offers the core features, and an enterprise edition, which includes support for real-time motion capture and Vizard script clustering. Visit WorldViz’s website for more information.

Intel Core i7-8086K 40th Anniversary Model

Intel's 8086, the company's first processor to use its ubiquitous x86 instruction set architecture, debuted on June 8, 1978. Forty years later and by some stroke of fortuitous timing, Intel's desktop CPU portfolio is loaded with eighth-generation Core processors. So it was only fitting, then, that after a bit of prodding by a well-known chip analyst, Intel announced that it'd pay homage to the 8086 with a 40th-anniversary limited-edition Core i7-8086K.

Core i7-8086K is based on the same Coffee Lake architecture as Core i7-8700K, right down to its six Hyper-Threaded cores able to work on 12 threads concurrently. But it features a higher base frequency and more aggressive Turbo Boost bins, which tell us that Intel carefully picked out the best dies to use in these chips. This is the first Intel processor to ship with a 5 GHz Turbo Boost bin, matching AMD's record with the FX-9590. And if you're only looking at clock rate, the -8086K represents a 1000x multiplication of the original 8086's 5 MHz frequency.

Incidentally, the -8086K is also Intel's first six-core processor with a 4 GHz base frequency, though that specification isn't as eye-catching.

Intel kicked off its anniversary celebration with a giveaway of 8086 Core i7-8086Ks. If you didn't win one, you'll have to purchase the processor like we did. Your window of opportunity won't be large, though: our sources confirm a production run of just 50,000 units. We expect collector's items to sport premium pricing, and Intel doesn't disappoint in that department. As of this writing, the -8086K sells for $75 more than the once-flagship Core i7-8700K.

So what is this processor's appeal, other than the obvious nostalgia? Core i7-8086K comes from a higher-quality bin than Core i7-8700K, so enthusiasts with deep pockets can expect to receive the very best example of Coffee Lake silicon available. Of course, most folks won't consider the extra $75 worth paying for moderate gains at stock clock rates. But again, this is a limited-edition piece of hardware steeped in history.

Intel Core i7-8600K

The 6C/12T Core i7-8086K is manufactured on Intel's 14nm++ process, just like its other Coffee Lake CPUs. Like the company's Core i7-8700K, its 95W Core i7-8086K also features 13MB of L3 cache, support for up to 64GB of dual-channel memory at DDR4-2666, an unlocked multiplier to facilitate overclocking, and Intel's integrated UHD Graphics 630 engine that can boost up to 1.2 GHz. For more information about the Coffee Lake architecture, check out our Core i7-8700K review.

The -8086K's real differentiation involves its modified Turbo Boost frequencies. But in an effort to maintain a 95W thermal design power rating, Intel only increased this chip's base clock rate by 300 MHz. Intel also increased the single-core clock rate to 5 GHz. We were able to sustain 5 GHz in tasks confined to a single core, such as Cinebench and LAME. However, the busy scheduling environment in a modern desktop operating system, which finds threads migrating frequently between cores, prevented 5 GHz operation in even mainstream tests like our gaming benchmarks. In other words, don't expect to see 5 GHz very often.

We've heard reports that some motherboards don't support Intel's 5 GHz single-core Turbo Boost bin. However, updated firmware could fix that in the future. Regardless, it's a shame that Intel didn't port over Turbo Boost 3.0 technology to pin lightly-threaded tasks to the CPU's fastest core. Overclockers might have more luck coaxing higher clock rates from the -8086K: our sample easily stretched up to 5.1 GHz with a bit of extra voltage.

We normally don't cover processor packaging, but it is relevant given the Core i7-8086K's status as a collector’s item. Like all of Intel's K-series SKUs, the -8086K doesn't include a bundled heat sink or fan.

The box tell us us that this is a limited-edition CPU. Intel even includes a certificate of authenticity, along with a signed statement from former CEO Brian Krzanich.

MORE: Best CPUs

MORE: Intel & AMD Processor Hierarchy

MORE: All CPUs Content

Forty Years Of x86

Compared to the 16-bit 8086, Intel's Core i7-8086K represents a quantum leap in technology. Whereas a modern CPU can spend four years in the design process, Intel brought its 8086 to market in just 18 months. Stephen Morse, then 36 years old, was the lead architect. The 8086 was originally designed to be a filler product before Intel released the 8800, but Morse designed it to be the first in a line of chips that shared a common architecture to ensure forward compatibility.

And thus, the x86 instruction set architecture was born. Over the course of 40 years, Intel continually enhanced the x86 ISA, adding more than 3500 new instructions like MMX, SSE, TSX, and three flavors of AVX, among many others. Amazingly, the 64-bit Core i7-8086K is capable of running original 16-bit 8086 code. That's a testament to the x86 instruction set's longevity.

The original 8086 was fabbed on a 3200nm nMOS process using mercury arc lamps. Meanwhile, 40 years later, Intel is on its third-gen 14nm CMOS process that's manufactured with argon fluoride exerciser lasers.

Transistor measurements are no longer based strictly on feature sizes, but we can derive some basic comparative metrics. Die sizes have increased from the 8086's 33mm² to the -8086K's 149mm², and transistor counts are up from 29,000 to ~3,000,000,000 per processor, respectively. That means the original 8086 featured 879 transistors per square millimeter, while Core i7-8086K comes with 20,134,228 transistors per square millimeter for an astounding 22,905x density increase.

Interfaces have also changed as Intel added more cores, cache, new buses, expanded memory support, and on-die graphics. The original 8086 dropped into a 40-pin quasi-PGA interface, whereas the eighth-generation Core processors employ an LGA 1151v2 socket that boasts 1151 pins. If we widen the scope to Intel's 28-core enterprise behemoths, some interfaces pack a whopping 4637 pins.

MORE: Best CPUs

MORE: Intel & AMD Processor Hierarchy

MORE: All CPUs Content

Silicon Lottery, Overclocking & Test Setup

Alternately, Silicon Lottery procures batches of processors and delids them to replace Intel's thermal paste with liquid metal Thermal Grizzly Condoctonaut. According to the company, this reduces operating temperatures by 15°C to 25°C, depending on the workload. The improved thermal transfer material helps facilitate more aggressive overclocks. Silicon Lottery sells the modified processors at a premium price, and with a one-year warranty (rather than Intel's standard three-year coverage).

Silicon Lottery compiles statistics about the samples it modifies and shares them publicly, giving us a reasonable gauge of what's coming out of Intel's foundries. Some enthusiasts speculate that reserving the highest-quality silicon for Core i7-8086K would hurt the chances of scoring a higher-clocking -8700K. But as we can see, the percentage of -8700Ks able to hit anywhere from 5 to 5.2 GHz actually increased during the period of time we would have expected Intel to set aside top-binned dies for its -8086K. Then again, it looks like samples able to hit 5.3 GHz disappeared entirely, possibly representing those precious -8086K-capable dies.

Silicon Lottery also shares statistics on the Core i7-8086K, and its probability of receiving top silicon is markedly better than what we see from the latest round of Core i7-8700K data. Nearly all of the company's -8086Ks reach 5 GHz, and the top 14% are capable of reaching 5.3 GHz.

Our own Core i7-8086K achieved 5.1 GHz with a 1.35V Vcore and default load line calibration settings. In addition, we adjusted our AVX offset by -1 and saw a peak temperature of 86°C during AVX-heavy workloads using Corsair's beefy H115i closed-loop cooler. Although we successfully dialed in DDR4-3466 rates with 14-14-14-24 timings, we feel we could have pushed even higher with more time for tuning.

Instead of splurging on a Core i7-8086K, you could always purchase a modified Core i7-8700K from Silicon Lottery capable of hitting the same 5.1 GHz that we achieved. Unfortunately, those models sell for $479, making the -8086K's $425 price tag attractive in comparison. If you're chasing the highest overclock possible, the company does sell a Core i7-8086K capable of 5.3 GHz for $849. As with all Silicon Lottery chips, however, you lose two years of warranty coverage in the exchange. 

Comparison Products

Test Systems

Like many other vendors, MSI motherboards feature a default Enhanced Turbo feature that allows the processor to run at its maximum Turbo Boost bin on all cores, at all times. For the Core i7-8086K, you're looking at 5 GHz across all six cores.

This setting modifies the CPU's clock rate and voltage to deliver higher performance, which is basically factory-sanctioned overclocking. Again, MSI enables it by default in the BIOS, similar to most of the competition. But performance, power consumption, and heat are all affected when it's on. We manually disable the feature for our stock CPU testing to best reflect Intel's specifications.

MORE: Best CPUs

MORE: Intel & AMD Processor Hierarchy

MORE: All CPUs Content

VRMark, 3DMark & AotS: Escalation

VRMark & 3DMark

Out of the box, Core i7-8086K effectively tied Intel's Core i7-8700K in the DX12 CPU benchmark. But its higher overclocked frequency outstripped the tuned -8700K by a decent margin. Both processors traded places in the DX11 test, though again, overclocking propelled Core i7-8086K past the -8700K.

VRMark found the -8086K and -8700K offering virtually the same performance at stock and overclocked settings. The Core i5-8600K and -8400 both beat the Core i7 models though, suggesting that this benchmark rewards configurations without simultaneous multi-threading technology exposing logical cores. 

Ashes of the Singularity: Escalation

At stock and overclocked settings, Core i7-8086K and the Core i7-8700K performed almost identically in Ashes of the Singularity: Escalation.

MORE: Best CPUs

MORE: Intel & AMD Processor Hierarchy

MORE: All CPUs Content

Civilization VI Graphics & AI, Dawn of War III

Civilization VI AI Test

Civilization's AI test measures performance in a turn-based strategy game and tends to favor per-core performance.

At stock settings, the Core i7-8086K surprisingly trailed Core i7-8700K and -8700, though just barely. Overclocking provides a minuscule boost over the other tuned Intel processors.

Civilization VI Graphics Test

Core i5-8600K dominated at both stock and overclocked settings, which tells us that this benchmark prefers physical cores over logical resources. The overclocked Core i7-8086K fell next in line with a lead over competing Core i7 and Ryzen 7 models. However, it trailed the -8700 by 1 FPS on average at stock settings.

Warhammer 40,000: Dawn of War III

Dawn of War responds well to Intel's high clock rates, so it was no surprise to find overclocked Coffee Lake-based CPUs at the top of our chart.

Although the overclocked Core i7-8086K landed in first place, it's clear that the outcome in Dawn of War III was limited by graphics performance up top.

A stock Core i7-8086K beat the -8700K. But the difference between them was so small that the -8086K's 300 MHz peak Turbo Boost advantage didn't seem to help much.

MORE: Best CPUs

MORE: Intel & AMD Processor Hierarchy

MORE: All CPUs Content

Far Cry Primal, GTA: V & Hitman

Far Cry Primal

Tuning provided small average frame rate boosts to Intel's six-core Coffee Lake-based CPUs. Meanwhile, Core i7-8086K only offered a slight advantage versus the less expensive eighth-gen Core i7s. 

Grand Theft Auto V

Grand Theft Auto V favors Intel architectures and, more generally, multi-core designs with high clock rates.

The stock and overclocked Core i7-8086K yielded a small advantage over the Core i7-8700K.

Hitman

Our Hitman benchmark was rendered almost useless by a patch that imposed a 90 FPS cap on performance. A few weeks ago, though, a subsequent update restored our Hitman test to its prior glory.

With the frame cap removed, Intel's overclocked processors hit a performance ceiling that may be imposed by available graphics horsepower.

Shifting focus to the stock configurations, Intel's six-core CPUs were clearly faster than last generation's quad-core flagship.

MORE: Best CPUs

MORE: Intel & AMD Processor Hierarchy

MORE: All CPUs Content

Shadow Of War & Project CARS 2

Middle-earth: Shadow Of War

Not all games respond to increased host processing resources; some of them are wholly limited by available graphics horsepower. Middle-earth: Shadow of War is definitely one of those graphics-bound titles, demonstrating a 4.5 FPS average variance from the slowest sample in our pool to the fastest. As a result, it was no surprise to see Core i7-8086K and -8700K tied.

Project CARS 2

Project CARS 2 is purportedly optimized for threading. However, our 6C/6T Core i5-8600K beat the overclocked 8C/16T Ryzen 7 2700X, so it's clear that parallelism isn't the most influential factor in defining this game's performance.

MORE: Best CPUs

MORE: Intel & AMD Processor Hierarchy

MORE: All CPUs Content

Office & Productivity

Adobe Creative Cloud

At stock settings, Core i7-8086K offered minimal improvement over the Core i7-8700K in our overall Adobe Creative Cloud score. Even though this suite has a few parallelized workloads, the final score is heavily influenced by the lightly-threaded tasks common in most desktop applications. So, it wasn't surprising to see the Core i7-8086K's superior overclock beat the tuned -8700K by 6.5%.

Web Browser

The Krakken suite tests JavaScript performance using several workloads, including audio, imaging, and cryptography. Core i7-8086K trailed the -8700K in stock form, though overclocking changed the story.

The MotionMark benchmarks, which emphasize graphics (rather than JavaScript), are exceedingly sensitive to CPU clock rates. Yet, Intel's stock Core i7-8086K trailed the -8700K again, reinforcing our opinion that some motherboard firmware versions aren't fully optimized to exploit the 5 GHz single-core Turbo Boost bin.

Productivity

The application start-up metric measures load time snappiness in word processors, GIMP, and Web browsers under warm- and cold-start conditions. Other platform-level considerations affect this test as well, including the storage subsystem. Core i7-8086K exploited its clock rate advantage in stock and overclocked trim to provide snappy performance.

Our video conferencing suite measures performance in single- and multi-user applications that utilize the Windows Media Foundation for playback and encoding. It also performs facial detection to model real-world usage. Core i7-8700 beat the -8086K by a hair; however, its locked multiplier prevents it from vying for chart-topping performance. Core i7-8086K posted a lead at stock frequencies during the writing benchmark, reinforcing that win after overclocking.

The photo editing benchmark measures performance with Futuremark's binaries using the ImageMagick library. Common photo processing workloads also tend to be parallelized, so it's no surprise the tuned Ryzen 7 processors lead by a large margin. The overclocked Core i7-8086K fared best among Intel's CPUs, but there's no substitute for core/thread count in this workload. 

MORE: Best CPUs

MORE: Intel & AMD Processor Hierarchy

MORE: All CPUs Content

Rendering, Encoding & Compression

Rendering

Threaded workloads remain an uncontested strength of AMD's Zen-based processors and their hefty core counts. But tasks that are also affected by memory performance, such as Blender, allow Core i7 to claim a lead. The multi-core Cinebench and POV-Ray tests are dominated by the Ryzen line-up.

At stock settings, Core i7-8086K lead the -8700K in our single-core POV-Ray and Cinebench benchmarks. Overclocking opened up  a much wider gap between the two CPUs.

Encoding & Compression

LAME is a single-threaded workload that typically illustrates the advantage of higher clock rates and IPC throughput. Not surprisingly, then, Core i7-8086K's frequency advantage lead to a win.   

Our threaded compression and decompression tests work directly from system memory, removing storage throughput from the equation. Thus, we found that performance scaled according to core/thread count.

y-cruncher, a single- and multi-threaded program that computes pi using AVX instructions, kept itself isolated to one core during our single-threaded test, allowing the Core i7-8086K to flaunt its higher frequency relative to the -8700K. Conversely, the multi-threaded y-cruncher test reminded us that both processors have the same multi-core Turbo Boost frequencies.

MORE: Best CPUs

MORE: Intel & AMD Processor Hierarchy

MORE: All CPUs Content

Final Analysis

Core i7-8086K's higher base and single-core Turbo Boost frequencies delivered small speed-ups throughout our test suite. But because of Windows' busy nature, those gains were somewhat unpredictable. Although the -8086K rarely stayed in its single-core Turbo Boost bin for long, the same can be said for most CPUs. Regardless, Core i7-8086K earns recognition for becoming the fastest gaming processor on the market, if only just barely.

Our charts below plot gaming performance with a geometric mean of the 99^th percentile frame times (a good indicator of smoothness) converted into a frame-per-second measurement. We also have price-to-performance charts that get split up to include the CPUs-only, plus extra platform costs. For the models that don't come with a bundled cooler, we add $25 for a basic heat sink. We also add $20 if overclocking requires a more expensive motherboard (as is the case for Z370).

If you're only looking to use Core i7-8086K at its stock settings, the processor provides nearly the same gaming experience as the Core i7-8700K at 4.9 GHz. Its advantage is minor at 1920x1080, and it'll shrink at higher resolutions. Granted, overclocking is one of the -8086K's selling points. But the extra 200 MHz you get compared to our -8700K just doesn't justify a $75-higher price tag. And as with all K-series SKUs, you need to buy your own cooler and 300-series motherboard.

We see similar trends throughout our application tests: Core i7-8086K is strikingly similar to the -8700K, and overclocking opens a slight advantage due to our sample's increased headroom. It's only a shame that Intel didn't have the margins to also improve Core i7-8086K's multi-core Turbo Boost frequencies. Such a move would have yielded bigger gains across the board.

You could always purchase a delidded Core i7-8700K, or do the risky work yourself, to match the -8086K's overclocking potential. But if you go the Silicon Lottery route, expect to pay even more than a brand new Core i7-8086K costs and lose two years of warranty coverage.

Core i7-8086K is probably overkill for most of our readers. Both Intel and AMD have far more economical options that provide similar performance through our benchmark suite. Given the limited supply of Core i7-8086Ks, however, we don't expect them to be available for long, and competitive positioning probably isn't the top priority for this CPU's target market.

In light of the anecdotal evidence we've heard from Silicon Lottery, you can rest assured that the -8086K represents Intel's very best Coffee Lake silicon. There are those among us who always seek out the best performance possible, regardless of price. If that describes you, then Core i7-8086K is the fastest gaming chip out there. Just be aware that you're paying dearly for a bit of overclocking headroom.

Moderate gains at stock clock rates mean Core i7-8086K isn't worthwhile for most of Intel's customers. But if you're willing to pay a premium for a piece of history that just so happens to perform well, the -8086K is a cool, enthusiast-oriented option.

MORE: Best CPUs

MORE: Intel & AMD Processor Hierarchy

MORE: All CPUs Content

Children love to obsess over things. My niece knows every scene from "Moana," my nephew regularly quizzed everyone around him on the names of every character in "Thomas the Tank Engine," and I personally used to watch every minute of "Power Rangers" that aired. It comes as little surprise, then, that my childhood obsession with stunt doubles beating on each other in spandex jumpsuits would lead me to Chroma Squad.

Chroma Squad is what happens when you put "Power Rangers," a management simulator, and a turn-based tactical RPG in a blender. The game tasks you with managing a new TV studio devoted to a show that's so similar to "Power Rangers" it almost hurts. The individual episodes play out as turn-based tactical RPGs where you have to simultaneously defeat your opponents and keep the audience captivated by doing specific moves.

I love everything described in that paragraph. Some of my favorite games--Final Fantasy Tactics, Shining Force, and several members of the Fire Emblem series--are tactical RPGs. Having something to do in between fights is also welcome; some of those games can feel tedious because they're basically a series of battles with cutscenes in between. Managing the studio and winning over the imaginary audience keeps things fresh.

It's a good thing Chroma Squad's presentation lives up to its core design. The game's developers clearly understood that part of the reason "Power Rangers" was (and can still be) so appealing is the fact that it embraces the kitsch. Characters don't need to be deep; they can fit into an archetype. Scripts don't need to make sense. Nothing matters as much as having fun, and that's exactly what this game offers.

All of this is wrapped up with pixel art and a soundtrack reminiscent of the early days of modern gaming. Chroma Squad isn't the best looking or sounding game out there, even among the nostalgic throwbacks it unabashedly emulates, but it still carries its own charm. That's exactly what I want when I look for something to play in between work, competitive titles like Overwatch and League of Legends, and other responsibilities.

Because it was released in 2015, you can find Chroma Squad on GOG and Steam for just $15, and it falls as low as $2 when it goes on sale at various other stores. If you're looking for a tactical RPG, or if you simply want to see what managing a "Power Rangers" production would be like if the actors had to beat each other up, Chroma Squad is worth checking out. As I would've told you a few decades ago: It's morphin' time.

In addition to AMD's desktop-oriented Radeon Software Adrenalin Edition, it's also time for the company to introduce its Radeon Pro Software Adrenalin Edition.

Although we initially had trouble getting AMD's press build installed, a second version of the driver made available to reviewers did behave as expected. It included an extra bit of text in the file name: CWG-Enabled-Gaming. Those first three letters stand for CreatorWhoGames, so we weren't surprised to discover that game profiles were automatically available to our Radeon Pro WX 7100, even though it's a pure workstation card. Makes us wonder what advantages the Radeon Pro WX 9100 will offer over Radeon Vega Frontier Edition, particularly when the former costs $600 more than the latter.

The Update Cycle

Back to the driver itself. Enterprise-class means that you count on this software in terms of planning, stability, updates to support a changing operating environment, and optimizations for the professional applications that matter. Certified or not, confidence in the hardware and its underlying infrastructure plays a role in your purchasing decision. AMD is on a quarterly update schedule with fixed dates for publication so that roll-outs can be planned in advance. In fact, we already know the drivers go live on the second Wednesday of the second month each quarter.

Radeon ProRender

ProRender isn't new, but it's one of those features that AMD continues to develop. With its Radeon Pro Software Adrenalin Edition update, this physically-based rendering engine gains support for interactive viewport denoising. Known information is mixed with 3D real-time content to generate a clear render preview. The advantage is obvious: you don't have to wait forever for the render to finish, yet can still identify possible mistakes.

The Game Engine Importer is also improved, making it possible to pull geometry and materials into real-time engines for viewing in VR. The boundaries between building a project, visualizing it, and even interacting with it in virtual reality continue to blur. This is by no means a gimmick. It utilizes the mature functionality and high performance of platforms like the Unreal Engine not for entertainment, but productivity. Suddenly, AMD's idea to enable workstation cards with gaming optimizations makes a lot more sense. After all, not everything coming out of game engines is diversion these days.

Physically-based rendering facilitates the use of realistic materials and lighting models, yielding a better-looking real-time preview than you'd get from static and artificial Phong shading. This PBR shader is now available for Blender, joining 3ds Max and Maya 2018 on the PC. If you're on macOS, you get support via Maya and Blender. In addition, you have to mention Cinema4D, where AMD's Radeon ProRender feature plays an interesting role, enabling content creation with decent acceleration.

ReLive And Overlay

AMD's ReLive feature also makes its way into the workstation space, delivering functionality that was previously only available through third-party software. Similar to what ReLive does on the desktop for games, you're able to save on-screen content as an image or video, with or without sound, then store, edit, or share that content with others. The new overlay, which makes it possible to fade-in your own camera using chroma keying, brings video work into a familiar interface. A lot of folks are going to find this workflow more comfortable than learning an entirely new piece of software.

Virtualization

AMD continues its push in the virtualization space with MxGPU technology, first discussed back in 2016 and enabled with a previous generation of GCN-based hardware. For this release, the company is talking about its open-source KVM host OS driver, available on GitHub. The argument in favor of a solution like this is that it can save costs and provide more transparency than proprietary solutions. We only have to hope its performance can keep up with those competing solutions.

Performance: From 2016 To Today

In an effort to track AMD's performance over time, we pulled out its Radeon Pro WX 7100 and installed the Radeon Pro and AMD FirePro Software Enterprise 16.Q4 driver. We determined the card's performance mid-year with Radeon Pro Software Enterprise 17.Q3 (from July '17), and then tested with the new driver, Radeon Pro Adrenalin 17.5.

We also tested the Vega-based Frontier Edition and realized that nothing really changed between its launch this summer and today. Obviously, that older 16.Q4 driver doesn't support any of the Vega cards. Actual improvements must have happened in the first half of 2017.

The benchmarks show that AMD was able to boost some of its SPECviewperf scores, though other changes between 2016 and 2017 are more modest. Fluctuations of about one percent are normal, and thus fall below our measurement tolerances.

Using the full version of professional applications yields more interesting results. After all, freely-accessible metrics like SPECviewperf receive a disproportionate amount of optimization time. Gains achieved in a test everyone can run are more likely to be reproduced around the Web, which makes for great marketing. And that's why we make it a point to run a more diverse benchmark suite.

Conclusion

Aside from the driver's initial failure to install, we come to the same conclusion here as our U.S. team covering AMD's desktop-oriented Radeon Software Adrenalin Edition release: this update furthers AMD's cause against a relentless competitor. Improved features, new functionality, and a well-defined roadmap make it clear that AMD has its eye on the prize, and understands what an enterprise customer needs from a professional product. Keep up the momentum, AMD. We certainly approve of your work.

MORE: Best Graphics Cards

MORE: Desktop GPU Performance Hierarchy Table

MORE: All Graphics Content

The wait is over: You can now download the Windows 10 Creators Update via the Windows 10 Update Assistant.

Microsoft is set to release the Creators Update to all users on April 11. However, you can install the update now if you want to stay a little ahead of the curve. All you have to do is head to Microsoft's website, click the "Update now" button, and run the Windows 10 Update Assistant after it downloads. It should then check for updates and let you know that Build 15063--the official build number for the Creators Update--is now available to install.

We're about to install the Creators Update ourselves to learn more about what's been added to the operating system, what's changed, and what remains the same, for better or worse. We do know that we have a few things to look forward to: a Game Mode that promises to boost performance, Beam live-streaming from the Game Bar, a picture-in-picture mode for streaming videos while pretending to work, and improved privacy settings.

Other changes include improved performance and security for the Microsoft Edge web browser, a night light mode that reduces blue light emissions, and the ability to lock your PC via Windows Hello on your smartphone. Microsoft also included a new 3D Paint app that seems poised to compete with Blender, the popular freeware tool used for 3D graphics and animation. There's also improved support for Windows Mixed Reality-compatible products.

Intel's Pentium CPUs Get Hyper-Threading

Enthusiasts and casual users alike have suffered from the slow trickle of CPU innovation over the last several years. Each new generation brings smaller improvements, and lately, stagnant pricing. Intel's 14nm Kaby Lake architecture, which marks the company's transition to an extended tick-tock-tock cadence, sets the stage for even less excitement from each generation. We appreciate faster transistors that provide higher clock rates, along with Intel's improved media capabilities, but the rewarmed Skylake design won't inspire anyone with a fairly modern PC to upgrade.

Although the high-end processors give us little to talk about, Intel's recent realignment of the Core i3 and Pentium families are a bit more newsworthy. First, the company launched an unlocked Core i3-7350K, and though it doesn't offer the value we expect from an i3, it is a fun chip for tuners.

Intel also infused the Kaby Lake-based Pentiums with 100-200 MHz of extra frequency. More important, they now enjoy the benefit of Hyper-Threading technology. In the past, Hyper-Threading was a key differentiator between the Core i3 and Pentium CPUs, but the dual-core chip's ability to operate on four threads simultaneously could put today's Pentiums on-par with some of yesterday's low-end Core i3s. Hyper-Threading can boost performance up to 30%, and though we usually see ~20% gains in applications optimized for parallelization (mileage varies, of course), this move also opens the door to games that require four threads.  

Kaby Lake Pentiums still include 3MB of last-level cache shared across the die, which is another differentiating feature compared to the 4MB-equipped Core i3s. Thankfully, the Pentiums do include a heat sink, which will help value-seekers keep costs down.

The 51W Pentium G4620 is the family's highest-end model. Its 3.7 GHz base frequency is only 100 MHz higher than the previous-gen G4520. As with all Pentiums, Turbo Boost is not supported. The chip does feature HD Graphics 630, though, and the Gen 9.5 graphics architecture provides fixed-function hardware for HEVC 10-bit decode/encode, VP9 8/10-bit decode, and VP9 8-bit encode. The G4620 offers promising performance, but its $93 price tag comes uncomfortably close to the Core i3 series. 

The 54W Pentium G4560 appears to offer better value with its 3.5 GHz base clock rate and $64 price tag. That's 31% less money for a 200 MHz sacrifice. The G4560 even challenges low-end Core i3 CPUs. Although it operates at a lower frequency than the 3.9 GHz Kaby Lake i3-7100 and 3.7 GHz Skylake i3-6100, it retails for $53 less. The Pentium G4560 drops you back to HD Graphics 610 with a lower 1050 GHz turbo clock rate, but most Tom's Hardware readers will probably pair the Pentium with a mainstream add-in graphics card. 

Intel did make a few adjustments to prevent the Pentiums from plundering sales of its own more expensive models. The company nixed support for AVX/AVX2 instructions and TSX-NI, though we don't expect those omissions to hurt low-cost gaming machines much. It also trimmed Optane support, which is one of the few reasons to upgrade to a 200-series motherboard. We don't know what price Intel's Optane caching will command when it comes to market later this year, but we're fairly confident that the technology won't be aimed at entry-level machines. Although the H270 and B250 platform controller hubs offer more connectivity than their predecessors due to increased HSIO lane allocations, if you don't need those features, low-cost 100-series motherboards are plenty attractive (and less expensive).

Some games benefit more from high clock rates more than any other specification, and most titles played on a mainstream gaming system will be graphics-bound before a CPU bottleneck rears its ugly head. Either one of the Pentiums we're reviewing complement low-cost motherboards and sub-$200 graphics cards for reasonable 1080p performance. But at the price points we're talking about, we want to really optimize for value. Let's see if the G4620's slightly higher clock rate is worth the big premium.

MORE: Best CPUs

MORE: Intel & AMD Processor Hierarchy

MORE: The History Of Intel CPUs

Gaming Benchmarks

Game Testing Particulars

Enthusiasts who shell out enough money for a Titan X aren't going to have an overclocked dual-core CPU under the hood. In an effort to balance the host processors we're testing with graphics horsepower, we're complementing them with a mid-range Asus Strix RX 470 4GB. We're also benchmarking at the 1920x1080 resolution most mainstream gamers will aim for in this price range. We present the AMD Athlon X4 750K with a 4.3 GHz overclock.

Comparison Processors

Ashes of the Singularity

Our entire test pool falls below Oxide Games' official minimum spec of a Core i7 or equivalent. Ashes of the Singularity is CPU-intensive and scales well with increased core counts and frequency, so the Core i5-7600K naturally assumes its position at the top of the chart. The battle at the other end of the price spectrum favors Intel's Core i3-6320. However, the Pentiums also turn in a decent showing.

The G4620 averages 43.8 FPS, edging out the G4560's 40.3 FPS. The Haswell-based Pentium G3258 does a great job of showing why you want Hyper-Threading and higher clock rates. After all, the Pentium G4620 and G4650 provide ~2x the performance. The difference is even more pronounced in our variance and unevenness test results.

Terrible frame time results from the Pentium G3258 and Athlon X4 750K necessitate removing them from the chart. They muddy the outcome, so we're creating two sets of line graphs when it helps clarify our presentation.

Battlefield 4

Battlefield 4's single-player campaign is notoriously graphics-bound, so we don't observe much scaling as we swap dual-core CPUs out for quad-core models. Only a few FPS separate the Kaby Lake CPUs. The unevenness and FPS/frame time difference results reveal just how close the G4620 and G4560 are when we pair them with a mid-range GPU.

You could drop down from the Ultra-quality preset to facilitate higher frame rates, but both Kaby Lake-based Pentiums can obviously push the RX 470 to its limit with these settings. The Haswell-era Pentium G3258 suffers frequent frame rate dips and worrying frame time variance.

Hitman (2016)

The Athlon X4 750K and Pentium G3258 struggle with Hitman's CPU-centric workload, thus providing unplayable performance.

The Hyper-Threaded Pentium G4620 and G4560 demonstrate incredible gains over the old G3258. Although the average FPS advantage is impressive, it's the higher minimum frame rate that has us so impressed. The G3258 also suffers extreme frame time variance during the test, while the Kaby Lake Pentiums compete with the pricier Core i3s.

The G4620 provides higher average and minimum frame rates than the G4560, but once again the unevenness index shows that there is little differentiation from a smoothness perspective.

Grand Theft Auto V

The quad-core CPUs bunch together during the opening sequence of our benchmark, but separate into a clearly-defined hierarchy as the workload becomes more processor-bound.

The Pentium G4620 pulls ahead of the G4560, but it's hard to say a 3.9% lead in average frame rate is worth the 31% price difference.

Core i3 and i5 CPUs perform best, but the Pentium G4620 and G4560 enable playable performance. That's a big jump over the woeful Pentium G3258 and Athlon X4 750K.

Project CARS

Slightly Mad Studios designed the Project CARS game engine specifically to promote parallelism by breaking tasks down into smaller chunks across available resources. The end result is a sophisticated engine that scales well with additional CPU cores and higher clock speeds. The developer recommends a minimum of a Core 2 Quad Q8400 or AMD Phenom II X4 940, but the G4620 and G4560's additional threads push them into contention with the beefier dual-core chips.

Once again, there is very little difference in the average frame rates achieved by our Kaby Lake Pentiums. But Project CARS does expose one of the most pronounced deltas between the two processors in our unevenness test results.

Although the Core i3 series provides a tangible advantage over the Pentiums, Intel's Core i3-7350K has the least attractive value proposition of the test pool. Yes, you can overclock the -7350K for additional performance, but what you have to spend on a heat sink and higher-end motherboard to leverage the chip's unlocked multiplier is prohibitive.

Metro: Last Light Redux

The Pentium G4620 barely slips past Intel's Core i3-7350K, but this game is obviously graphics-bound with modern processors.

Metro: Last Light Redux lists a 2.2+ GHz dual-core CPU as its minimum requirement, so our entire test pool technical qualifies. But the Athlon X4 750K and Pentium G3258 continue to lag the field due to their aging architectures.

Middle-earth: Shadow of Mordor

Middle-earth: Shadow of Mordor's minimum CPU requirement is either a Core i7-750 or AMD Phenom II X4 965.

The Athlon X4 750K is the only processor that struggles mightily with this game's built-in benchmark. Meanwhile, a scant 0.5 FPS separates the other CPUs we tested. Intel's Pentium G4260 and G4560 again show little variation in the FPS/frame time difference and unevenness categories.

3DMark Time Spy & Fire Strike

We selected the Time Spy metric to illustrate DX12-based CPU scaling due to its demanding physics simulation, occlusion culling, and procedural generation operations. DX12 offers better scaling than DX11 and exploits the quad-core processors to great effect; we recorded a 1952-point gulf between the high-end dual- and quad-core CPUs.

The Pentium G4620 and G4560 leverage their additional logical cores to generate a big lead over the older Pentium G3258. The difference between the G4620 and G4560 is subtler. Intel's Pentium G4620 only provides ~4% more performance during the DX12 test.

The DX11 Fire Strike benchmark runs 32 parallel soft and rigid body physics simulations that tax the processor specifically. In this test, the Pentium G4620 only provides a 6% advantage over the G4560.

Office & Productivity Applications

Blender 2.78b

The multi-threaded Blender test scales well when you back it with multiple physical cores. But it also responds well to higher frequencies and Hyper-Threading.

Notice the pronounced jump in render times when we switch to Pentium processors? Blender uses AVX instructions on compatible CPUs, and the Pentium's lack of AVX support shows through. They're instead forced to utilize the SSE instruction set.

Some rendering and encode/decode programs are optimized for AVX2 extensions. And while most software can fall back to older instructions, performance is typically lost in the process.

Cinebench R15

Maxon's Cinebench single-core benchmark yields predictable results, given what we know about each architecture. The Core i5-7400 uses its top 3.5 GHz Turbo Boost bin to provide a slight advantage over the 3.5 GHz Pentium G4560. Intel's Haswell-based Pentium G3258 employs a lower 3.2 GHz frequency, so it naturally lands below the G4560. The Athlon suffers from an architecturally-imposed IPC deficit and falls to the bottom of our test pool.

The multi-core test scales well. Intel's Pentium G4560 and G4620 enjoy a healthy lead over the G3258 due to their higher frequencies and Hyper-Threading support. But this test prefers the Core i3's more aggressive clock rates and increased cache, along with the Core i5's additional physical cores.

Although Cinebench's OpenGL workload is mostly a test of graphics performance, our Radeon RX 470 behaves differently depending on the host processor it's paired with.

HandBrake

Our HandBrake workload involves converting a 4.19GB movie file into an MP4, so it's a long and repeatable threaded benchmark. Like our Blender results, we notice a big drop in performance as we transition to the AVX-deprived Pentium series. Hyper-Threading helps the benchmark along compared to Intel's older Pentium G3258, but the company's Core i3s definitely enjoy an advantage due to their ISA enhancements.

LAME

Our suite favors future-looking threaded applications, but the nuances of Intel's Turbo Boost technology are most perceptible in lightly-threaded tests. We use LAME to characterize single-threaded performance in a simple .WAV to .MP3 conversion.

Although we're accustomed to seeing the highest available Turbo Boost frequencies in this metric, Intel's Core i5-7500 only hit 3.7 GHz (instead of its 3.8 GHz ceiling), allowing the 3.7 GHz Pentium G4620 to keep pace. We also noticed the Core i5-7600K behaving similarly; it jumped to 4.2 GHz briefly, but fell to 4.1 GHz for the remainder of our test.

7-Zip 16.02

7-Zip utilizes all available execution resources, rewarding processors with more cores and higher clock rates. The Pentium G4560 lags behind its more powerful relative by a mere 4%.

Notably, the Pentium G4560 beats the G3258 by 40%, which is a result of both its faster clock rate and the additional logical cores. 

Adobe After Effects CC

After Effects is one of the few Adobe applications that launches several concurrent threads to spread work across multiple cores. As a result, physical cores are favored, though Hyper-Threading does facilitate greater utilization of the dual-core models. Unsurprisingly, there is a slim margin between the Pentium G4560 and G4620.

Adobe Illustrator CC (64-bit)

A responsive storage subsystem and high clock rates come into play during the Illustrator workload.

Adobe InDesign CC (64-bit)

AMD's Athlon X4 750K stands out due to its extremely low performance, while the Core i5-7400 falls between Intel's Pentium G4620 and G4560.

Adobe Photoshop CC (64-bit)

The Photoshop Light test performs simple tasks like changing the color balance and auto-leveling, while the Heavy sequence employs more stressful manipulations. The CPU-centric benchmarks perform to our expectations, and the results of both tests align similarly.

Advanced Photo Editing & 4K Video Editing

Photo and video editing is a universal task that benefits greatly from OpenCL acceleration. The conventional test relies on host processing, while the accelerated results quantify the benefit of heterogeneous computing with a Radeon RX 470.

There's a one-second difference between the Pentium G4620 and G4560 in our photo manipulation workload, which grows to seven seconds under a more demanding 4K video editing benchmark. OpenCL acceleration has a potentially tremendous impact on performance, in this case whittling the difference between Pentiums down to an imperceptible level.

The Athlon X4 750K is unbearably slow on its own as we edit 4K video. Fold in the power of heterogeneous compute, though, and it becomes competitive.

Microsoft Excel 2016 - Word, Excel, & PowerPoint

Mundane Excel tasks don't require a lot of computational horsepower, so our test pool offers generally acceptable performance across the board.

Power & Thermals

Unfortunately, BCLK-based overclocking is mostly fruitless. Intel encouraged motherboard vendors to disable the feature, even via third-party clock generators, with recent BIOS updates.

We didn't see any thermal bottlenecks with our Corsair H100 v2 cooler and these mainstream CPUs, so there's frankly not much to say about heat. We did record temperatures with AIDA, but a changing ambient environment likely exaggerates any reported delta between our test runs.

The Pentium G4620 averaged just 29W during the stress test, and the G4560 dipped to 24W. In comparison, the Core i5-7600K averaged 46W, and the Core i3-7350K consumed 29W. The low-power Pentiums perform fine with Intel's stock air cooler, helping you save money for other components. 

The number of configurations we tested was so large that our custom graphics charts couldn't accommodate another sample and remain readable, so we dialed in a relatively simple 4.3 GHz overclock on the aging Athlon X4 750K at 1.43V rather than bore you with stock results. The tuned settings push AMD's power consumption to over 100W, which has an understandably detrimental effect on performance per watt comparisons.

Conclusion

In the face of AMD's impending Ryzen launch, Intel seems resigned to bulking up its low-end offerings to stave off the competition's historically competitive mainstream CPUs. No matter the motivation, Hyper-Threaded Pentiums are a ray of sunshine for budget-conscious builders. The technology facilitates a big boost over Intel's Haswell-based Pentium G3258, which came to be derided for its poor performance in games that expected at least four threads. Although we did notice a performance hit in workloads optimized for AVX extensions, you can use an OpenCL-compatible GPU to augment performance in applications written with heterogeneous computing in mind.

Despite Intel's careful segmentation, the more powerful Pentium family unavoidably takes the shine off of lower-end Core i3s, which cost more and don't always deliver the extra performance to match. But both of the Pentiums we looked at today are still slower than Core i3s, i5s, and i7s in threaded tasks like media encoding and file compression. If you truly need the muscle of a quad-core CPU, don't rely on Hyper-Threading to match Intel's beefier processors.

The Pentium G4620 is a compelling offering, and if integrated graphics fit in with your productivity-oriented plans, HD Graphics 630 is as good as it gets from a Pentium. Expect to pay just under $100 for marginal gains over the Pentium G4560, though. In many tasks, particularly games, a GPU bottleneck masks the gains of a faster, more expensive CPU, so you'll need to gauge the G4620's benefit based on what you're doing and what other components might limit performance.

We picked up the Pentium G4560 for a scant $64 with free shipping. While it's true that we measured lower performance from the G4560 in several games, the nimble chip always landed a few percentage points away in average and minimum frame rate results. Hyper-Threading technology and higher clock rates definitely yield big gains compared to the unlocked Pentium G3258, at times doubling the unlocked CPU's performance. As important, it provided the same smooth frame delivery as Intel's Pentium G4620.

There's little reason to pair a Pentium with a Z270-based motherboard, so you can save some money on something with an H270 or B250 PCH. All told, it should be possible to build a capable platform for a few hundred bucks, leaving more room in your budget for a faster graphics card.

AMD historically does well in the value space, and many believe it'll claw back lost market share with Ryzen. We do expect AMD to be aggressive with its pricing, and the company's chipsets will once again be modernized, so that's a good thing. But we can't deny the Pentium family praise based on what AMD might do. By adding Hyper-Threading to the Pentium series, Intel boosts performance in a way that might even cut into some of its Core i3 sales. Sounds like a win to us.

MORE: Best CPUs

MORE: Intel & AMD Processor Hierarchy

MORE: The History Of Intel CPUs

Introduction

It seems like only yesterday that we were introducing you to Intel’s Core i7-5775C and i5-5675C, the first socketed CPUs based on the company’s Broadwell design. In reality, that was about 60 days ago. So why on earth are we talking about Skylake, another new architecture, two months later? Aren’t these big reveals supposed to happen once every couple of years?

It’s no secret that Broadwell encountered delays. We should have been looking at products built on the 14nm die shrink a long time ago. But desktop enthusiasts were among the last in line to get their hands on Intel’s efforts. Perhaps in anticipation of today’s launch, Broadwell didn’t receive much fanfare during Computex 2015. Those C-series parts aren’t even readily available online today. Nevertheless, we marveled at the performance of their graphics engines, sporting 48 EUs and 128MB of embedded DRAM. It was great to also see them arrive multiplier-unlocked, ripe for the affections of power users building highly-integrated HTPCs. The only bummer was that a dialed-back TDP and lower stock clock rates meant Core i7-5775C was slower through our benchmark suite than the previous generation’s Core i7-4790K.

Today we’re pulling the wraps off of the true Devil’s Canyon successors, Core i7-6700K and Core i5-6600K. Our exploration is going to feel a little incomplete though, because Intel isn’t going into detail on the Skylake architecture until IDF later this month. What we have, then, are a couple of CPUs, motherboards based on the Z170 chipset, some killer DDR4 memory kits and a bunch of questions about what makes Skylake special.

It’s almost like someone at Intel decided, “Hey, enthusiasts are going to be so excited to see sixth-gen Core processors at gamescom that they won’t care what’s inside of them.” And you know what? I’m tired of trying to figure out how decisions like these are made. Assume the "better together with Windows 10" angle had something to do with it. So let’s just roll with the information we do have. It’s so much less frustrating that way.

Intel Core i7-6700K And Core i5-6600K CPUs

Tools like CPU-Z with preliminary Skylake support still identify Core i7-6700K as a 95W part. In fact, all of Intel’s early documentation carves out a 95W category. But Intel now says the -6700 and Core i5-6600K have 91W TDPs. They fit into LGA 1151 interfaces keyed differently than its LGA 1150 CPUs, so there’s no way to drop one into an incompatible motherboard. Not that this should be a problem. Because Skylake is architecturally dissimilar from prior Core designs, it requires new core logic. In other words, Core i7-6700K and i5-6600K necessitate platform upgrades, too. Fortunately, Intel isn’t slapping a big premium on either processor. Core i7-6700K should be priced around $350, while Core i5-6600K is expected to sell for $243.

Both CPUs feature four IA cores—we know that much for sure. Like so many line-ups before, the i7 comes equipped with Hyper-Threading. It leverages the power of simultaneous multi-threading to more thoroughly utilize available resources. To Windows, each of the Core i7’s physical cores appears as two logical processors. The Core i5 doesn’t have this feature; its four cores work on four threads.

We’ve also seen these cache configurations before. The higher-end i7-6700K includes 8MB of last-level cache, while the i5-6600K sports 6MB. If CPU-Z is to be believed, Intel’s Skylake architecture continues to arm each core with 256KB of L2 cache, along with 32KB of L1 data and 32KB of L1 instruction cache. Neither the i7-6700K nor the i5-6600K include embedded DRAM, which we saw on the Core i7-5775C and i5-5675C.

The Core i7-6700K’s base clock rate is a nice, even 4GHz. Its Turbo Boost profile is quite conservative, though. With just a single core active, the -6700K only pushes up to 4.2GHz. Meanwhile, Core i5-6600K starts at 3.5GHz and stretches as high as 3.9GHz. We’ll go into more depth on overclocking shortly, but of the samples we have in the lab currently, -6700K seems comfortable up to around 4.7GHz. We also spent time at 4.9 and then 4.8GHz, but couldn’t get either stable enough under prolonged stress tests. Five gigahertz wouldn’t boot completely.

Intel’s Haswell-E-based Core i7s already incorporate quad-channel DDR4 memory subsystems. However, Skylake promises to make the technology more mainstream with its own dual-channel controller. In actuality, the architecture officially supports DDR4 at data rates as high as 2133 MT/s and DDR3L at up to 1600 MT/s. All of the boards in our lab currently are DDR4-only though, so anticipate needing a new motherboard and memory kit if you step up to one of the two new enthusiast SKUs.

Most enthusiasts probably won't care, but Intel’s Skylake-based K-series SKUs come armed with the graphics configuration referred to as GT2. The company seems to be saving all of its graphics fanfare for IDF, which is understandable when that’s not a meaningful part of today's conversation. What this means, though, is that the impressive gains we saw from Broadwell’s GT3e arrangement are gone. Those 48 EUs become 24 in Skylake’s middle-of-the-road implementation, branded as HD Graphics 530, and they operate at up to 1150MHz.

Notably, the Gen9 engine implemented in Skylake does support DirectX 12, OpenCL 2.0, OpenGL 4.4 and OpenGL ES compared to Gen8’s DirectX 11.1 and OpenCL 1.2. When the engineers do start talking about graphics at IDF, expect them to spend a lot of time on architectural optimizations targeting power. Of course, performance is tuned as well, primarily through better utilization of available bandwidth (hierarchical Z and memory color stream compression). Through its Y-, U-, H- and S-line platforms, Gen9 should touch everything from 4W to these 91W parts. There’s much more to discuss on the display side (GT4e should be amazing), but we’ll save what we know for another day.

Both Skylake-based processors feature 16 lanes of PCIe 3.0 connectivity for add-in cards, just like their predecessors, configurable as 1x16, 2x8 or 1x8/2x4 links. Where they differ, however, is the four lanes that Intel sets aside for attaching a Platform Controller Hub. In generations past, the company called this interface DMI 2.0. It offered up to 2GB/s of bi-directional throughput—enough to accommodate most combinations of mainstream storage, peripherals and networking. But the 100-series chipsets are more feature-rich, necessitating greater bandwidth to the host processor. As such, Intel gives Core i7-6700K and Core i5-6600K a Direct Media Interface 3.0 theoretically capable of moving almost 4GB/s.

Intel Z170 Chipset

What, exactly, is capable of utilizing that plus-sized pipe? Arguably more exciting than the new Core CPUs themselves are Intel’s 100-series chipsets. The company hasn’t formally announced every model yet. Rather, it’s only talking about the flagship Z170—and that’s fine with us, since, from an enthusiast standpoint, everything else is a subset of the Z170’s feature set.

Certain familiar features naturally carry over from the 9-series PCH family. There’s an integrated gigabit Ethernet MAC with a single-lane connection for attaching a PHY. And you still get six SATA 6Gb/s ports for storage, along with HD Audio support.

There’s a lot more that changes, though. USB connectivity evolves to enable as many as 10 USB 3.0 ports and 14 USB 2.0 ports. Z97, for example, offered six USB 3.0 and 14 USB 2.0 ports. That’s nice—but nowhere near as notable as the 20 lanes of PCI Express 3.0 originating from the PCH. Together with software optimizations in Intel’s Rapid Storage Technology driver, we end up with the first desktop platform designed with PCIe-based storage in mind.

Now, you don't get 20 lanes of PCIe, six SATA ports, M.2 connectors, all of the USB, and GbE at the same time. Z170 is highly configurable, so motherboard manufacturers have to pick and choose how they utilize its "ports". Six of 26 are used up by USB 3.0 (which is where "up to 20 PCIe lanes" comes from). With the remainder, your vendor of choice can tap into another four USB 3.0 ports, SATA, and PCIe-based storage. But they all eat into Z170's connectivity. Populating all six SATA 6Gb/s ports, for example, eats into six lanes. A PCIe storage device monopolizes four. On the Z170A Gaming M7 motherboard we’re using to test, MSI exploits the flexible PCH with four PCIe x1 slots and one PCIe x4 slot, an ASMedia ASM1142 USB 3.1 controller that consumes two lanes and two M.2 slots that appear to share a four-lane link. MSI juggles the various upgrade paths with PCIe switches.

Intel's RST driver continues to support RAID 0, 1, 5 and 10 across SATA devices, but then adds RAID 0 and 1 functionality for PCIe-based SSDs in the M.2 slots, according to MSI’s manual. You can use one of those PCH-derived slots for a graphics card, and the bandwidth increase over DMI 3.0 would undoubtedly improve your experience over previous-gen platforms. MSI says its board will do three-way CrossFire this way. Conversely, Nvidia continues to limit CPUs with 16 lanes of third-gen PCIe to two-way SLI.

Beyond those capabilities easily covered by a block diagram, Z170 is the only chipset with official support for overclocking (that’s full support on K-series parts and partial overclocking on certain non-K parts). Intel seems to be taking tuning much more seriously with Skylake, too. Many of the limitations that worked their way into previous-generation architectures are now gone, giving you the freedom to explore your processor’s potential, even if there’s no guarantee you’ll be breaking clock rate records with the new capabilities.

Overclocking

The overclocking community isn’t always happy with the decisions that Intel makes in pursuit of greater integration, improved efficiency or even just lower cost. We’ve seen lower-quality thermal interface material, restrictive ratios and integrated voltage regulation affect tuning in different ways. Overall, though, the company seems receptive to feedback from the enthusiasts pushing its products beyond their factory specifications.

Sometimes the alterations are superficial in nature. Devil’s Canyon saw Intel make adjustments to the Haswell architecture’s power delivery and thermal performance. But we were still stuck with BCLK ratios, around which the base clock could only be manipulated plus or minus a few megahertz. A new architecture like Skylake gives Intel the opportunity to reevaluate more fundamental subsystems and change their behavior. With that in mind, the Core i7-6700K and Core i5-6600K represent a significant step forward in many ways, even if certain details remain unavailable (for example, what’s Intel using between its die and heat spreader this time?).  

To begin, the Platform Controller Hub’s reference clock to the PCIe bus and I/O is fixed at 100MHz. A separate BCLK signal from the PCH facilitates tuning the processor’s cores, cache, graphics subsystem, memory controller and system agent in 1MHz increments up to 200MHz.

That signal is multiplied against a number of different ratios to dial in optimized clock rates. The cores, for instance, support ratios of up to 83x—higher than Haswell’s 80x ceiling, but far beyond the CPU’s practical limit in both cases. As with Haswell, Skylake’s ring bus ratios are not locked to the cores. Adjustments are available up to 83x as well, though you can detune the ring if you suspect it of holding back a more aggressive overclock elsewhere. The graphics engine offers ratios up to 60x, similar to Haswell and Ivy Bridge. And if you have a mobile Skylake CPU with embedded DRAM, adjustable ratios facilitate a first opportunity to improve its performance through overclocking.

Intel’s early overclocking documentation suggested that memory ratios would be available up to 24x (at 133MHz) and 31x (at 100MHz), creating a maximum of around 3200 MT/s. However, the company's launch material mentions data rates of up to 4133 MT/s. We have kits in-house capable of 3600 MT/s, so there’s still room to scale up as motherboard vendors continue optimizing. Whereas previous architectures exposed ratios that stepped memory data rates up and down in 200/266 MT/s increments, Skylake is more granular with 100/133 MHz steps. The addition of XMP 2.0 simply accounts for a new DDR4 specification. Intel’s XMP certification process remains unchanged, though.

Gone is the fully-integrated voltage regulator, which many enthusiasts faulted for making their Haswell-based CPUs run hotter and require higher-end cooling. It remains to be seen how much impact this has on real-world results, particularly since we’re dealing with a completely different architecture.

Our Hands-On Experience

It’s early in Skylake’s life, and we’re still testing pre-production samples. We weren’t particularly optimistic about the architecture’s scalability given Intel’s 4GHz base clock rate and 4.2GHz peak Turbo Boost frequency on the Core i7-6700K. However, our experience so far suggests that 4.7GHz could be a reasonable target with minimal voltage increase.

One of our samples even cruised along at 4.9GHz using a 1.41V setting, but it wasn’t stable under load, and we’re not comfortable with that voltage.

We didn’t go to the trouble of trying to find our i7-6700K’s breaking point using single-megahertz BCLK adjustments. But just to show Intel’s more flexible BCLK controls do work, we set the reference clock to 115MHz and snapped the following screen shot:

We also devoted more time to testing DDR4 memory scaling. Corsair and G.Skill both sent in kits capable of 3200 MT/s. Additionally, Corsair followed up with a 3600 MT/s configuration. Using MSI’s Z170A Gaming M7 motherboard, we tested at 2133, 2400, 2666, 2933 and 3200 MT/s. Both companies’ kits were rock solid using MSI’s default 3200 MT/s settings and their XMP profiles. But we weren’t able to get 3600 MT/s dialed in; Windows consistently crashed as it started up.

In the hours before launch, MSI did send over a firmware update that got 3466 MT/s stable, albeit at lower bandwidth than 3200 MT/s. It's only a matter of time, though, until even more aggressive memory settings become viable. Intel's early overclocking guidance suggested that ratios enabling 3200 MT/s would be possible and now the company is throwing around numbers like 4133 MT/s. Motherboard vendors are quickly optimizing for the enthusiast-oriented memory kits popping up in anticipation of Skylake, so expect more on this front soon.

Bandwidth continues to scale with data rate. Starting at 2133 MT/s, a >23 GB/s result certainly isn’t bad for a dual-channel controller. By the time you get to 3200 MT/s, though, you’re up above 32 GB/s—an almost-40% increase.

Of course, real-world performance doesn’t improve nearly as much. Intel’s desktop configurations typically aren’t starved for data given the workloads we hit them with, so a best-case speed-up in WinRAR is more like 6-7%. And that’s a test we know to be at least somewhat sensitive to memory performance. Pretty much everything else we run is less responsive.

Interestingly, it’s not at the highest data rate where performance is best, either. DDR4-2933 appears to be the peak, after which bandwidth keeps going up while our timed benchmarks slide ever so slightly.

How We Tested

A switch to LGA 1151 meant we had to start our search for a reference platform all over again. We're already utilizing a full complement of MSI motherboards, so we asked the company to show us what it had planned for Z170...and were impressed. The Z170A Gaming M7 is a beastly piece of hardware, both on its specification sheet and aesthetically.

To be sure, MSI takes full advantage of the platform it's working with. Sixteen lanes of PCIe from the host processor are configurable in x16 or x8/x8 links. The PCH's flexible I/O is set up to accommodate as many as two M.2 slots, two SATA Express connectors (or four SATA 6Gb/s ports), two additional 6Gb/s connectors, six USB 3.0 ports, an additional two 10Gb USB 3.1 ports attached to a two-lane PCIe link, and a host of x1 and x4 expansion slots.

As mentioned, Corsair and G.Skill both sent in memory kits designed to push the capabilities of Skylake's memory controller. It's truly notable that both vendors had no trouble hitting 3200 MT/s (that's already 1066 MT/s beyond the official 2133 MT/s specification) before Intel's architecture even launched. Moreover, we got Corsair's 3600 MT/s kit to 3466 MT/s before it tripped up. Expect companies like MSI to shore up compatibility as quickly as possible.

Most of the benchmarks from today's story were run by Tom's Hardware Germany, using the same hardware that goes into generating our 2015 CPU Charts. In addition to desktop performance analysis, we also go into depth on power consumption, thermals and alacrity in workstation-oriented applications.

Results: Desktop Publishing And Multimedia

Adobe CC

We’re using Photoshop, InDesign and Illustrator, all of which are included in Adobe’s CC package, as well as PCMark 8 Professional to control the workloads. The details of each benchmark are available in the table below.

The storage subsystem and background processes influence the benchmark results, since they include opening and closing each application, as well as loading and saving files. For this reason, PCMark 8 natively reports back the geometric mean of three benchmark trials (GEOMEAN).

In stark contrast to our other benchmarks, we noticed significant performance differences between Windows 8.1 and 10. Consequently, wherever these differences were significantly above the margin of measurement error, we included both results in our graphs. And of course, we're using the most up-to-date drivers available for each platform.

Adobe Photoshop Light

Adobe Photoshop Heavy

Adobe InDesign

Adobe Illustrator

Intel’s Skylake CPUs show much-improved performance under Windows 10 in some applications compared to both their competitors and their own Windows 8.1 performance. We’ll certainly look into this more in the future, since some of the gains are massive. Why AMD’s A10-7850K does so well with Illustrator also remains a mystery to us. We repeated the benchmark with the same results.

Results: Office Productivity

Microsoft Office 2013

No collection of desktop benchmarks is complete without Microsoft’s popular Office suite. We’re leaving control over the workloads (as well as computing and reporting the geometric mean of the three benchmark runs per application) to PCMark 8 Professional once again.

Microsoft Word 2013

Microsoft Excel 2013

Microsoft PowerPoint 2013

As expected, Intel’s two Skylake-based CPUs lead the field. It’s interesting to see that the Core i5-6600K is at least as fast as, or even a bit faster than the i7-6700K, in spite of its lower clock rate. SMT seems to reduce overall performance a bit in Microsoft Office.

Results: Rendering, Encoding, Compression, Arithmetic

Blender (Rendering)

Blender has an efficient rendering module that runs exclusively on the CPU, even though rendering our benchmark file using GPU acceleration is much faster. A tile size of 16 pixels has proven to be the most efficient for host processors, so that's what we’re using. Both Skylake-based processors fall in line as expected.

Cinebench R15 (Rendering)

This benchmark, which is based on Maxon’s Cinema 4D, provides the interesting option to have the CPU render in single- or multi-threaded mode. The ratio between the two says a lot about efficiency, illustrating the difference between physical cores and simultaneous multi-threading implementations.

This relationship between single- and multi-threaded performance is of particular interest. And we come up with one very interesting result: Intel’s Core i5-6600K falls right between the i7-4790K and i7-4770K.

Adobe CC Media Encoder (Encoding)

We’re testing with a video file saved as H.264 with 3840x2160 resolution, 25 FPS, progressive VBR, one pass, 6000 Mb/s target and 8000 Mb/s maximum. It has a 320 Kb/s and 48kHz AAC stereo soundtrack. We’re using the integrated software renderer, which makes optimal use of all possible threads. The results aren’t really surprising, though it's interesting that the Core i7-6700K approaches Intel's six-core -5820K.

WinZip Pro 19 (Compression)

The trick with this benchmark is to compress different types of content, such as text, pictures, multimedia files, videos and applications, without producing troublesome overhead due to time-sensitive file operations. This is why we copy all 3.02GB worth of data to an ISO file that can be compressed in one go. We’re using the CPU, purposely circumventing GPU acceleration via OpenCL.

SiSoft Sandra 2015 (Arithmetic)

If you compare the overall results for single- and multi-threaded performance, putting heavy emphasis on integer and 32/64-bit floating-point throughput, then Intel’s new Core i7-6700K makes quick work of the older i7-4790K, whereas the i5-6600K is found just below the i5-4690K.

If you're only looking at single-threaded performance, the SMT-equipped processors fare significantly worse, since their eight threads are scheduled through four physical cores. Even though Broadwell wins the day and the older Core i5-4690K edges out the i5-6600K, at least the i7-6700K dominates its predecessor.

Results: Workstation Applications

The following benchmarks are based on AutoCAD 2015, Cadalyst 2015 and three modules of SPECviewperf 2015. They were specifically chosen to represent CPU performance, even though we have a GeForce GTX 980 installed instead of a workstation-class card.

AutoCAD 2015 2D And 3D Performance

AutoCAD is a popular application from Autodesk. First, we’re testing its "2D" performance with Cadalyst 2015. Quotes are used there because AutoCAD deals with 2D the same way many other applications do: through DirectX's D3D interface. This way of implementing 2D is worth testing since there really hasn't been any hardware acceleration for 2D through the kernel-mode driver since Windows Vista. Graphics cards with unified shader architectures don’t have dedicated 2D units anymore, either.

The finishing order in this benchmark is determined solely by the CPU, since the graphics card's performance goes unchanged. The two new Intel processors set a new bar in both metrics with similar 2D and 3D results.

Maya 2013

Viewport 2.0 isn’t part of our Maya 2013 results on purpose, since it’s based on DirectX. This means that this benchmark is based exclusively on OpenGL. The render modes used for this benchmark are shaded, ambient occlusion, multi-sample anti-aliasing and transparency, and the model consists of 727,500 vertices.

Showcase 2013

Showcase 2013 is another DirectX-based benchmark. Autodesk might be the only major developer to make the jump to DirectX, but many smaller companies are taking the plunge as well.

The model used for this benchmark includes eight million vertices, as well as render modes like shading, projected shadows and self-shadowing. In the end, clock rate and IPC determine the winners and losers.

SolidWorks 2013 SP1

SolidWorks 2013 by Dassault Systèmes is a classic. The different models for our workloads range in size from 2.1 to 21 million vertices. Individual tests use the software's many rendering modes, including a shaded mode, shaded with edges, ambient occlusion, shaders and environment maps. Compared to SPECapc for SolidWorks 2013, the CPU test is gone, the number of models is lower and a benchmark with parallax effects was added.

HD Graphics 530: Gaming

Bioshock Infinite @ 1920x1080 (DirectX 11)

Bioshock Infinite is no heavyweight when it comes to graphics load. Still, even after we picked the game's low settings, the integrated graphics engine (and not the IA cores), limited performance.

We are disappointed to see Intel take a big step backward here. After enabling stellar frame rates from the 65W C-series Broadwell processors, the company neuters its enthusiast-oriented 95W Skylake CPUs with HD Graphics 530, based on the GT2 configuration. Consequently, the results we measure are a lot worse than what we saw a couple of months ago:

In spite of this, and partly due to its much faster x86 cores, Intel’s new offerings keep pace with AMD’s best APUs, if just barely. Then again, AMD enjoys a substantial price advantage. To be sure, you won't want Core i7-6700K or Core i5-6600K for their 3D capabilities.

Half Life 2: Lost Coast @ 1920x1080 (DirectX 9)

Half Life may be old, but it does represent a challenge for most IGPs, giving us the opportunity to evaluate playable games on entry-level graphics hardware. We’re using 2x MSAA, so the host processing complex isn’t stressed too much.

There’s a marked increase in performance compared to the Core i7-4790K’s HD Graphics 4600. The new processors also narrow the performance gap between Iris Pro 6200 seen in our Bioshock Infinite benchmark.

Grand Theft Auto V — Welcome To The Entry Level

This chart compares a system with an inexpensive CPU and entry-level or older graphics cards to AMD’s current APUs and Intel’s two Skylake CPUs running integrated graphics.

Broadwell’s Iris Pro 6200 had us flying high, and the two Skylake CPUs have us crashing back to the ground. The reason that AMD’s APUs are still within reach is that their x86 cores are much weaker, get overwhelmed by GTA V and severely bottleneck the integrated graphics.

Intel's decision to step back (and down) on graphics compared to Broadwell certainly leaves a noticeable mark on the results. Then again, most enthusiasts aren't going to buy one of these unlocked processors and use its graphics engine for anything other than its Quick Sync functionality. Instead, they'll drop in a more powerful discrete card.

Although we certainly enjoyed seeing what the Core i7-5770C could do without the help of an add-in board, most of our readers correctly countered that dedicating a lot of transistors to integrated graphics would be a waste in a gaming PC.

HD Graphics 530: Workstation

AutoCAD 2015 2D And 3D Performance

We’ve already talked about the way we test AutoCAD. Suffice it to say that the CPU needs to help out quite a bit when it comes to 2D graphics, since this type of acceleration hasn’t been possible through the GPU since Windows Vista. Neither the driver model nor the unified shader architecture provides this functionality.

Of course, that means this benchmark is more dependent on host processing than the graphics card. And sure enough, our benchmarks heavily favor threading on CPUs with a lot of cores.

If the rendering is actually done in 3D, then the order changes dramatically. Broadwell, with its beefy graphics architecture, springs into the lead. Still, Skylake keeps up surprisingly well, given the anemic GT2 configuration. AMD’s APUs just can’t compete as a consequence of their less effective host processing cores.

Maya 2013 (OpenGL)

SPECviewperf uses OpenGL exclusively for its Maya component, manipulating a model made up of 727,500 vertices with shaded, ambient occlusion, multi-sample anti-aliasing and transparency effects.

Graphics processing limits the test's performance, since there isn't much of a CPU load applied. Intel’s Core i7-5770C with Iris Pro 6200 provides up to 36 percent more performance than AMD’s Radeon R7 on the A10-7560K. That'd be a more painful loss of the two chips were closer in price. What hurts more, though, is that Skylake’s HD Graphics 530 are even slower. Anyone looking to use Intel processors in an office and design environment without discrete graphics would probably want to stick with Broadwell or consider other options.

Showcase 2013 (DirectX)

The next benchmark, which we've seen once already, is based on DirectX. Again, the test file for Showcase 2013 uses eight million vertices and, among other capabilities, shading, projected shadows and self-shadowing.

The results make it abundantly clear that integrated graphics make for a lackluster experience. The two Skylake CPUs fit right into this trend with pitiful performance that’s also way worse than Broadwell’s. Then again, all of these contenders are really just theoretical in nature; none of them provide anywhere close to usable performance.

Cinebench R15 (OpenGL)

Finally, Cinebench R15’s OpenGL graphics benchmark puts a bit more emphasis on the CPU.

A quick look at our CPU Charts shows what's possible with a discrete card installed. If, however, you stick with integrated graphics, the on-die engine becomes a distinct bottleneck.

Skylake falls in line between Intel’s Broadwell-based processors and AMD’s APUs once again.

Power And Temperature

In response to feedback from our Broadwell coverage, we're increasing the benchmark run time from one to 10 minutes, and setting the graph's time interval to 500ms, which seems sufficient to achieve meaningful averages.

Idle Power Consumption

The two new CPUs consume even less power than their Broadwell counterparts. With this small decrease, they end up under 4W with their IGPs active and trucking along at 800MHz. This is a great result. In fact, it’s the best we've seen from any CPU.

Gaming Power Consumption

Let’s think back to our GTA V test for a moment. In order to generate measurements that are easy to reproduce, we rendered a scene using the integrated recorder that takes a lot of graphics power at a resolution of 720p for the IGP and 1080p for Nvidia’s GeForce GTX 980.

The combination of GeForce GTX 980 and Core i5-6600K yields an average power consumption of 53W. This is comparable to what we measured for the i7-5775C a couple of months ago. Surprisingly, Intel’s Core i7-6700K comes in 23W above this number, landing at 76W. Our i5-6600K test sample might be great, or our i7-6700K particularly bad, depending on how you want to look at it. At the very least, it demonstrates the variability between CPUs in the same family, though we also have to keep in mind that these are engineering samples.

Now we pop out the add-in card and look at power consumption with on-die graphics. Broadwell’s power consumption stays the same in this test, since a host processing bottleneck keeps it there. That's not the case for Skylake. The two new CPUs’ graphics performance are very similar, with a difference of only 2 FPS. The i7-6700K averages 80 FPS, while the i5-6600K almost manages to keep up at 78. However, the i7-6700K’s power consumption is 83W, whereas the i5-6600K consumes just 69W. That's a significant 14W difference.

Maximum Power Consumption (Stress Test)

We’re creating the maximum possible CPU load, including its x86 cores, the FPU, the cache and the IGP, and then putting power consumption under a microscope. To better illustrate the outcome, each processor gets its own graph, offering a clearer view of the differences between individual performance and load states.

The Core i5-6600K remains relatively frugal with an average of 73W, whereas the i7-6700K helps itself to a full 100W (a 27W delta). Even though the i7 is Hyper-Threaded, allowing it to be better-utilized, with an average clock rate that's 300MHz higher, our measurement just isn't really acceptable, particularly since it's 5W over the CPU's claimed TDP.

Power Consumption Overview

The picture that emerged from our detailed results is confirmed when the two new processors are compared to other CPUs and APUs.

Unfortunate outliers, such as AMD’s FX-9590 under load and the FX-4350 at idle are pieces of the puzzle. Anyone wondering why a Core i7-5960X looks almost like a low-power CPU in comparison needs to remember that most of Intel's processors are operated right around their sweet spots, whereas AMD’s larger FX models and a few of Intel’s chips are pushed into inefficient territory at their stock frequencies.

We’ve deliberately selected a large number of CPUs for this comparison, covering a wide range of performance and, as a result, power consumption.

The new Skylake processors’ low power consumption at idle immediately jumps out in a positive way. These results represent a substantial improvement.

Intel’s Core i5-6600K scores some major points for the new platform when it comes to gaming with an add-in graphics card, even though Broadwell already set the bar for this very high. Part of the explanation for this result can be found in the fact that integrated graphics draws practically no power in this scenario, whereas the older processors with integrated graphics always had at least some losses. Only the Core i7-6700K sticks out a bit; we’ve already discussed some of the reasons that might be the case.

Under full load, we see a similar picture. A difference of approximately 21W compared to the older Core i7-4790K is too high in our opinion. We’ll try to rerun these numbers after the launch with a retail CPU to see if our results are due to our lab sample and update this article accordingly if it proves necessary.

Let’s compare the two CPUs in a table:

Temperatures at Full Load (During the Stress Test)

The Core i5-6600K’s temperature across all cores averages approximately 48 degrees Celsius after 30 minutes, and the i7-6700K ends up at approximately 64 degrees Celsius. Even during peaks, it never surpasses the limit of 73 degrees Celsius.

Due to reader feedback, we reverted back to our reference air CPU cooler, as opposed to the all-in-one water cooling solution. In light of this, these results are certainly decent, even though the i7-6700K’s higher power consumption drives up the temperatures.

The bottom line is that both CPUs can be cooled on air very effectively, which wasn’t really possible with Intel’s slower Core i7-4770K, if it was even just slightly overclocked.

Conclusion

It’s difficult to draw conclusions about hardware still shrouded in mystery. Although there’s plenty we know about Skylake based on Intel’s official disclosures, information leaked elsewhere and the diagnostic tools we use in the lab, this launch just doesn’t feel complete without the background we’d expect to accompany a new architecture. Fortunately, we have the fundamentals: we know how Core i7-6700K and Core i5-6600K perform, we know how much they cost, we have a good sense of their strengths and we’ve seen the platforms they’ll drop into.

Arguably, motherboards are the most exciting part of the story (when was the last time you heard that?). The changes introduced by Intel’s Z170 chipset pave the way for an era of PCI Express-based storage, blasting through the throughput and latency limitations imposed by SATA 6Gb/s and AHCI. Of course, it’s nice to see greater proliferation of USB 3.1 controllers, DDR4 support and granular BCLK overclocking as well.

The Core i7-6700K has a lower Turbo Boost ceiling than Core i7-4790K. But its superior IPC translates to better performance in most single- and multi-threaded workloads. Under the heaviest tasks it typically only trails six- and eight-core Haswell-E parts. Of course, the benefit of Intel’s mainstream platforms is their more modern feature sets. So do you want all of the features I just listed off, or do you need lots of processor-based PCIe and more cores?

Intel will go into more depth on Skylake during IDF later this month, and I don’t see any reason to rush into a purchase before then (provided processors surface for sale right off the bat, that is). Should everything we learn support the data we generated today, then I think it’s safe to say Skylake will become the first architecture to really get enthusiasts excited since Sandy Bridge—and not even entirely because of the processors themselves.

MORE: Best Gaming CPUs For The Money
MORE: All CPU Content

Igor Wallossek is a Senior Contributing Editor for Tom's Hardware Germany, covering CPUs and Graphics.

Chris Angelini is a Technical Editor at Tom's Hardware. Follow him on Twitter and Google+.

Follow Tom's Hardware on Twitter, Facebook and Google+.

Introduction

The time has finally come for Intel to introduce its socketed Broadwell processors. They’re fabulously late, and you’re probably not going to want to buy them, what with Intel’s Skylake architecture close at hand. But they’re frankly fantastic little CPUs, particularly if you’re a history buff.

What do I mean by that? Well, exactly two years ago, the company was unveiling its Haswell-based desktop processors and Tom’s Hardware’s sentiment was captured somewhat succinctly, I think, by the title of our launch story, The Core i7-4770K Review: Haswell Is Faster; Desktop Enthusiasts Yawn.

Prior to that piece’s publication, I had just spent a couple of days with Intel in Santa Clara learning about its Haswell architecture. Most of the emphasis was (understandably) focused on the mobile effort. We talked about optimizations for power, the Iris Pro Graphics 5200 that’d propel high-end notebook performance forward without a discrete GPU and increased attention on x86 architectures that’d scale down to tablet form factors. It was all well and good. What we saw was exciting, and what we were promised pointed to a future filled with performance-enhancing integration driven by efficiency.

There was just one thing missing, though. Desktop users didn’t get the most exciting goodies. The highest-end socketed CPUs were stuck with decidedly mainstream HD Graphics 4600. Worse, perhaps, the company eliminated the “limited overclocking” previously available on non-K-series models.

What reason did we have to upgrade, then? A little more IPC throughput? Meh. We were underwhelmed, and not particularly subtle about letting Intel know.

Broadwell For Desktop: A Change In The Air?

Massive though the company may be, it isn’t deaf to feedback. Representatives readily admitted it’d be both difficult and time-consuming to incorporate the requests we were making into its stack during the Haswell generation. But it actually did. We wanted a more enthusiast-oriented flagship and it came up with Devil’s Canyon (Core i7-4790K Review: Devil's Canyon Tantalizes Enthusiasts). We asked for a multiplier-unlocked model able to compete against AMD’s affordable Athlon X4s and received the Pentium G3258 (Intel Pentium G3258 CPU Review: Haswell, Unlocked, For $75). Intel even threw HEDT enthusiasts a bone with an eight-core -5960X, plus an attractive six-core -5820K.

Two years ago, I distinctly remember asking about the possibility of a desktop-oriented part with Iris Pro Graphics 5200. This manifested as the Core i7-4770R, and it was great in platforms like Gigabyte’s Brix. But BGA packaging limited its utility to enthusiasts.

This brings us full circle. Intel’s Core i5-5675C and Core i7-5775C are the first socketed desktop processors with Intel’s most advanced on-die graphics engine, Iris Pro Graphics 6200. At last! Both CPUs are compatible with the LGA 1150 interface, supported by existing 9-series motherboards after a firmware update. They’re also multiplier-unlocked, appealing to power users with a penchant for pushing additional performance.

What’s not to like, then? Most pointedly, Intel’s Skylake architecture is expected in a few short months. That’s a “tock” in the company’s cadence, representing a new microarchitecture leveraging the 14nm manufacturing process adopted for Broadwell. Beyond the improvements rolled into Skylake-based processors, Intel’s 100-series chipsets introduce a host of upgrades that enthusiasts will most certainly want (a faster Direct Media Interface, PCIe 3.0 from the PCH and more flexible overclocking on K-series parts—more on all of that later). We typically don’t hold off on recommending hardware, hoping for more from the next generation. But in this case, a distinct lack of interest in Broadwell from most of the system builders we’ve talked to or from Intel itself, really, more than suggests something better is on the horizon.

Still, I can’t help but admire what Broadwell on the desktop achieves. If only for the sake of deconstructing technology and discussing its significance, let’s take a closer look at Core i5-5675C and Core i7-5775C.

Four Cores, Lots Of Graphics And 65W

Both models sport 65W TDPs, so it’s little surprise that Intel says Broadwell is optimized for all-in-ones and mini PCs (though perhaps the soldered-down BGA models will fit better in those environments). Enthusiasts, these aren’t going to replace the Devil’s Canyon processors in your gaming PCs (or even older pre-refresh Haswell-based CPUs). Existing H97 and Z97 motherboards will support Broadwell, provided they receive new firmware. But it’s difficult to imagine a situation where upgrading makes a lot of sense. If you’re on an older Ivy Bridge or Sandy Bridge platform, Broadwell would require buying a new board, doubling your reasons to wait.

In that 65W power envelope, however, Intel crams four Broadwell-based IA cores, a dual-channel memory controller, lots of cache, 16 lanes of third-gen PCI Express connectivity and, most notable, the Iris Pro Graphics 6200 engine, which Intel is confident will circumvent your desire for discrete graphics in the compact form factors it’s targeting.

As host processors, Core i5-5675C and Core i7-5775C should be marginally faster than Haswell-based CPUs at similar clock rates. The issue, of course, is that they employ lower frequencies than a number of previous-gen chips. So, they'll actually post lower scores in workloads that emphasize host processing (like the Sandra Arithmetic benchmark, above). We discussed Broadwell and the process used to manufacture it in Introducing Intel's 14nm Node and the Broadwell Processor. Briefly, though, IPC improvements are only claimed to be around 5%, so performance gains are going to be muted. This is expected on a “tick” cycle; Intel’s focus was on transitioning a mature design to 14nm, after all. The Core i7 does benefit from Hyper-Threading technology, addressing up to eight threads concurrently, while the Core i5 is limited to one thread per core.

The two new BDW-LGA CPUs bear tell-tale signs of a mobile-oriented die configuration (aside from their muscular graphics engines). One is a smaller last-level cache. The Core i5-5675C features 4MB, while the Core i7-5775C comes with 6MB. At least on the desktop, both brands would be expected to sport an additional 2MB.

Then there’s the memory controller with DDR3L-1600 support. Yes, we got away with DDR3 at 1.5V; however, Intel specifies 1.35V modules.

And then there’s the Iris Pro Graphics 6200 engine, Intel’s crown jewel, imbued with more rendering, compute and media horsepower. Architecturally, it’s still an evolution of the HD Graphics design. But the shift to 14nm manufacturing gives Intel room in its transistor budget, opening the door for extra fixed-function and programmable resources. Amazingly, the company more than doubles the shader count of CPUs like Core i7-4770K with Core i5-5675C and Core i7-5775C, while cutting max power by 20W. Iris Pro Graphics 6200 is the GT3e implementation of Intel’s biggest Broadwell die, meaning it’s also accompanied by 128MB of on-package L4 cache. Intel’s API support is extensive, too. The company lists DirectX 11.2 and OpenGL 4.3, adding that it’s DirectX 12-ready and supports OpenCL 2.0, OpenGL ES 3.1 and Renderscript.

As you can see, the Core i7-5775C’s base clock rate is 3.3GHz, and Intel’s Turbo Boost technology pushes the frequency up to 3.7GHz in single-threaded workloads. Again, the CPU includes 6MB of L3 cache and 128MB of eDRAM. Hyper-Threading allows the quad-core chip to schedule eight threads at a time, while official DDR3L-1600 memory support facilitates up to 25.6GB/s across two channels. In a nod to enthusiasts, Intel ships the Core i7-5775C with an unlocked ratio multiplier.  

The Core i5-5675C is also overclockable through an unlocked multiplier — convenient since a 3.1GHz base clock rate and 3.6GHz peak Turbo Boost frequency aren’t that aggressive compared to several existing Haswell-based parts. Still, four Broadwell cores should suffice for all but the most taxing workloads, even if the Core i5 lacks Hyper-Threading support. Intel further differentiates the -5675C’s graphics performance with a maximum dynamic clock rate of 1100MHz for its Iris Pro Graphics 6200 engine, compared to the Core i7’s 1150MHz ceiling.

MORE: Best Gaming CPUs For The Money

Iris Pro Graphics 6200

From one generation to the next, Intel changes up the way it handles productization of its graphics. In the Sandy and Ivy Bridge families, high-end desktop processors came equipped with the company’s top trims: HD Graphics 3000 (12 execution units) and HD Graphics 4000 (16 EUs). Haswell saw the company deploy HD Graphics 4600 (referred to as GT2, with 20 EUs), saving HD Graphics 5000, Iris Pro Graphics 5100 and Iris Pro Graphics 5200 (that’s GT3, GT3 and GT3e, respectively, all rocking 40 EUs) for soldered-down CPUs.

In the image above, I numbered the six domains composing Haswell’s GT2, otherwise known as HD Graphics 4600. Domain three demarcates the Sub-Slice—a building block containing EUs, texture samplers, L1 instruction cache and a Media Sampler. Domain two is referred to as Slice Common, and it hosts the rasterizer, pixel back-ends and L3 cache. Together, those blocks make up the Slice.

A Slice in Haswell’s GT2 config included one Slice Common and two sub-slices, totaling 20 EUs. For Broadwell, Intel juggles the organization of resources to optimize for performance and power—each Sub-Slice is made up of eight EUs, rather than 10. But as a result of its shift to 14nm manufacturing, Intel can put a third Sub-Slice on GT2, yielding 24 EUs and more sampling throughput/cache per EU (and still reducing power versus Haswell, according to Jason Ross, graphics architect at Intel). The EUs themselves receive targeted improvements that relate to the architecture and implementation, bettering their performance and cutting power. For instance, the two SIMD floating-point units in each EU now support native 32-bit integer operations. Previously, only one did. The result is a doubling of integer computation throughput within each EU. The execution units also get native 16-bit floating-point support.

Broadwell’s GT3 adds a complete second Slice, doubling the already-faster GT2’s resources, including its fixed-function media capabilities. The math on that adds up to 48 EUs—a 2.4x increase compared to Core i7-4790K’s HD Graphcis 4600. And because there are three Sub-Slices per Slice instead of two, texture sampler performance increases 1.5x, while the FLOPS-to-texture ratio falls from 40:1 to 32:1. 

The gains are palpable. For a 140% increase in EUs, we measure between a 109% and 141% performance improvement, depending on the operation in question.

GT3e further incorporates 128MB of embedded DRAM on the processor package, behind its shared L3 cache on a dedicated ring bus stop. Not only does this benefit performance, but Intel says there are also advantages to power (and thus efficiency) as you avoid transactions that would have previously gone to system memory. The eDRAM operates in its own clock domain and, according to the firmware on MSI's Z97A Gaming 6, runs at 1.8GHz. At that frequency, and given read/write buses capable of 32 bytes/cycle, you’re looking at bi-directional throughput of over 57 GB/s.

Of course, as we know from last generation’s Iris Pro Graphics 5200, the eDRAM isn’t married to the graphics engine; it’s available to the IA cores as well.

Beefing Up Media In Broadwell

Intel has a storied history of design decisions rooted in simultaneous performance and power gains. More than four years ago, it introduced Quick Sync, again leveraging its manufacturing advantage to build a fixed-function engine for media encode/decode acceleration. The company lobbied ISVs to support its hardware, and a number of apps surfaced right off the bat to exploit it. Over time, Quick Sync has evolved to accelerate the latest formats, while giving developers more balance between quality and performance (target usages).

With Broadwell, Intel continues its quest to push more work at the fixed-function blocks optimized for specific tasks. These are faster than parallelized programmable logic (like EUs), which are in turn quicker than general-purpose IA cores. Because they involve fewer transistors, they also use a lot less power. That’s a win on two fronts—if you can afford to throw hardware at the problem. Intel, with its 14nm process, can.

So what does Broadwell on the desktop enable above and beyond Haswell? The Multi-Format Codec engine gets native support for 4096x2048 content, accelerating HEVC decode at up to 4Kp30 and VP9 at up to 4Kp24. This isn’t handled by a fixed-function block, though. Rather, Intel describes an approach involving the IA and graphics cores. This isn’t ideal, and the company is obviously working on a fully hardware-accelerated solution, but it’s better than nothing.

AVC/H.264 encoding receives a more substantial speed-up by virtue of the additional Sub-Slices (and the second Slice on GT3), since there’s a fixed-function Media Sampler—responsible for motion estimation—in each one. And because the EUs are used for rate control and mode decision, several steps along Intel’s familiar two-stage encoder run faster.

The Ivy Bridge graphics architecture included a sixth domain called the video quality engine, using dedicated hardware for video and image processing at very low power. Prior to that, those jobs were handled by the EUs. With Broadwell, the VQE is purportedly up to 2x faster.

Taken together, these improvements should have a profound impact on media performance, particularly in the context of desktop Haswell’s GT2 engine versus desktop Broadwell’s GT3e. One Multi-Format Codec becomes two. One Video Quality Engine becomes two, each with up to 2x throughput. Two Media Samplers become six, also sporting up to 2x throughput each.

SiSoftware Sandra 2015 appears to use Quick Sync for encoding, and these are the transcode results we measure. The H.264->H.264 task gets a 39% speed-up compared to Core i7-4790K and the WMV->H.264 workload enjoys 44% more throughput on Core i7-5775C.

Intel is also touting its end-to-end 4K support, which could be more relevant to the Core i5-5675C and Core i7-5775C than most other Broadwell-based processors, provided these wind up in small form factor and media PCs. The CPUs accelerate AVC/H.264 encode and decode at 4Kp60, along with HEVC decoding at 4Kp30 through the EUs and IA cores. Intel’s display controller can do up to 3840x2160 at 60Hz using DisplayPort 1.2 or 4096x2160 at 24Hz with HDMI 1.4. Unfortunately, HDMI 2.0 support didn’t make the cut.

How We Tested

Because Intel’s Broadwell-based desktop parts are compatible with its existing LGA 1150 interface, we didn’t need any special pre-production platforms to test it. We simply sought out an early firmware update to one of the recent additions to our reference collection: MSI’s Z97A Gaming 6.

This board employs familiar core logic, and is subject to the limitations of Intel’s 16-lane PCIe controller and PCIe 2.0-capable Z97 PCH. Still, MSI manages to set this platform apart by giving it USB 3.1 support through a Type-C port on the I/O panel, powered by ASMedia’s ASM1142. The company also exposes SATA Express connectivity and PCIe-based M.2, opening the door to higher-performance storage.

Although MSI touts the overclockability of its memory subsystem (claiming DDR3 data rates of up to 3200 MT/s), we’re more interested in the board’s ability to support lower-voltage DDR3L modules. G.Skill sent over its F3-12800CL9D-8GBXM kit. A 1600 MT/s data rate at CAS9 is ideal if you’re looking to match Broadwell’s specifications. A 1.35V rating also plays an important role in keeping these CPUs’ memory controller in-spec.

Most of the benchmarks from today's story were run by Tom's Hardware DE, using the same hardware that goes into generating our 2015 CPU Charts. In addition to desktop performance analysis, we also go into depth on power consumption, thermals and alacrity in workstation-oriented applications.

Power And Temperature In Detail

Idle Power Consumption

If our measurements and the sensors are to be believed, then the two CPUs’ power consumption at idle is well under 5W without integrated graphics. Running with integrated graphics enabled and without an external graphics card results in less than 6W. Desktop graphics that only consume 1W are unheard of. Whether you're talking about previous-gen Intel or current AMD solutions, every other implementation uses a lot more power.

Gaming Power Consumption

Let’s think back to our GTA V performance test. We’re rendering a power-hungry scene (driving a car at night) using the integrated recorder to create a load for our measurements. The average power consumption of Intel's Core i5-5675C comes in at a moderate 42W, whereas the Core i7-5775C consumes 52W. The peaks show that these numbers might go up if the CPU is maxed out. We’ll take a closer look at this soon with our stress test.

So what happens when the integrated graphics engine takes over rendering duties? This limits the processor's x86 cores, and its peaks are consequently not as pronounced as they were (starting toward the middle of our test run). The power consumption of the two CPUs becomes almost identical in this scenario, which makes sense, since the Iris Pro implementations are identical. The power consumption averages approximately 62W.

Maximum Power Consumption (Stress Test)

We’re putting the whole processor under maximum load to see how the picture changes. Both Broadwell-based models are plotted in separate graphs to make them easier to see. These charts show the telemetry’s frantic efforts to adjust the load nicely. This is similar to what we’re used to seeing from today’s graphics cards.

The more mainstream Core i5-5675C comes in at 65W, which is just shy of its 66W TDP. Intel's Core i7-5775C is more liberal, which is to say that it consumes 74W, or approximately 10W more than its smaller sibling. The reason for this is its higher clock rate, as well as Hyper-Threading technology keeping the available cores better-utilized. That latter feature yields a significant speed-up in tasks able to exploit threading (such as our stress test), though it does affect power consumption.

We’ll get to how all of these results stack up to Intel's previous-gen architectures and AMD's competition on the next page. Before that, though, we need to cover the by-product of power consumption: heat.

Temperatures at Full Load (During the Stress Test)

All parts of the processor (including the integrated graphics) need to be stressed to generate a significant amount of waste heat. The average temperature across all cores after 30 minutes is 52 degrees Celsius for the Core i5-5675C and 58 degrees Celsius for the Core i7-5775C. The package temperatures are a maximum of 40 and 43 degrees Celsius, respectively. We’re cooling the processors on our usual benchmark system with a Raijintek Triton 360 All-In-One open-loop compact liquid CPU cooler.

It’s hard to compare these temperatures to those generated by discrete graphics cards, since the new architecture's power consumption is significantly lower. If the performance-to-power consumption ratio is taken into account, then Broadwell stays significantly cooler.

Power Consumption Overview

Power Consumption Overview

An overview of power consumption confirms what we saw on the previous page. Comparing the two new processors to other CPUs and APUs shows the magnitude of Intel's optimizations.

Unfortunate outliers, such as AMD’s FX-9590 (under load) or the FX-4350 (at idle) are just one piece of the puzzle. If you're wondering why the Core i7-5960X looks almost like a low-power processor by comparison, remember that most of these CPUs operate close to their sweet spots, whereas AMD's larger FX models and a few Intel CPUs don't run efficiently, even at their stock settings.

There are a lot of CPUs in the charts that follow, spanning a wide range of the performance (and power consumption) spectrum. As you can see, Intel's two new processors consume a lot less power than the previous models at idle, which makes for a great start.

Broadwell sets new power consumption standards for gaming with a discrete graphics card as well. The reason for this is the Iris Pro's almost nonexistent draw. Other CPUs with on-die graphics engines tend to lose some power through them.

The two new processors do well, even under full load and with a fully active IGP. They consume significantly less power than comparable older models, in spite of providing equal or better performance.

Bottom Line

Intel's previous-gen and AMD's current architectures drive home how impressive the Broadwell design is, and how far Intel has come with it. Although integrated graphics is often derided, we're seeing the company's emphasis on dedicating more resources to graphics paying off in a big way. Based on our power consumption results, Intel's balance of host and graphics processing is a resounding success.

Iris Pro Graphics 6200: Gaming

Bioshock Infinite at 1920x1080 (DirectX 11)

Bioshock Infinite isn’t particularly demanding when it comes to graphics load (we excused it from our benchmarking suite quite a while back). However, even with our purposefully entry-level quality settings, the on-die graphics engine, not the CPU, limits performance.

Still, it's surprising that the Iris Pro 6200 with its 48 EUs offers more than twice the performance of HD Graphics 4600 found on Intel's Core i7-4790K. The company's newest design also beats AMD’s fastest APU by a massive 49 percent. The two Broadwell processors serve up 22 and 21 FPS at 1920x1080 with Ultra settings. Stepping down to the Medium preset gets you an average of 44 and 41 FPS.

These performance numbers are right around the level of an overclocked AMD Radeon R7 250X or Nvidia GeForce GTX 560 (non-Ti). That's nothing short of amazing when you consider that Iris Pro consumes somewhere between 10 and 12W.

Half-Life 2: Lost Coast at 1920x1080 (DirectX 9)

This classic has been gathering dust for a while now, but it provides a challenge that any integrated graphics engine should be able to master. Here, we have the chance to evaluate a title that's truly playable.

We’re using 2x MSAA to shift some load away from the CPU. As a result, the performance increase going from HD Graphics 4600 to Iris Pro 6200 is even more extreme. Broadwell enables frame rates three times higher. AMD's fastest APU, the A10-7800K, falls even further behind as well.

Grand Theft Auto V – Entry Level Battle

Our last benchmark is more modern, and decidedly more demanding. We’re comparing a budget-oriented system with entry-level or older graphics cards to AMD’s current APUs and Intel’s new Broadwell-based processors with Iris Pro 6200 graphics.

We also paired the fastest graphics card in this line-up with Intel’s Core i7-5775C to ensure host processing isn't limiting performance. As it turned out, the average frame rate didn't increase much compared to our machine with an AMD CPU. However, the minimum frame rate jumped quite a bit to 45 FPS.

Clearly, these results look pretty good for Intel’s graphics effort. The company's new processors are definitely faster than a Radeon R7 250 with GDDR5 memory, while consuming a lot less power. AMD’s fastest APU gets destroyed; Iris Pro 6200 is twice as fast, even with its slow connection to the shared DDR3-1600.

Bottom Line

AMD’s APUs do suffer the lower IPC throughput of their host processing architectures. However, Iris Pro 6200 is still significantly faster than any integrated graphics solution that we’ve ever tested, even without help from the Broadwell architecture's efficient x86 cores. Sure, the delta would shrink if we were testing lower-clocked CPUs. But there's just no way around it: the ball is in AMD's court now.

Iris Pro Graphics 6200: Workstation

AutoCAD 2015 2D and 3D Performance

AutoCAD is a popular application from Autodesk. First, we’re testing its "2D" performance with Cadalyst 2015. Quotes are used there because AutoCAD deals with 2D the same way many other applications do nowadays: through DirectX's D3D interface. This way of implementing 2D is worth testing since there really hasn't been any hardware acceleration for 2D through the kernel-mode driver since Windows Vista. Graphics cards with unified shader architectures don’t have dedicated 2D units anymore, either.

Consequently, this benchmark adds value to our review, since most of the 2D calculations are executed via the CPU these days. This means that they’re more dependent on host processing than the graphics card. This carries through to our results, which heavily favor the higher-clocked Haswell-based CPUs.

The picture changes as soon as 3D performance is involved. Intel's Broadwell architecture leads, and AMD’s APUs don’t stand a chance due to their weaker x86 cores.

Maya 2013 (OpenGL)

The SPECviewperf software suite uses OpenGL exclusively for Maya, manipulating a model made up of 727,500 vertices. 

Graphics processing limits this benchmark's performance, since the CPU load isn't particularly demanding. Intel’s new Core i7-5770C with Iris Pro 6200 provides up to 36 percent more performance than AMD’s Radeon R7 on the A10-7560K. Ouch.

Showcase 2013 (DirectX)

The next benchmark is based on DirectX. Autodesk might be alone in the field of large application vendors, but many smaller companies are making the move to DirectX as well. The benchmark for Showcase 2013 uses eight million vertices and, among others, shading, projected shadows and self-shadowing.

Based on those weak frame rates, it's pretty clear that integrated graphics isn't the way to go for an optimal experience. Still, Iris Pro 6200 comes out a whopping 109 percent ahead, even if that’s not enough to provide usable results.

Cinebench R15 (OpenGL)

Cinebench R15’s integrated OpenGL graphics benchmark places a bit more emphasis on the CPU, which becomes apparent when looking at the GeForce GTX 980’s different frame rate results. If you're only using processor graphics, however, then that becomes your bottleneck.

Desktop Publishing And Multimedia

Adobe CC

We’re using Photoshop, InDesign and Illustrator, all of which are included in Adobe’s CC package, as well as PCMark 8 Professional to control the workloads. This way, we’re covering a relatively a large range. The details of each benchmark are available in the table below.

The storage subsystem and background processes influence the benchmark results, since they include opening and closing each application, as well as loading and saving files. For this reason, PCMark 8 natively reports back the geometric mean of three benchmark trials (GEOMEAN).

Adobe Photoshop Light

Adobe Photoshop Heavy

Adobe InDesign

Adobe Illustrator

Bottom Line

Both Broadwell-based processors fall in line with their Haswell predecessors according to clock rate. The only notable exception is Adobe InDesign, where somewhat higher FP64 performance, including better scaling across multiple threads, sways the results upward a bit. It’s also interesting to see that AMD’s APUs fare well. They're even significantly better than the aging FX family in some places.

Office Productivity

Microsoft Office 2013

No desktop benchmark suite is complete without Microsoft’s popular Office suite. We’re leaving control over the workloads (as well as computing and reporting the geometric mean of three benchmark runs) to PCMark 8 Professional once again.

Microsoft Word 2013

Microsoft Excel 2013

Microsoft PowerPoint 2013

Bottom Line

Both CPUs perform as we'd expect, given their clock rates. It’s interesting that the Core i5-5675C is at least as fast as, if not a bit faster than, the Core i7-5775C. If Hyper-Threading was disabled, the two new processors would appear similar to their Haswell-based counterparts. This is probably due to small problems with the beta BIOS or Intel’s microcode.

Rendering, Encoding, Compression, Arithmetic

Blender

Blender has an efficient rendering module that runs exclusively on the CPU, even though rendering our benchmark file using GPU acceleration is much faster. A tile size of 16 pixels has proven to be the most efficient for CPUs, so we’re using it for our testing. Both Broadwell-based processors fall in line as expected.

Cinebench R15

This benchmark, which is based on Maxon’s Cinema 4D, provides the interesting option to have the CPU render in single- or multi-threaded mode. The ratio between the two says a lot about efficiency, illustrating the difference between physical cores, modules and simultaneous mult-threading implementations.

That relationship between single- and multi-threaded performance is of particular interest. The former is the same across both new processors, and comparable to Haswell. Multi-threaded performance, however, is better. This leads us to conclude that Broadwell's multi-core efficiency is improved compared to Haswell, as long as the BIOS and microcode play their parts.

Adobe Media Encoder CC

We’re using a UHD video file with an audio track that we’ve recorded ourselves. It’s saved as an H.264 file with 3840x2160 resolution, 25 FPS, progressive VBR, one pass, and a 320 Kb/s and 48kHz AAC stereo soundtrack. We’re using the integrated software renderer, which makes optimal use of all possible threads. The results aren’t surprising.

WinZip 19 Pro – Compression

The trick with this benchmark is to compress different types of content, such as text, pictures, multimedia files, videos and applications, without producing troublesome overhead due to time-sensitive file operations. This is why we copy all 3.02GB worth of data to an ISO file that can be compressed in one go. We’re using the CPU, not the GPU acceleration via OpenCL.

SiSoftware Sandra 2015 – Arithmetic

If the overall results for the single- and multi-threaded benchmark runs are compared, putting heavy emphasis on integer and 32/64-bit floating-point performance, then the Core i7-5775C is found right between Intel's Core i7-4790K and Core i7-4770K, in spite of the new processor’s lower core frequency.

The Core i5-5675C manages to inch out the much higher-clocked Core i5-4690K. This advantage only emerges if an application performs a lot of 64-bit floating-point operations, though. Otherwise, Broadwell isn't quite as fast. It looks like synthetic metrics, at least, can extract more FP64 performance from Broadwell, though integer and FP32 throughput doesn't really improve.

If you're only looking at single-threaded performance, the SMT-equipped processors do significantly worse. This helps explain why Intel’s Core i5-5675C does better in some applications than its bigger brother, the Core i7-5775C.

Workstation Applications

The following benchmarks are based on AutoCAD 2015, Cadalyst 2015, and three modules of SPECviewperf 2015. They were specifically chosen to represent the CPU performance in this context well, in spite of us using a Palit GeForce GTX 980 Super JetStream OC graphics card and not a workstation graphics card.

AutoCAD 2015 2D and 3D Performance

We’ve already described why and how we’re using AutoCAD in the integrated graphics section. Suffice it to say here that the CPU needs to help out quite a bit when it comes to 2D graphics acceleration, since this type of acceleration hasn’t existed via the GPU since Microsoft Windows Vista. Neither the driver model nor the unified shader architecture provides this functionality.

The positions in this benchmark are solely determined by the CPU, since the speed of the graphics is the same all around.

It’s interesting that the smaller CPU comes out a little bit ahead for 3D yet again. This trend reverses once SMT has been turned off, though.

Maya 2013

With Autodesk’s Maya 2013, a current and popular application was added to the benchmark suite. Viewport 2.0 isn’t part of the results on purpose, since it’s based on DirectX. This means that this benchmark is based exclusively on OpenGL. The render modes used for this benchmark are shaded, ambient occlusion, multi-sample anti-aliasing, and transparency, and the model consists of 727,500 vertices.

Showcase 2013

Showcase 2013 is a DirectX-based benchmark. Autodesk might still be the only major company to have made the jump to DirectX, but many smaller developers have taken the plunge as well. The benchmark model used for this benchmark uses eight million vertices, as well as render modes such as shading, projected shadows, and self-shadowing. What it comes down to in the end is the clock frequency.

SolidWorks 2013 SP1

The different models for our SolidWorks 2013 workloads range in size from 2.1 to 21 million vertices. Individual tests use the software's many rendering modes, including a shaded mode, shaded with edges, ambient occlusion, shaders and environment maps. Compared to SPECapc for SolidWorks 2013, the CPU test is gone, the number of models is lower and a benchmark with parallax effects was added.

Conclusion

Operating at lower clock rates with less shared L3 cache and 65W TDPs, Intel’s two new socketed Broadwell-based CPUs clearly aren’t intended to usurp the existing enthusiast-oriented Haswell processors. Power users, breathe easy. That’s not what today’s launch is about. The real star of the show is Iris Pro Graphics 6200, which absolutely destroys anything Intel previously offered for its LGA 1150 interface (not to mention AMD’s best effort to make APUs look good).

In fact, if you were to summarize Broadwell on the desktop in one run-on line, it’d be that a 14nm manufacturing process gives Intel a distinct advantage, which manifests as four IA cores fast enough for desktop workloads and a significantly more complex graphics engine able to hang with many mainstream add-in cards, all crammed into a modest 65W TDP.

From there, you can get into the intricacies: the Broadwell architecture’s increased IPC throughput, a more than doubling of EUs on the graphics engine, major enhancements to the media processing pipeline, 128MB of L4 cache that no other LGA 1150 CPU has enjoyed and optimizations for power, which amplify Intel’s efficiency story. Could you imagine if the company had a version of its quad-core, GT3e-equipped die designed for an 84 or 88W TDP?

Given what we do have our hands on, though, most Tom’s Hardware readers are going to ask, “What’s the point?” Anyone with an LGA 1150 motherboard already has a Haswell-based processor, limiting the allure of an upgrade to Iris Pro Graphics 6200. But if you already have a desktop PC with Haswell inside, you probably have a discrete graphics card too, since HD Graphics 4600 is…well, modest by modern PC gaming standards. Those of you on an older platform would need not only the -5775C or -5675C, but also a new motherboard. You’ve waited this long—why not hang tight for a few months for Skylake and start anew with 100-series chipsets, DDR4 and the return of unlocked 95W K-series CPUs?

Even if the Core i5-5675C and Core i7-5775C aren’t particularly practical right now, we still have to commend Intel for listening. Two years ago, we asked for a socketed CPU sporting the company’s grand effort to showcase integrated graphics, preferably in an enthusiast-friendly configuration. These processors come close to what we envisioned. They’re just victims of unfortunate timing.

MORE: Best Gaming CPUs For The Money

Chris Angelini is a Technical Editor at Tom's Hardware. Follow him on Twitter, Facebook and Google+.

Igor Wallossek is a Senior Contributing Editor for Tom's Hardware Germany, covering CPUs and Graphics. Follow him on Twitter, Facebook and Google+.

Follow Tom's Hardware on Twitter, Facebook and Google+.

A Spark Of Brilliance

Have you ever asked yourself why you need so many boxes in your living room? The cable box, the DVR, the gaming console, the Blu-ray player, the little silent-but-weak PC substitute that checks your Facebook page or Twitter feed—you don’t need all of that, right? Cable companies have gotten on-board with video recorders the size of small PCs, and console makers include Blu-ray drives. And then you can always retreat to your desk for PC gaming, right?

I’ve always said that a home theater PC should be able to do everything all of those other devices do, and faced strong push-back for it from those who think set-top boxes only need to play back media, silently. That a media PC should be limited in order to assure complete silence. That even a few decibels of noise audible in a quiet room is too much. Rather than continue arguing, I ditched the plans I had for covering living room entertainment, put away my own HTPC, and kept on writing reviews of more traditional desktop-class hardware.

Maybe I gave up too easily? Steiger Dynamics figured out that a strong message reaches beyond the silent movement's din (irony?) and into the hearts of true performance geeks. Here we have a machine that gives the impression of silence, relying on the logic that most people either can’t or have great difficulty discerning sound pressure levels lower than 16 decibels in natural environments.

The price for this machine is only $330 over the self-built option, with us using the closest-matching $400 OrigenAE case. If you subtract the $49 overclock fee and $99 cable service, Steiger only has about $189 in mark-up. And even if we deduct another $80 for the LED panel not present on the special Maven case, we’re still impressed by the value Steiger Dynamics offers the high-end PC market. But is the machine equally impressive?

Getting To Know The Maven Pure

With a thick face panel that wraps around the sides, Steiger Dynamics’ Maven chassis resembles the super-expensive S15V from OrigenAE, but with a smaller access panel and no digital display. Rather than hide the Blu-ray burner behind a door, Steiger Dynamics uses a slot-loading version.

The smaller panel hides a simplified connector set, with two USB 3.0 ports, headphone and microphone jacks, and an SD flash media interface.

The Maven chassis doesn’t have any rear-panel fan mounts, and instead has a 140/120 mm mount on its lid. This gives Steiger Dynamics more room for the super-quiet, closed-loop liquid cooler in our test configuration.

Note that the I/O panel port section represents Asus’ Z97-A. A recent switch to the similar Z97-AR helped Steiger Dynamics ditch the integrated VGA and DVI-D ports, while retaining the DisplayPort and HDMI outputs. The removed connectors are even less important to anyone who orders their system with a discrete graphics card, as shown.

Corsair’s H60 quietly cools Intel’s Core i5-4690K, EVGA’s GeForce GTX 770 pushes pixels quickly without much acoustic output, Seasonic’s Platinum Power SS-660XP2 feeds those devices at nary a whisper, and the entire assembly relies on high-efficiency components and reduced thermal output to retain low fan speeds.

Two Kingston HyperX 120 GB SSDs in RAID 0 offer exceptional performance noiselessly, while a 3 TB Western Digital Red drive provides NAS-oriented mechanical storage for user data. Super-thick panels keep most of the hard drive’s noise from being transferred out of the case, and foam padding on all large surfaces helps prevent it from being reflected out of the Maven’s vents.

Maven Pure Customizations

Every machine that Steiger Dynamics sells is a custom order, yet some are more customized than others. In the case of our Maven Pure Custom, that included a $49 CPU overclock service (4.30 GHz) to go with the manufacturer-overclocked graphics card and DDR3-1866 CAS 9 SDRAM.

The machine also comes with a cool little kit built into its soft foam packing.

Our configuration came with the power supply's bag and accompanying cable, along with a pair of white cloth gloves. It also included an 802.11ac/Bluetooth combo card, so Steiger Dynamics bundled the corresponding antennas.

A $15 add-in, the BlueRigger DVI-to-HDMI cable wasn’t tested, so we left if off the price sheet.

A three-ring binder includes a CD sheet with manufacturer-supplied driver discs, an owner’s guide, and a pouch for each component’s documentation.

Remaining leftover pieces from the case, motherboard, power supply, and graphics installation kits are found inside the accessories box.

Steiger Dynamics also sent along a CouchMaster Basic. Unfortunately, I don't have a suitably-photogenic couch. Shown above on one of its couches. To be fair, this $174 add-on was also untested and left out of our pricing calculations.

How We Tested Steiger Dynamics' Maven Pure Custom

Building for reduced noise does have a few drawbacks; the Maven Pure Custom’s GeForce GTX 770 has to face off against the Radeon R9 290X and 290 from our previous System Builder Marathon. It’s at least 20 decibels quieter than my $1600 machine though, and for those who don’t remember the significance, decibels are on a logarithmic scale.

Results: Synthetic Benchmarks

Ah, the GeForce GTX 770. Quiet, efficient, and simply mid-range next to more enthusiast-oriented graphics cards. Yet, this same part is probably necessary for Steiger Dynamics to reach its goal of ultimate low-noise performance. AMD's cards certainly aren't known for their acoustic conservatism.

The Maven Pure Custom blazes through PCMark, partly because of its RAID array and partly because of Steiger Dynamics' excellent overclock. The lack of Hyper-Threading can make a Core i5 easier to overclock than a Core i7, but I was still embarrassed by the excessive voltage it took to reach 4.3 GHz on my System Builder Marathon machine.

The Core i5 even performs well in Sandra’s CPU tests, which is why I’m happy that SiSoftware's synthetic suite isn’t used in our overall performance calculations.

The XMP defaults for the Maven Pure Custom’s DDR3-1866 C9 perform nearly as well as my own custom DDR3-2133 C9 overclocks.

Results: 3D Games

Ultra HD might be the next standard in super-quality home theater displays, but they're still relatively rare in our living rooms. Our System Builder Marathon machines are instead set up for extreme desktop gaming. Trendsetters might like 4K, however, 5760x1080 would look pretty awesome as well on a trio of 48” displays.

The Maven Pure Custom barely survives Battlefield 4 at Ultra quality and 4800x900, forcing anyone looking to run at even higher resolutions (like 3840x2160) to drop to Medium quality defaults. Our best machine reaches 5760x1080 at Ultra quality, but again leaves little room for a 4K upgrade.

Barring any unforeseen circumstances, both of our noisier comparison machines appear to have enough power left to run Grid 2 at Ultra quality  and 4K. The Maven Pure Custom is probably tapped-out at 5760x1080, but remember what I said about a wall of huge 1080p displays?

Arma 3 hammers all three machines, though the Maven Pure Custom barely stays above 20 FPS at 4800x900 and Ultra quality. Though the noisy machines are faster, all three need to drop to lower quality presets before running smoothly at 3840x2160.

At Ultra quality, Far Cry 3 finally pushes the Maven Pure Custom down to 1080p. That machine might have enough potential to hit 4K at the game’s High Quality preset, but the comparison systems are our only somewhat-secure bet against the future of high-quality panels.

Results: Media Encoding And Creativity

The virtually-silent Maven Pure beats our conventional desktops in iTunes and LAME MP3 encoding. That's what you get when you can get single-threaded apps running quickly on one core.

The benefit of logical cores and added cache are apparent in HandBrake and TotalCode Studio as a performance match between the $1600 SBM machine’s stock CPU and the Maven Pure’s overclock.

Nvidia’s technology works a little better in Photoshop’s OpenCL-based filters, so the Maven Pure takes its lead.

Extra cache and Hyper-Threading are nice features for the $1600 PC to have in Adobe Premiere, but Acrobat’s single-threaded engine is far more clock rate-dependent.

Results: Productivity And File Compression

ABBYY FineReader, Blender, and 3ds Max are all optimized for threaded architectures, which is what our cheaply-built $1600 machine’s Core i7 offers. But the Maven Pure Custom still beats our Core i5-based $1200 gaming PC.

We’ve seen this type of result in Visual Studio before, but I've never been able to figure out the exact cause. In this case, it’s probably related to the RAID driver, as I believe former application-specific performance problems were also found on some “Intel Enterprise Storage”-loaded machines.

WinRAR and 7-Zip seem to think that the $1600 machine’s Core i7 is great, placing the Core i5-based Maven Pure and our own $1200 machine on-par. WinZip EZ breaks the back of our $1200 machine, and only its builder could tell you why.

Power, Heat And Noise

Moderating noise usually involves a combination of quiet case technology, large coolers, and low-speed fans. Steiger Dynamics tackles the fan issue using the knowledge that less power in means less heat to put out.

A low-voltage overclock helps the Maven Pure Custom demonstrate moderate power levels in spite of its high performance.

Large coolers help the Maven Pure Custom keep heat levels down in spite of its low fan speeds.

The combination of low power and high performance gives the Maven Pure Custom exceptional efficiency. But how quiet is this thing, really?

The Maven Pure Custom is so quiet that I had to measure its noise level from a fraction of a meter away! But what happened to the other machines?

Our System Builder Marathon tests don’t include noise because the authors don’t have matched equipment and environments, which means that our measurements wouldn’t be comparable. On the other hand, I do enough audio measurements to give a good estimate that my own Q2 System Builder Marathon $1600 machine was running in the low 30s and fighting in the mid-40s from one meter away. Because decibel levels are logarithmic, that means it has over 100 times the sound energy. And because of the way people hear, it also sounds over four times as loud.

One meter is an important distance to remember, since it’s the distance that many industries use as a baseline. Speakers are rated at 1 W/1 m. Fan noise is also usually rated at 1 m. Yet, three meters is also important because it’s roughly the industry-standard distance for large-screen TV viewing.

(UPDATE: At 35 dBA measured, my Q3 System Builder Marathon machine was quieter than my Q2 build.)

High-End Gaming Meets Silence

Targeting the living rooms of high-end gamers, the Maven Pure Custom we received from Steiger Dynamics produced a scant 8.8 decibels at three meters and full 3D load. That’s about as loud as your own heartbeat. Most people would need to be inside a noise chamber to hear it. I'm sure the silence snobs will still snub the machine for producing some noise. Any noise. I say let them go back to the Via C3 platforms they were lauding back in 2001.

Performance enthusiasts know the score. If you can’t hear it, the noise doesn’t matter. I personally fought the purists reading Tom's Hardware with the argument of near-silence back when I was just a reader. And now that I’m doing the reviews, I’m very happy to see that a major builder has picked up the gauntlet. At least now Steiger has an editor on its side.

I’d love to give this machine our ultimate Tom's Hardware Elite award, as its craftsmanship and performance-to-noise is unsurpassed by anything we’ve tested. The only problem is that I’d need to compare it to something in the same class for a more definitive conclusion.

Similarly, I think it deserves a Tom's Hardware Smart Buy award for being only a couple hundred dollars costlier than a home-built machine using similar components and software. However, I’d need a similar machine from another vendor to prove that this one is a better value.

And so the Maven Pure gets our Approved recognition, recognizing its quality and value. For those of you who love the concept but disagree with the award, Steiger Dynamics even gives you the blueprint. That must be worth something to the most die-hard do-it-yourselfers, right?

More Performance, More Value

System Builder Marathon, Q3 2014: The Articles

Here are links to each of the four articles in this quarter’s System Builder Marathon (we’ll update them as each story is published). And remember, these systems are all being given away at the end of the marathon.

To enter the giveaway, please fill out this SurveyGizmo form, and be sure to read the complete rules before entering!

Day 1: The Budget Gaming PC
Day 2: Our Mainstream Enthusiast System
Day 3: The $1600 High-End Build
Day 4: Performance And Value, Dissected

Our latest round of System Builder Marathon machines saw Paul and Don chasing bigger overclocks while I simply tried to fix mine. Purchased just before Intel launched Devil's Canyon, my machine last quarter was stuck with a mere 4.2 GHz CPU overclock that required a massive 1.28 V to reach. Lacking the Haswell update's cooling advantage, my -4770K appeared to be nothing more than a reject, cast off from Intel’s binning process as the company began stockpiling anything resembling a good die for its next new model. Or maybe it was just bad luck-of-the-draw.

Paul switched his $500 PC to Intel’s low-cost overclocking CPU, the Pentium G3258, after noting a new way to use cheap boards with unlocked CPUs.

Don took advantage of a long-standing $75 discount on Zotac’s factory-overclocked GeForce GTX 770, putting any savings on his $1000 PC towards a larger CPU cooler.

Meanwhile, I avoided the binning tomfoolery altogether by ordering its flagship Haswell-based Core i7-4790K, using a recent motherboard price drop to offset the CPU upcharge.

But wait, didn’t we call these $600, $1300, and $1600 builds? In theory, we’re supposed to have $500, $1000, and $1500 to cover mandatory hardware, and negotiations with all three builders yielded $100, $200, and $300 for stuff that wouldn’t be needed to make the system operational (the platform).

That leads to $600, $1200, and $1800 budgets including the operating system, case, optical drive, and accessories. Paul can’t fit anything more than the OS into his $100, so he refers to his platform budget at $450. Don overspends, so he just changed the name of his $1200 machine to $1300. And though more money was available to me, I’m still trying to fit all of my hardware into a 3x multiple of Paul’s total hardware budget.

Though I blamed a lackluster CPU sample for my previous overclocking woes, I wanted to remove all doubt from your minds concerning this quarter’s build. Choosing a cheaper case allowed me to spend more on CPU cooling. Don had the same idea, but chose to add the cost of a similar cooler on top of his budget. He did keep the total cost below $1250 though, so I still would have probably called it a slightly over-budget $1200 PC rather than pretend its budget was higher. Then again, I’m treating my $1800 budget as if it were $1600…

How We Tested

Anyone familiar with my motherboard reviews or even my Q3 overclocking comments knows why I keep Haswell cores under 1.30 V. Longevity is why I gave up at 4.60 GHz rather than shoot for the moon at 4.7 GHz at 1.31 volts. Some enthusiasts, on the other hand, have been lucky enough to keep Haswell-based CPUs alive for extended periods at 1.35 volts, and anyone willing to saw-off a few memory fins probably isn’t worried about something so trivial.

With full specs including base height (for DRAM clearance) plus a full page of installation notes in our 2011 NH-D14 review, informed readers need not face these challenges.

Results: 3DMark And PCMark

3DMark shows progressive CPU and GPU scaling on all three of our machines, with the system that costs nearly 3x as much producing a little over twice the performance of Paul’s $600 PC baseline.

PCMark scaling is a little more rounded, as the $600 PC’s mechanical hard drive falls into a lower performance class compared to the SSDs of $1300 and $1600 systems.

Results: Sandra

Even though Don’s $1300 machine overclocked as well as my $1600 PC, Sandra's Arithmetic module reports that his overclock performance is less than my baseline. Since both Haswell-based processors have the same number of physical cores, I can only credit Intel's Hyper-Threading technology and a little extra last-level cache for allowing my CPU to outperform by such a wide margin.

Sandra's Cryptography routine shows closer performance scaling between the two quad-core PCs, while Paul’s little dual-core chip suffers from Intel’s disabling of various high-end features (including much of its cache and AES-NI).

Tired of my good-natured jabs over his previous memory benchmarks, Don Woligroski used DDR3-2400 to prove his machine’s metal in Sandra's Memory Bandwidth test, achieving 28 GB/s. Conversely, my lower-profile DDR3-1866 modules reach only 30 GB/s when manually configured to the same frequency and latency.

Results: Battlefield 4

Everyone wins in Battlefied 4 at Medium Quality. Paul lacks the performance potential to test at 5760x1080, but the target of his $600 PC is only 1920x1080. Reaching 60 FPS at 4800x900 without overclocking makes Paul the smart shopper.

Mr. Henningsen's $600 system drops out of the triple-wide race at Battlefield's Ultra quality preset, but still hits its 1920x1080 target. Don’s $1300 PC might be playable at 5760x1080, but that’s more of a sure thing as I approach 40 FPS with my overclocked $1600 machine.

Results: Grid 2

To say that the Grid 2 High Quality results from Don’s $1300 PC are incredible would be an understatement: I’d love to toss my drive on his machine and take another run at it.

We do know that this benchmark becomes CPU- and DRAM-bound at lower resolutions, and find it tough to believe that Nvidia’s architecture has that much less impact on these components. It's implausible, but still possible. I’ll let him have those numbers and keep my fingers crossed that a card I have coming my way will confirm them.

Those DRAM and CPU bottlenecks go away at higher settings, and that’s where we see reasonable performance scaling. Paul’s little $600 PC pulls through with 39.2 FPS at 4800x900 (and without overclocking), even though it only needed to reach 1920x1080 to be rated a success.

Results: Arma 3

Arma 3 plays smoothly enough at standard settings that any discussion of FPS is usually academic. It’s still nice to see that the cheap PC is powerful enough to play at 4800x900, while the bigger machines have frames per second to spare at 5760x1080.

This game remains playable on the $600 machine at Ultra quality, but only when overclocked and only up to 1920x1080. Even Mr. Hacksaw’s mighty $1300 machine coughs and sputters at triple-screen resolutions. At this point, our most taxing settings are an exclusive club with a $1600 minimum order.

Results: Far Cry 3

I’ve always been curious about Don’s low-resolution performance scaling in Far Cry 3, since it’s something neither Paul nor I have been able to replicate (even when I’m using the same hardware). Maybe it’s not that we’re doing something wrong—maybe it’s that he’s benchmarking differently.

The $600 machine survived all of its Far Cry 3 tests at our lower test settings, but becomes unplayable at our higher settings. According to Paul, smooth gameplay is possible by either dropping from 1920x1080 to 1600x900 or from Ultra to Very High quality.

The $1300 machine makes a valiant effort to reach 5760x1080 at Ultra quality, but even the $1600 PC struggles. Thankfully, the drop between average and minimum is small in this title, with a 22 FPS minimum recorded for the $1600 machine’s baseline test.

Results: Audio And Video Encoding

Paul’s $600 PC enjoys substantial performance-to-price advantages in single-threaded applications like iTunes and LAME MP3 encoding. The expensive machines win, but only because they’re clocked higher.

I could probably take a break and return from it while the $600 PC finishes its video encoding tests. My $1600 machine looks moderately faster than Don’s $1300 competitor, but it’s also only moderately more expensive.