There are some new signs of life in the next-gen AMD Ryzen development cycle. On Monday, Finalwire added “preliminary support for next-generation AMD desktop, server and mobile processors.” Yes, that means the system utilities development software team will have received some early information morsels about 'Ryzen 10000 series' processors, if AMD follows the expected naming convention that is. Twitter/X user HXL spotted the update, but at the time of writing, there has been yet another update to the latest beta mentioning processors such as the Intel Core Ultra 3 (AKA Arrow Lake-H/S) series.

As you can see, the first beta of the next-gen AIDA64 is graced by the preliminary support for AMD’s next-gen chips. However, before we go on, readers should be warned to rein in any excitement. Looking back at AIDA64's history, such an announcement typically comes around a year ahead of any product launch. Therefore, if this plays out as expected, 'Ryzen 10000' chips will likely be central to AMD’s product launch plans at Computex 2026.

Medusa Ridge coming to desktop

Even though it may be a year off, we have a few leaks-based information morsels about the 'Ryzen 10000' lineup that are worth summarizing here. On the desktop side of things, probably of greatest interest to our PC enthusiast audience, leaked info from March suggests upcoming Zen 6 architecture processors will remain compatible with the AM5 socket. These CPUs are apparently codenamed Medusa Ridge.

It would indeed be a great nugget of news for AM5 platform adoptees if it were to be confirmed that 'Ryzen 10000' chips would be compatible with the existing socket. Moreover, the potential move to 12-core chiplet dies (CCDs) with the next generation could be a major step for consumers.

We think it is likely that AMD will use TSMC's N3P (3 nm-class) process for its Zen 6-based products in 2026. With AMD’s 3D V-Cache packing processors proving to be a powerful lure to gamers, it is also expected to deliver Zen 6-based Ryzens into the X3D subfamily.

Will the 'Ryzen 7 10800X3D' be the next ‘gaming legend’ from AMD’s stables? Check out our Ryzen 7 9800X3D review to see the devastating mix of price and (gaming) performance it will have to eclipse.

Medusa Point on mobile

TL;DR How to Check Your PC's CPU Temperature

• Every CPU has a built-in temperature sensor that can be used to periodically monitor the health of your CPU.
• Tools like CoreTemp or NZXT's CAM provide a graphical interface to sensors and enable users to check their temps at-a-glance.
• Other tools to monitor CPU temperature include AIDA64, HWiINFO and HWMonitor.
• Each CPU has a safe temperature range in which they operate.
• CPUs at idle run at around a temperature of 50C but under heavier loads the CPU temperature can increase to 80C and even 100C for 12-14th Gen Intel i9.

Checking your CPU temperature isn't hard to do and you don't need to open the case as every CPU has built-in digital temperature sensors. You don't need to check your CPU temperature everyday, but you should check every few months to make sure your system operates at its peak potential. Checking your CPU temperature can also be used to diagnose your system and spot drastic performance slow-downs.

Luckily, checking your CPU temperature is easy and doesn't require opening your PC and sticking a thermometer inside. Every CPU has built-in digital temperature sensors, so you only need a software tool to read the measurements in Windows 11 or other operating systems. The easiest way to check your CPU temperature is to download a simple utility, like Core Temp or NZXT's CAM, to see your CPU temps.

Ultimately, keeping your CPU temperatures under control improves performance and reliability. This is especially true if you regularly strain your system with heavy loads like you'll do with the best CPUs for gaming, or if you're an enthusiast who strives for the best performance in CPU benchmarks.

Below, we'll explain how to check your CPU temperature, define the safe range of temperatures for a CPU, and outline what to do if your CPU's temperatures are too high.

How to Check Your CPU Temperature in 2025

Checking your CPU temperature is as easy as installing and using a software tool and then reading the output. There are multiple CPU temperature monitoring programs to choose from, with the best tools for checking CPU temp being any of the following (click to download):

• NZXT's CAM
• Core Temp
• AIDA64
• HWiINFO
• HWMonitor

However, there are other tools you can use to monitor CPU temperature. If you're overclocking your CPU and want more in-depth measurements, Intel's eXtreme Tuning Utility (XTU) and AMD's Ryzen Master software are designed by chipmakers and also offer expansive tuning options.

These are just a few examples, but we'll show you how NZXT's CAM and Core Temp work because our testing found that these two are the easiest to install and use. You can download CAM here.

While the CPU temp monitor CAM is intended to be used with NZXT's products, it works well as a casual monitoring tool in Windows 11, even if you don't have any NZXT hardware. You can use the software in Guest Mode to avoid creating a user account, and you can also disable the program from starting automatically with Windows if you don't plan on using it often.

Once installed, CAM offers an easy-to-use interface. The first block (PC Monitoring) lists the CPU's status, which shows the load, temperature, clock speed, and cooler fan speed. You can click on this block to access further details. As you can see, the current temperature of this system's CPU is 36 degrees Celcius (C), which is a healthy temperature for your CPU when idle.

CAM also has an overlay, which automatically turns on when you enter a game while CAM runs. This overlay can show your CPU's status and display temperature measurements while you play games.

You can also use the Core Temp tool, one of the best CPU temperature tools for Windows 11, to monitor the temperatures (download here). This is a simpler tool with a more basic interface. Just be sure to untick the freeware option during installation.

Core Temp provides temperature measurements for every core in your CPU. The left-hand side of the measurements shows the real-time temperature of each CPU core, and you can also see the measurements in the taskbar. But we're more interested in the Min and Max columns, which show the absolute minimum and the absolute peak temperature recorded while the program runs. For example, the CPU in the above image has a minimum temperature of 32C and a maximum of 51C, both of which are normal temperatures.

If you want to check your PC's CPU temperature during normal use, you can leave Core Temp running while you perform common tasks. However, if you want to measure the maximum possible temperature, you can check by running a Prime95 stress test for about 30 minutes (download here). Just be aware that this program puts an extremely heavy load on your PC, so you might encounter a blue screen of death (BSOD).

The best way to check your CPU temperature while gaming is to game for an hour or more and then check the program to see the maximum recorded CPU temperature. You should be concerned if this figure is at or beyond 95C. Anything between 80C and 95C (or above) leaves room for improvement.

How to Check Your CPU Temperature in the BIOS

You can check your CPU temperature in the system BIOS or UEFI, but be aware that this will only show your CPU temperature when the system is idle. That means you will see much higher temps when you are booted into Windows 11 or other operating systems, and the chip is under load. Not all motherboards list the CPU temperature in the BIOS, but nearly all enthusiast or DIY motherboards do.

It's pretty simple to enter the BIOS to check your CPU temperature. On most platforms, you simply reboot the system and click delete or F2 repeatedly as it restarts. Once you're in the BIOS, you will see the temperature listed in the Hardware Monitor section, which can also be named PC Health Status or Status, among other similar names. You should also check from within Windows 11 to get a better idea of CPU temperatures when using the PC (see below).

What's a Safe CPU Temperature?

When the CPU is idle but inside of the operating system (at the desktop), a great temperature is anything under or around 50C. Under higher loads, such as when playing a game, rendering a video, or other intensive tasks, your CPU consumes more power and, thus, runs at a higher temperature. This 'load temperature' is more important than idle temperatures (assuming idle temps are fine), so you'll want to periodically check your CPU temperature under load to ensure it's adequately cooled.

Under load, you want your CPU to stay under 80C, though some CPUs may run hotter in gaming laptops or small-form-factor (SFF) computers. Additionally, certain newer chips are designed to run at much higher temperatures, so it's important to double-check the specifications for your processor. For instance, AMD's Ryzen 5000 processors are designed to operate at up to 95C with a stock air cooler, while the Ryzen 7000 series is also designed to run at 95C even with a watercooler (notably, some models, like the Ryzen 9 7950X3D chips, have a lower 89C threshold - be sure to check).

Meanwhile, Intel's highest-end Core i9 12th-Gen Alder Lake, 13th-Gen Raptor Lake, and 14th-Gen Raptor Lake Refresh processors can run up to 100C during normal operation. These temperatures are within spec and expected for these specific processors. Luckily, Intel's newer Arrow Lake processors tend to run at much more ammenable temperatures, making them easier to cool.

As such, while the 80C threshold serves as a good general indicator, it is important to ensure that your processor doesn't exceed its 'TJ Max' (Temperature Junction Maximum). This is the safe limit for any given processor. This can also be listed as the TJunction or the Max Operating Temperature.

The TJ Max varies by chip, but most monitoring software lists the value, or you can check the chip's specifications on the manufacturer's page.

For older chips, you have some wiggle room to creep past 80C, but anything above 95C is typically critical. At this point, some CPUs will begin throttling, meaning the chip will reduce its clock speed and slow down to ensure it doesn't overheat, and your PC may turn off. This indicates the chips' self-protection mechanisms have been activated.

More advanced users who want the utmost confidence that their CPU can handle aggressive workloads should stress test their CPU to 100% using a program like Prime95 or AIDA64. When running a stress test, keep a close eye on the CPU temperatures and back off once they reach too high (above 95C). We consider an ideal stress test to be one hour long, though your maximum temperature will likely level off after 10-15 minutes. If you're looking for specific advice about temperatures and overclocking, head to our How to Overclock a CPU guide.

How to Fix High CPU Temperatures

If your CPU temperature exceeds its safe temperature thresholds under load, you should check your system to ensure the CPU's cooling is adequate. Here's a checklist of things to look for:

• Basic clean-up: Is your PC clean and dust-free (including the heatsink, radiator and intake fans/filters)? Canned air is a great solution for blowing out the built-up dust and grime.
• Are all your PC's fans spinning under load?
• When did you last apply fresh thermal paste between your CPU and CPU cooler? Thermal paste can degrade, so if it's been over three years, consider re-applying it with one of the best TIMs.
• Right-size your CPU cooler: Does your model CPU cooler specify a higher cooling capacity than your CPU's rated TDP?

For SFF PCs and laptops, it's possible that there's minimal cooling, as the device was never intended to be used under high loads for extended periods of time. For example, most laptops come with very compact cooling solutions that work well for short-term performance bursts but need to slow down during extended gaming sessions to stay below the shut-off threshold. Gaming laptops are often bulky because they're packed with extensive cooling systems.

If you are using a full-size gaming PC, however, and think your cooling should be adequate, you may want to re-apply thermal paste to your CPU. Most thermal paste's performance seriously degrades after about three years. Applying fresh paste (here's a list of the best thermal pastes) and cleaning the system from dust can offer much better cooling power and significantly better performance. This applies both to pre-built and custom-built PCs, and perhaps doubly so to pre-built systems; these systems often use sub-par paste that isn't applied consistently. It is always a best practice to re-apply thermal paste to any new pre-built system.

Additionally, as part of any PC tune-up, you'll want to ensure that your cooler is adequate. Many of the stock CPU coolers that come either bundled with the CPU or installed in pre-built systems aren't adequate to give you the full performance of the chip. This often applies to Intel CPU coolers on Core i5 and above processors. You'll need to follow the stress testing steps outlined above to check CPU temperatures under load in Windows 11 to ensure that the cooler doesn't get overwhelmed by the CPU when it is under heavy load.

FAQs: Frequently Asked Questions

• MORE: How to Overclock a CPU
• MORE: Best CPU for gaming
• MORE: CPU Benchmark Hierarchy
• MORE: Intel vs AMD

One of the many Zen 5 architectural changes announced at AMD's Ryzen 9000 Computex keynote was the architecture's bandwidth doubling of the L1 and L2 caches. HLX on X (formally Twitter) discovered an Aida64 benchmark run of a Ryzen 5 9600X engineering sample, confirming AMD's L1 and L2 cache tweaks.

The Aida64 benchmark reveals that the Ryzen 5 9600X offers almost 3,800 GB/s of read bandwidth on the L1 cache. Write bandwidth was rated at nearly 1,900 GB/s, and copy speed was tested at almost 3,800 GB/s, just like the read speed, all with a blazing sub-1 nanosecond cache latency, which we would expect from a first-level cache.

The L2 cache was rated at almost 1,900 GB/s for read bandwidth, write speeds was rated at almost 1,800 GB/s, and copy speed was rated at almost 1,900 GB/s. Latency was 2.8ns.

The same source that provided the Aida64 benchmark also provided an Aida64 memory benchmark of a Ryzen 5 7600X Zen 4 processor for comparison. The results confirm what AMD disclosed in its Computex announcement. The Zen 4 part boasts nearly 2x less raw bandwidth in its L1 and L2 cache, compared to the Zen 5-based Ryzen 5 9600X.

The 7600X boasts a 2,000 GB/s read and copy speed and a 1,000 GB/s write speed (with the same latency). The L2 cache is rated at just barely over 1,000 GB/s across the board, read, write, and copy speeds (with a slightly less but indistinguishable 2.6ns latency result).

CPU cache is one of the most important components in modern CPUs. A good caching system will keep the CPU cores fed with data consistently in a number of workloads with minimal downtime. A poor caching system, or a lack thereof, will lead to poor CPU performance as the cores have to wait for data to be transferred from much slower system memory.

We have yet to see how Zen 5's L1 and L2 bandwidth improvements apply to real-world performance, or even synthetic benchmark performance. But undoubtedly, these heavy bandwidth improvements are part of what helps boost Zen 5's IPC improvement by a reported 16% over Zen 4.

The Ryzen 5 9600X is AMD's latest mid-range desktop CPU based on the Zen 5 architecture. The chip boasts 6 cores and 12 threads, a peak turbo frequency of 5.4GHz, and 38MB of cache.

Der8auer reports that Lamptron is selling illegal AIDA64 keys with its LCD-equipped products  — including LCD air coolers and dedicated LCD hardware monitoring displays. FinalWire Ltd., makers of AIDA64, confirmed to Der8auer that Lamptron is bundling free invalid AIDA64 keys with its products and has been doing so for some time. FinalWire has requested that Lamptron stop this shady practice, but it has not. Der8auer suspects that Lamptron is using some sort of key generator to generate fake AIDA64 keys that will validate, but only on select versions of the software. 

The German overclocker first discovered Lamptron's shady tactics through one of his employees who heard about Lamptron AIDA64 product keys not working on a forum two years ago. Deciding to see if this was true or not, Der8auer pulled out a "free" AIDA64 key from his Lampton ST060 air cooler to see if it would work.

He found that the key works as advertised. The box says the AIDA64 key is a "free version and does not support online upgrades." Plugging the key into AIDA64 activates the product without any issues. However, updating AIDA64 caused the key to become invalid (just as the box info says).

After contacting FinalWire, the AIDA64 makers confirmed that Lampton's "free" AIDA64 keys are invalid and illegal. It reports that Lampton registered as a reseller two years ago but only placed a single order for AIDA64 Extreme. FinalWire finally reveals that it is unclear if Lampton ever bundled valid licenses with its devices.

"We understand your concerns regarding the "FREE VERSION" license included with the cooler. Unfortunately, we've discovered instances of Lampton bundling invalid AIDA64 licenses with their products. When we became aware of this, we immediately contacted Lamptron and requested they cease this practice. Although they registered as a reseller two years ago and placed a single order for AIDA64 Extreme, it's unclear if they ever bundled valid licenses with their devices."

Der8auer tested several other Lampton AIDA64 keys from several other products and discovered that all of them were invalid. Some of them were labeled as different versions as well, including AIDA64 Extreme and AIDA64 Business, even though Lampton only officially bought AIDA64 Extreme at one time.

Lampton is using its free AIDA64 keys to power its LCD displays. The company's displays don't have any software included and thus requires 3rd party software to operate. AIDA64 has a feature that allows users to make their own OSD specifically for secondary hardware monitoring displays.

The German overclocker shared all five additional keys he tested to FinalWire. FinalWire reported back that all keys were invalid, which strongly suggests Lampton is using some sort of key generator.

"Unfortunately, none of the product keys you shared are valid in our system. This strongly suggests they were generated illegally using a tool like keyGen," stated FinalWire Ltd.

For now, no one has taken legal action against Lampton for selling illegal AIDA64 keys. But if it continues to provide these keys against FinalWire's wishes, the company is obviously at continuous risk of serious consequences.

FinalWire on Tuesday announced that the latest version of its AIDA64 diagnostic and benchmarking software has gained preliminary support for AMD's upcoming processors based on the Zen 5 microarchitecture. In addition, it "improved" support for Intel's forthcoming Meteor Lake, Arrow Lake, and Lunar Lake processors.

AIDA64 Engineer 6.90 software can now recognize AMD's processors based on the next-generation Zen 5 microarchitecture, according to release notes for the program. This essentially means that AMD has provided FinalWire a set of identifications for its upcoming CPUs, an important development milestone. 

Unfortunately, this does not give any clue about when AMD is set to introduce these processors. For example, FinalWire added preliminary support for Intel's codenamed Meteor Lake CPUs back in late 2021, and these products are set to arrive only later this year. Lunar Lake, among the chips receiving "improved" support, is expected to arrive in late 2024 or early 2025 and is expected to feature a brand-new microarchitecture. 

In addition, AIDA64 Engineer 6.90 now has "improved support" for Zhaoxin's upcoming Yongfeng processors for client PCs and servers that are set to feature up to 32 cores.  

Last but not least, the latest version of AIDA64 comes with an AVX2-optimized benchmark for Intel's Raptor Lake Refresh processors, though it is not completely clear whether this benchmark has something special particularly for the refreshed versions of Raptor Lake, or if it also includes other chips.

AMD has already formally announced that its Zen 5-powered processors will belong to the Ryzen 8000 family that is planned for release in 2024. AMD's Zen 5-based client CPUs —Granite Ridge for desktops and Strix Point for laptops — are expected to incorporate RDNA 3.5-powered integrated GPUs. 

TL;DR How to Check SSD Health in Windows 10 and Windows 11

• Download and install CrystalDiskInfo, then run the app
• Several SSD makers provide their own free drive health utilities that provide more than just the stats. Download the tool from the manufacturer of your SSD.
• Check from the command line: Click the Search button from the Taskbar > Type in "CMD" and hit enter > Type "WMIC" and hit enter > Type "diskdrive get status" and hit enter

To keep your PC running in peak condition, it's essential to check the vitals of critical components periodically. One of the most important components is your SSD, which typically stores your OS and application data. To ensure your SSD operates in peak condition, it’s advisable to perform a health check periodically. This will also tell you how much of the write endurance of the SSD you have left.

Fortunately, Windows 10 and Windows 11 include a built-in tool that enables you to monitor SSD health, and third-party apps are also available. Check your SSD health if:

• Your PC runs slower than usual overall
• Applications launch slowly or crash
• Blue Screen of Death errors frequently occur
• Your SSD flags itself as read-only

Luckily, there are a few ways to monitor your SSD’s health using SMART (Self-Monitoring, Analysis, and Reporting Technology) built into storage devices and software utilities that interface with it.

Check SMART Using CrystalDiskInfo

CrystalDiskInfo is a helpful utility that taps into SMART on your SSD and reports those values back in an easy-to-use interface.

1. Download and install CrystalDiskInfo from the app publisher’s website.

2. Open CrystalDiskInfo, which immediately presents you with all pertinent information relating to your SSD(s) and HDD(s)

In this case, you can see that the SSD in this system has a “Power On Hours” value of 18,857 hours and has accumulated 7,654GB worth of writes during its lifetime. You’ll also find details like current SSD temperature and overall health. The installed SSD has 95 percent life remaining (Good), which refers to its wear leveling count.

Using the Windows 11 Command Prompt to Monitor SSD Health

While you can certainly download and install apps like CrystalDiskInfo and AIDA64, Windows 11 also has built-in analysis tools.

1. Click the Search button from the Taskbar

2. Type in "CMD" and hit enter

3. Type "WMIC" and hit enter

4. Type "diskdrive get status" and hit enter

You will now be presented with one of two results in the command prompt. If you see "OK," your SSD is operating nominally. If you see "Pred Fail," it's likely that your SSD is malfunctioning, and it would be wise to back up your data as soon as possible.

Check SMART Using AIDA64 Extreme

1. Download and install AIDA64 Extreme (30-day trial) from the publisher’s website.

2. Open AIDA64 Extreme and click on the Storage icon from the main menu.

3. Click on SMART.

Like CrystalDiskMark, AIDA64 Extreme reports the SMART data, but in a slightly different way. AIDA64 indicates that the drive has 100 percent of its estimated useful life and reports life writes in terabytes. It also lists the total “Power-On Time” at 785 days. Also noted are the same SMART attribute values (threshold, current, and worst).

Using Manufacturer-Specific Tools to Monitor SSD Health and Scan/Fix Errors

Several SSD makers provide their own free drive health utilities that provide more than just the stats.

Seagate’s SeaTools provides similar SMART reading capabilities as AIDA64 Extreme and CrystalDiskInfo (including attribute tracking and power-on hours), but it also allows you to run several non-destructive diagnostic tests. It also works on non-Seagate drives.

The Short Self Test completes within a few minutes and simply reports if the test Passed or Failed. The Short Generic Test also gives a Pass/Fail grade and takes about a minute to run.

SeaTools can even attempt to fix errors (should they appear in the preceding tests) using the Fix All Short and Fix All Long commands. Fix All can repair and replace bad blocks, so the app warns you to back up your data before proceeding.

Samsung Magician Software is another popular utility with SMART diagnostic capabilities and the ability to monitor SSD health. As it's Samsung-specific, the software can also let you know if the SSD in your system is genuine. Samsung also loads up the software with a performance benchmark, support for over provisioning to boost performance/longevity and the ability to perform firmware updates for your SSD.

Yet another option is the Western Digital Dashboard, which is similar in concept to the Samsung Magician Software, as it allows you to check drive health/SMART status, keep an eye on temperatures and even upgrade to newer firmware all from one utility.

MORE: Best SSDs

MORE: Best Hard Drive

MORE: Best External SSDs

MORE: How We Test HDDs And SSDs

MORE: All SSD Content

Over the past few months, there have been numerous reports of Samsung 980 Pro SSD failures, with many complaints originating from China. We never heard an official statement from Samsung on the matter, but custom PC builder Puget Systems definitely took notice after it "received a surprising number of reports of failing Samsung drives." Puget worked with Samsung to identify the problem, and a firmware update is available for affected 980 Pro SSDs.

According to Puget, the failures traced back to the 3B2QGXA7 firmware and primarily affect 2TB versions of the 980 Pro (although lower capacities are not entirely immune). Puget doesn't detail the cause of the failures, but Reddit user reports suggest that the SSDs report a S.M.A.R.T. value of 0E (media and integrity error) or greater in CrystalMark or AIDA64. In addition, an 03 value (available spare) lower than 100 percent may also be reported. It's recommended to use the Samsung SSD Magician software to scan the drive for errors if the 03 or 0E value is out of spec.

"The newest one is one of the 2TB ones. Over 2000 errors already," reported user RaXelliX on Reddit. "And the middle one 2TB has already failed with over 32,000 errors and is locked to read only mode. I fully expect the two others to fail soon too."

"Bought a 2TB 980 Pro in February of this year and installed it a few months later in May. Manufactured August 2021," added Meme_Attack in a separate Reddit thread. "Lasted a grand total of 4 months before I noticed Media Errors ballooning from 300 to 6,000+ after a full Samsung Magician scan. Also found 9 red blocks through the same scan. Lost 5% of available spare, drive's at 95% as I speak. 4.1 TB written. Super careful and light use." 

So, what happens if your drive reports abnormal 0E or 03 values? Well, it usually means that your 980 Pro is on a one-way ticket to an early death. According to Puget, the most common issue that users face is that the SSD goes into read-only mode. Given that most users have their OS installed on the SSD, read-only access makes the boot drive unusable. Unfortunately, once your 980 Pro goes into this failure state, there is no way to recover your data, even with the firmware update.

However, the new firmware update from Samsung will prevent the failures from occurring in the first place if your drive is still functioning. So, if you have a 980 Pro running the 3B2QGXA7 firmware, you'll want to update to the newest firmware (5B2QGXA7) using the Samsung SSD Magician software ASAP. It should be noted that 980 Pro SSDs running the 4B2QGXA7 or 5B2QGXA7 firmware are not affected by this issue.

Puget Systems customers who have already experienced a Samsung 980 Pro SSD failure or have problems using the Samsung SSD Magician software can contact the company for guidance. All other 980 Pro users would need to reach out to their respective system builder or contact Samsung directly if the SSD was bought from a third party.

One of the world's most popular diagnostics and reporting software, AIDA64, has just received a significant update across its suite of products. The newly released v6.75 of the software brings support for AMD's upcoming Zen 4 processors and the AM5 motherboards they'll slot into. It's not all AMD though, as support for Nvidia's upcoming GeForce RTX 4090 has also been added. These additions bring support for some of the year's likely biggest hardware releases, and show just how closer these bits of hardware loom.

Besides improved support for AMD's Zen 4 - the company's  most important CPU release from the past years - AIDA64 v6.75 adds preliminary 64-bit multi-threaded AVX-512 and AVX2 optimized benchmarks for the architecture - a feature Intel dropped with the release of Alder Lake and its x86 take on Arm's Big.Little architecture. The release also adds detailed chipset information for the integrated memory controller of the new AMD processors.

Sadly for us all, there aren't that many details that can be gleaned from the added support for Nvidia's upcoming RTX 4090, which has been talked about as offering a 30% performance improvement over the RTX 3090 Ti. Do take note that all currently circulating information is nothing but alleged leaks and salted rumors, so adjust your expectations accordingly. That said, the addition of multi-threaded OpenCL GPGPU benchmarks, graphics processor, OpenGL and GPGPU details, temperature and cooling fan monitoring for this particular graphics card is a clear signal of its approaching release.

Besides support for Nvidia's next crown jewel of a GPU, the latest version of AIDA64 extends GPU diagnostics support for a plethora of graphics card across both AMD and Nvidia.

On the AMD side, FinalWire lists the AMD Radeon 6700, Radeon Pro W6600, Radeon RX 6650 XT, Radeon RX 6750 XT, Radeon RX 6950 XT and the AMD Rembrandt SoC. Chinese discrete GPUs also get some love - specifically the Glenfly Arise, which on paper should be roughly equivalent to an Nvidia GTX 1630. Finally, on the Nvidia camp, we're looking at improved support for the GeForce GTX 1630, GeForce MX570 A, GeForce RTX 3050 OEM, GeForce RTX 4090, H100 96GB, RTX A500 Laptop, RTX A1000 Laptop, RTX A2000 8GB Laptop, RTX A3000 12GB Laptop, RTX A4500 Laptop, RTX A5500 Laptop, T550 Laptop.

FinalWire's latest additions to its AIDA64 v6.75 are applicable throughout tits software suite, across its Extreme, Engineer, Business and Network Audit products.

Patch notes for a new beta version of AIDA64 -- version 6.60.5944, show preliminary support for a mysterious new Intel CPU platform, which carries the Alder Lake-X codename. We don't have an official word on what these new CPUs are, but on the surface, these chips could be new HEDT models of Intel's Alder Lake CPUs, meaning Intel could be returning to the HEDT market after a long hiatus.

In recent years Intel began using the letter X to symbolize its HEDT platform, including the chipset and the CPUs related to the platform. So it's easy to guess that Alder Lake-X will probably be a continuation of that tradition. However, it's been a while since we've seen a HEDT processor from Intel. The last HEDT lineup debuted in 2019 in the form of Intel's 14nm Cascade Lake-X, maxing out at 18 cores.

Intel produces Alder Lake and Sapphire Rapids on the Intel 7 node, formerly known as the 10nm Enhanced SuperFin process. Both processor families leverage Intel's Golden Cove cores, although Sapphire Rapids won't incorporate the Gracemont cores as Alder Lake did. It makes sense since the first targets data centers, whereas the latter aims for the mainstream market.

One theory is that Alder Lake-X could be the enthusiast-friendly version of Intel's Sapphire Rapids platform with an unlocked multiplier for overclocking and other features. However, there were early rumors that Intel may launch Sapphire Rapids-X, and if they're accurate, then Alder Lake-X and Sapphire Rapids-X would need to coexist. Nevertheless, it's not impossible, and we can see it working out for Intel.

For example, Intel could market Alder Lake-X with higher core counts than the consumer Alder Lake chips and maybe kill off the Gracemont cores, which we doubt will be helpful for HEDT enthusiasts. Alder Lake-X reminds us of Kaby Lake-X, which launched alongside Skylake-X. Intel's intention for Kaby Lake-X was to provide entry-level SKUs for its HEDT platform. On the other hand, Intel can sell Sapphire Rapids-X with similar core counts but with octa-channel memory and AVX512 support. Remember that Intel has fused off AVX512 from its Alder Lake chips for good. So, the only way to get AVX512 is through Sapphire Rapids, and Sapphire Rapids-X could be a budget-friendly version for consumers. Of course, all of this is just our estimated guess.

At any rate, Alder Lake-X will have to deal with AMD's latest Ryzen 5000 Threadripper Pro CPUs, spanning up to 64 Zen 3 cores. Unfortunately, AMD hasn't confirmed whether it will launch the non-Pro variants. The Ryzen Threadripper shortage started in the fourth quarter of last year and has been plaguing vendors, putting in doubt whether the regular Ryzen 5000 Threadripper lineup will launch or not.

Hopefully, we'll know more about Alder Lake-X in the future, but it seems Intel is looking into the idea of coming back to the HEDT market. Alder Lake-X will give AMD some other competition, which is always a welcome sight.

The latest version of FinalWire's AIDA64 benchmarking and diagnostic software brings in support for Nvidia's unannounced GA102F graphics processor. The chip is cloaked in mystery as Nvidia seems to be rolling it out very late in Ampere's lifecycle, and it has never previously added letters to designations of its GPUs.

FinalWire's AIDA64 version 6.60.5933 beta can read GPU information for Nvidia's GA102F processor, according to release notes (discovered by @momomo_us). The release notes do not disclose what product will be powered by the GA102F or how the chip will be used, but it was important enough for Nvidia share its ID and specifications with FinalWire to include in AIDA64.

Being the world's largest supplier of graphics processing units with about 85% market share, Nvidia can afford the luxury of developing GPU silicon aimed at very specific market segments. Just recently Nvidia introduced its GA103 chip (featuring a 496 mm^2 die size) that was specifically for the laptop-bound GeForce RTX 3080 Ti, but which is also being used in the desktop GeForce RTX 3060 Ti. The design costs for such a chip are in the hundreds of millions of dollars range, but as it can be used in a wide range of products from enthusiast-grade gaming laptops to performance graphics cards, its design is well justified.

But Nvidia's GA102 is a massive 628 mm^2 graphics processor containing 28.3 billion of transistors, so its design costs are in the hundreds of millions and its re-spin costs are extremely high as well. Making a new revision of this chip a couple of quarters ahead of the rumored Ada / RTX 40-series launch is an extremely odd move, given GA102 is only used for a relatively limited range of products. Nonetheless, just a couple of weeks before the company is expected to launch its flagship GeForce RTX 3090 Ti GPU, AIDA64 has added GA102F support.

Perhaps GA102F is indeed a version of GA102 with some changes designed to improve yields of full-fat silicon and make the GeForce RTX 3090 Ti somewhat more practical from manufacturing point of view. Or maybe GA102F is a version of GA102 designed specifically for cloud datacenters and therefore prone to high temperatures and constant load. Another possibility is that this is a chip aimed at cryptocurrency mining that's been adjusted in some fashion — or potentially the opposite, a chip designed specifically not to do well at mining.

Assuming the GA102F ends up in cards for the PC gaming market, it will certainly compete with the best graphics cards. We expect more details will come to light in the coming weeks.

According to the latest set of beta release notes for a popular PC benchmarking tool, we will soon witness the launch of a new breed of Nvidia GeForce RTX 3060 Ti desktop graphics cards. Outwardly the specs should remain the same, without a new model number/designation. However, buyers will find that the GPU under the shroud is the GA103 rather than the GA104 of old. 

HXL spotted this change in the AIDA 6.60.5918 beta downloads page. Please note that Nvidia hasn't confirmed the change, so take this information with a pinch of salt.

Earlier today, we were speculating how Nvidia's Mysterious GA103 GPU might find its way into the desktop graphics card market. The GA103 GPU came to light when a curious PC enthusiast in China decided to strip down his shiny new GeForce RTX 3080 Ti-powered laptop to take a closer look. Some expected this new mobile GPU to be a severely cut-down GA102, but instead, a brand new piece of silicon was discovered. We pondered specifically whether a desktop version of this GPU might target the area between the RTX 3070 Ti and RTX 3080, but as revealed by AIDA, we are looking at a rung or two lower at this time – a desktop GeForce RTX 3060 Ti.

See the comparison table above to get a grasp of what silicon surgery Nvidia is performing to create a GA103-based GeForce RTX 3060 Ti. You can see that the new RTX 3060 Ti card's GA103 GPU has 42% of the CUDA cores possible. At the time of writing, the GA104 is the most widely used GPU by Nvidia, but its production planning, with the updated laptops having just arrived, might have made it sensible to juggle a few SKUs. 

We can't be sure of the reasons behind the "refreshed" desktop RTX 3060 Ti, but it makes good economic sense for NVIDIA (i.e., economies of scale with production, die error rates and keeping things simple).

Before signing off, we must comment on the possibility that there may be slight performance differences observable by PC enthusiasts who get their hands on the GA103-based RTX 3060 Ti. While the key tech specs like core counts, clocks, and memory remain the same, there is a chance here that the "new" model could run cooler, or overclock better, for example. 

Nvidia would have been very careful to make sure there were no downsides to the change to avoid accusations of bait and switch. So it is reasonable to expect such slight benefits. With this in mind, we look forward to seeing tests of a refreshed RTX 3060 Ti, if and when it does materialize.

The latest version of popular hardware monitoring software AIDA64 (via HXL) supports Intel's Core i9-12900KS. The new update lends credibility to the recent rumor that the chipmaker is preparing a highly-binned Core i9-12900K.

It's been a couple of generations since Intel released a KS-series processor. We last saw one in the Coffee Lake days with the Core i9-9900KS. However, Intel is willing to give the current Alder Lake flagship a similar treatment, which seems fitting since the microarchitecture has put the chipmaker back on the map again and on the best CPUs list.

The Core i9-12900KS will most likely retain the same 8 + 8 configuration as the Core i9-12900K with eight Golden Cove cores and eight Gracemont cores. As the "S" suffix implies, the Core i9-12900KS is a special-edition SKU, meaning it'll come with higher clock speeds than the vanilla Core i9-12900K. The previous Core i9-9900KS boosted to 5 GHz on all cores, 300 MHz higher than the typical all-core boost clock speed for the regular Core i9-9900K. We can expect the Core i9-12900KS to arrive with a similar uplift.

Intel Core i9-12900KS Specifications

*Specifications are unconfirmed.

According to VideoCardz's sources, the Core i9-12900KS reportedly features a 5.2 GHz boost clock across all eight Golden Cove cores. Depending on the workload, that's approximately a 200 MHz increase over the standard Core i9-12900K with a 5 GHz all-core boost clock. The Core i9-12900K often boosts to between 4.8 GHz and 4.9 GHz, whether the workload includes AVX or AVX2 or just SSE. So it's feasible that the Core i9-12900KS will have no problems breaking the 5 GHz barrier right off the bat.

A higher boost clock speed also means higher power consumption. The Core i9-12900K already has a 241W PBP (Processor Boost Power) rating, Intel's fancy term for PL2. The Core i9-12900KS may push this value closer to the 300W range.

Being a "KS" chip, the Core i9-12900KS will unquestionably present a slight premium over the Core i9-12900K. The Core i9-9900KS costs 5% more than the Core i9-9900K at launch time. If we use the same margin, the Core i9-12900KS may end up selling for around $629 since the Core i9-12900K already retails for $599.

Of course, it's arguable that you can overclock the Core i9-12900KS to match the Core i9-12900K regarding the all-core boost clock. However, not everyone likes overclocking, and mileage varies from chip to chip. Therefore, the Core i9-12900KS will be a suitable option for consumers who want the fastest Alder Lake processor out of the box.

FinalWire just released the latest 6.60 version of AIDA64, its diagnostic and benchmarking software. The latest version brings in preliminary support for Intel's next-generation Raptor Lake processors as well as Nvidia's yet-to-be-announced top-of-the-range graphics processor for mobile PCs, essentially confirming the green company's plans to launch the rumored GeForce RTX 3080 Ti laptop GPU. We don't have detailed specifications for either of those yet, but assuming you manage to get your hands on an early sample, AIDA64 will apparently know what it is.

Enhanced CPU Support 

FinalWire's AIDA64 v6.60 update introduces multiple innovations, including AVX-512 and AVX2 accelerated benchmarks for Intel's Alder Lake (which do not officially support AVX-512). It also has support for Raptor Lake processors and preliminary support for the next generation Meteor Lake CPUs that are due to land about two years from now. In addition, the new version of the tool includes an AVX accelerated benchmark for various Zhaoxin and Hygon C86 Mukti/Dhyana processors.

But while it is nice to see enhanced support for Intel's Alder Lake and Raptor Lake CPUs as well as Chinese CPUs, the confirmation of Nvidia's GeForce RTX 3080 Ti for laptops looks particularly interesting. 

Nvidia's GA102 and RTX 3080 Ti Coming to Laptops

It's been rumored that Nvidia plans to introduce new Ti-branded GeForce RTX GPUs for desktops with more VRAM. Now it looks as though there will also be some new GeForce RTX 30-series Ti products for laptops in the near future. According to the press release, what's even more exciting is that a new GeForce RTX 3080 Ti will be coming to laptops.

The notes state, "GPU details for nVIDIA GeForce RTX 2060 12GB and GeForce RTX 3080 Ti Laptop." What they don't say is whether that GPU will be based on Nvidia's GA104 GPU (with up to 6144 CUDA cores) or if it will use GA102 GPU (with up to 10752 CUDA cores). However, Nvidia already has the GA104-powered GeForce RTX 3080 for laptops with 6144 CUDA cores, so using the same GPU with higher clocks or more VRAM may not make a lot of sense. What seems more likely is that Nvidia will use the considerably larger and more power hungry GA102 in a configuration appropriate for high-end gaming notebooks.

At present, Nvidia's GA102 is used exclusively for desktop graphics cards: the GeForce RTX 3090, GeForce RTX 3080 Ti, and GeForce RTX 3080. While those all require 320W or more power, Nvidia could drop the clocks and maybe even switch to GDDR6 memory (instead of GDDR6X) to fit the chip into a mobile profile. The biggest notebooks top out at around 150W of power for the GPU, and such a configuration should be possible. It would also unlock performance levels previously not available for mobile PCs. Of course that would impact battery life, but for desktop replacement notebooks that shouldn't matter too much.

Nvidia yet has to confirm its plans to launch the GeForce RTX 3080 Ti for laptops, so take this information with a grain of salt. However, AIDA64 isn't normally prone to leaking false information, so we strongly suspect we'll see the GPU in notebooks in the coming months.

A new Alder Lake benchmark has been posted with a for sale listing, potentially showing Alder Lake running in its new Gear 4 mode. According to @harukaze5719, this latest leak comprised a Core i5-12600K 10-core CPU running the AIDA64 cache & memory test, with a kit of DDR5 memory operating at DDR5-6400. Memory bandwidth was impressive at around 88-90GBps. However, memory latency took a huge hit, running at a sluggish 92.5ns. 

According to the listing, the system ran with Gear 4 enabled. In the last decade, Intel hasn't released a single architecture that approached 92.5ns of memory latency. Only AMD's original Zen architecture was capable of such laggy behavior, and it was known for its memory bottlenecks in that first-generation iteration.

Intel first introduced gear modes with its current Rocket Lake architecture as a way to improve memory support, but at the cost of latency. Rocket Lake offers up two modes:  Gear 1 and Gear 2. Gear 1 is the default mode that will run the memory controller and RAM at the same frequency, resulting in very low system latency. Gear 2 on the other hand reduces the memory clock to run at half the speed of the memory controller clock, or a 2:1 ratio. You sacrifice some memory latency, but you reach higher frequencies, and thus higher bandwidth, as a result. 

When we tested both gears in our Rocket Lake review, we noticed Gear 2 took a noticeable toll on system latency, directly impacting gaming performance. We tested a Core i5-11600K with Gear 1 mode featuring DDR4-2933 RAM, and DDR4-3200 RAM for Gear 2. Despite the faster memory speeds on Gear 2, we found that Gear 1 was consistently faster, with 5% better gaming performance. The memory latency results showed a similar trend, with 59.3ns for Gear 2 mode and 52.3ns for Gear 1. The latency results would be even more pronounced if both configurations ran at the same RAM frequency.

With Gear 4 mode, Intel takes things to a whole new level by cutting the memory controller frequency to just a quarter of the memory frequency. While this does allow the chip to support incredibly fast memory, it should still result in large latency losses.

So if the leaked latency results are true, with Gear 4 nearly doubling the amount of memory latency of Gear 1, then that setting will probably be quite useless for PC gaming and for other consumer-based applications. Generally, system latency is far more important than memory bandwidth, with only a couple of memory-hungry apps actually breaking that rule. So we'll have to see how these changes materialize once Alder Lake launches later this year.

The biggest unknown to all this is DDR5 memory, and what the highest memory frequency will be for the most expensive DDR5 kits. We've already seen DDR5 kits that can hit 12600MT/s, so Gear 4 mode could be advantageous at these speeds and beyond. But for now, it's only a theory. We'll have to wait for reviews (or at least a few more leaks) to see how the new Alder Lake architecture handles DDR5, and if higher gear modes will actually be useful.

Liquid cooling manufacturer Barrowch has released a new all-in-one water cooling kit that features the most futuristic looking CPU block you can buy. The block features a 2.9-inch,1440 x 1440p display with HDMI connectivity.

The screen resolution on Barrowch's waterblock is really impressive, and overkill if we're being honest. With over 700 pixels per inch, this little waterblock has nearly 2x the amount of PPI as a Samsung Galaxy S21 smartphone, giving you as much pixel density as you could ever ask for. The screen itself is a standard LCD display that uses a 5V micro USB interface for power.

The screen's primary use case is for monitoring your computer's sensors, ranging from CPU temperatures to GPU temperatures, system load and more. The screen can also be used in conjunction with AIDA64's monitoring tools to allow for more accurate readings. With the HDMI input, you could also use it as a very tiny computer monitor.

The all-in-one water cooling kit appears quite impressive even without the screen. The block is hooked up to a thick 240mm radiator with dual RGB 120mm fans. Barrowch was able to install a 17W pump and reservoir directly onto the radiator, including an acrylic windows next to the reservoir allowing you to check fluid levels

Because this AIO is a kit rather than a sealed unit, you can add additional radiators and blocks to the cooling system. Turning your 'simple' AIO unit into a full custom loop.

CPU support includes AMD's AM4 CPU socket, Intel's LGA 115x and LGA 2011 socket for HEDT compatibility.

Pricing and availability hasn't yet been revealed.

Asus often offers very complete software packages with its motherboards. This time around, the vendor is including two very helpful utilities for enthusiasts in its ROG Z590 series of motherboards.

Whether you're a casual user or an enthusiast, you should know about MemTest86. It's a popular piece of software that helps diagnose defective memory modules. The program has also turned into a tool for testing memory overclocks.

As spotted via Twitter user momomo_us, it would appear that Asus has struck a deal with PassMark Software to integrate MemTest86 directly into the motherboard's UEFI BIOS, helping users ensure their best RAM is actually running at its best. 

At the time of writing, Asus' ROG Z590 motherboard product pages aren't complete. The marketing refers to the boards using the standard, free version of MemTest86, but the screenshot shows the Pro version in operation, which commands a $44 price tag.

Asus is also bundling a 60-day subscription of FinalWire's AIDA64 Extreme with its ROG Z590 motherboards, as well as B560 ones. AIDA64 is a useful program for monitoring your system's vitals, auditing its performance and diagnosing potential problems.

FinalWire offers a free 30-day grace period for AIDA64 Extreme but with limited features. Asus' subscription unlocks the program's full functionality for 60 days. It would have been awesome if Asus had deliver a full one-year license, but then again, that would be asking for too much. AIDA64 Extreme retails for $39.95 for three PCs. 

Intel's 11th Generation Rocket Lake-S processors haven't landed on the market yet, so the best motherboards for the platform may still be to come. Newegg, however, currently has a few Z590 motherboards for sale and pre-order. That includes Asus' ROG Maximus XIII Hero and ROG STrix Z590-E Gaming WiFi that are selling for $499.99 and $379.99, respectively.

FinalWire has released the latest beta version (via Komachi_Ensaka) of its popular AIDA64 software, which confirms a ton of upcoming Nvidia GeForce RTX 30-series (Ampere) graphics cards. The newly added support insinuates that the graphics cards could be right around the bend.

The changelog didn't add any additional information to what we already expect from the GeForce RTX 3080 Ti. It does, however, confirm that the looming graphics card would utilize the GA102 silicon. The GeForce RTX 3080 Ti squeezes in between the GeForce RTX 3080 and RTX 3090 to contend with AMD's recently-released Radeon RX 6900 XT.

While Nvidia has launched the GeForce RTX 3060 Ti, the non-Ti variant is still behind closed doors. The GeForce RTX 3060 Ti shares the GA104 silicon with the GeForce RTX 3070. If FinalWire's information is accurate, the GeForce RTX 3060 will debut with brand new silicon, the GA106. There have been whispers that the GeForce RTX 3060 could possibly arrive with 3,840 CUDA cores.

We haven't heard much about the GeForce RTX 3050 up to this point, except that the graphics card might feature up to 2,304 CUDA cores. Apparently, Nvidia has a lot of different Ampere silicon stored away as FinalWire listed the GeForce RTX 3050 with the GA107 silicon, which hasn't been used in any Ampere graphics cards as of yet.

It turns out that Nvidia may break away from using the same silicon on both its desktop and mobile variants as the chipmaker has done in the past. The change is reflected in AIDA64's changelog as the GeForce RTX 3070 and RTX 3060 Mobile seemingly employ the GA104M and GA106M silicons, respectively. 

It's just conjecture, but the "M" suffix likely means mobile to differentiate the mobile silicon from the desktop equivalent. It remains an open question whether Nvidia will retain the core specifications on the mobile SKUs instead of the desktop versions.

The last tidbit of juicy details is related to the GeForce RTX 3070 Mobile, which popped up earlier today in a ray tracing benchmark. If the specifications are true, Nvidia could offer a 16GB version of the GeForce RTX 3070 Mobile. That's double the amount of memory that's on the GeForce RTX 3070. The idea certainly sparks our curiosity on whether we could potentially see a desktop version with 16GB as well. Even before Ampere's launch, there were already rumors of a GeForce RTX 3080 with 20GB of memory.

If you believe in hearsay, the GeForce RTX 3080 Ti, RTX 3060, and RTX 3050 are rumored to launch in January 2021. The mobile GeForce RTX 30-series' fate is still unknown, but there have been retailer listings of Ampere-powered gaming laptops, so mobile Ampere might not be far off either.      

Introduction and Test System

We hope you learned something interesting from How to Stress-Test Graphics Cards (Like We Do). In that piece, we introduced monitoring software, real-world games, synthetic benchmarks and artificially intense workloads that the Tom's Hardware team uses in our laboratories around the world to enhance coverage of graphics cards.

Now it's time to explore CPU and platform-oriented testing, including the apps needed to evaluate stability and CPU cooling, much like the ones we use in our CPU Benchmark Hierarchy. Setting CPU frequencies or voltages too high are a few of the variables that negatively affect an overclock, both of which can cause temperatures to rise. So, how do you stress test your own CPU thoroughly (yet safely), and which utilities should you use?

Important Warning about All Stress Tests

Before we begin, we have to warn you about the inherent dangers of running stress tests. The software that we’re presenting might not just produce a full load, but also potentially push beyond manufacturer-defined power limits. Using these so-called "power viruses" can result in damage to the system, especially if they're used for extended periods of time. Those who want to use these applications assume all responsibility for the outcome. Be sure to continuously monitor relevant parameters, including temperatures, with accurate and up-to-date utilities. At least you'll be able to abort your testing immediately if it becomes necessary.

Choosing the Right Monitoring Application

HWiNFO Monitoring Utility

How to Stress-Test Graphics Cards (Like We Do) covered a lot of the software we use for monitoring graphics card health. Keeping an eye on your entire platform's vitals may require a different approach, though. After all, there are a lot more sensors to watch.

HWiNFO64 is a good example of a tool that can do all of this. It can read practically any sensor output in real time, and write it to a log file, too. Download HWiNFO64 here.

To reduce any lag with the measurements, we recommend that you exclude unnecessary sensor readings (network, system, drives, etc.), from the monitoring altogether.

Choosing the Right CPU

We’re not using the familiar X299- and X99-based systems this time around. Instead, we're testing with an Intel Core i7-8700 and Z370 motherboard with 16GB of DDR4-3200 memory. This configuration represents high-end hardware fairly well, including the fact that gaming machines tend to lack some of the monitoring capabilities available on workstations.

However, our Alphacool Eiszeit 2000 Chiller generates a water temperature of exactly 20°C for us, as usual. This is what makes a direct comparison between different stress-testing applications possible.

The same Asus ROG Strix Radeon RX 560 from our previous stress testing story makes another appearance today. That means our results are comparable. Faster graphics cards don’t change our conclusions in any meaningful way; they just increase power consumption.

Test System and Methodology

We introduced our new test system and methodology in How We Test Graphics Cards. If you'd like more detail about our general approach, check that piece out. We've adjusted the CPU and the cooling system to better suit this article.

The hardware used in our lab includes:

MORE: Best CPUs For Gaming 

MORE: CPU Benchmarks Hierarchy

MORE: All CPUs Content

CPU Only: Prime95 With AVX Or SSE

Prime95 is to CPUs what FurMark is to GPUs: a true classic that continues receiving updates. Current versions support the AVX instruction set, which helps generate massive thermal loads. Not everyone finds it necessary to go all-out with temperatures, though. For those folks, the BIOS on motherboards offers AVX offset adjustments to reduce clock rates as AVX instructions make their way through the execution pipeline. Alternatively, you can download an older version of Prime95, before it was optimized for AVX support, or you can use newer versions that allow you to disable the instruction set.

Before we show you how to switch between instruction set extensions, you'll want to download Prime95 here.

Selecting the Instruction Set Extension

By default, Prime95 automatically selects the newest instruction set extension, such as AVX, AVX2, or even AVX-512. In order to change this behavior, Prime95 needs to be started and completely shut down again once. This will create the local.txt file. In it, exclusions are assigned a value of 0, whereas the code path that’s to be tested is assigned a value of 1. If you aren’t clear as to which SSE version is supported by your CPU, both can be set to 1. Prime95 will choose the correct fallback.

    CpuSupportsRDTSC=0 or 1    CpuSupportsCMOV=0 or 1    CpuSupportsPrefetch=0 or 1    CpuSupportsSSE=0 or 1    CpuSupportsSSE2=0 or 1    CpuSupports3DNow=0 or 1    CpuSupportsAVX=0 or 1    CpuSupportsFMA3=0 or 1    CpuSupportsFMA4=0 or 1    CpuSupportsAVX2=0 or 1    CpuSupportsAVX512F=0 or 1

Prime95 With AVX & Small FFTs

We start at the top and let Prime95 automatically activate AVX by skipping manual entries. The Small FFTs setting maximizes temperatures, even though memory isn’t taxed as much. Our results show this well:

Prime95 With AVX & Blend

Using the more conservative Blend test results in less load on the cores and more load on the memory. This is a viable alternative for pure stability testing meant to detect possible computational errors due to a potentially unstable overclock, especially since it’s suitable for prolonged test runs (providing your cooling solution is up to the task).

Prime95 With SSE & Small FFTs

Once AVX is manually disabled, Prime95 becomes a much better measure of stability. Most real-world applications don’t use AVX, after all. A longer test to detect possible overclocking-related instability makes more sense this way, even if we think you should still run a round of benchmarks using real-world workloads.

Prime95 With SSE & Blend

As before, using Blend instead of Small FFTs increases the memory load while easing up on the execution cores somewhat. Knowing when to pick this combination of settings is difficult, since it's perhaps the least-taxing of all. We consider it well-suited to systems without the best cooling, such as laptops and mini PCs.

MORE: Best CPUs For Gaming 

MORE: Intel and AMD Processor Benchmark Hierarchy Comparisons

MORE: All CPUs Content

CPU Only: OCCT With Four Options

OCCT offers a fairly challenging CPU stress test, which often gets overlooked. It might not generate the massive amounts of heat we measure under Prime95 with AVX, but it’s still a great choice for long stability tests. The software can log a number of important parameters, running in an endless loop or stopping after a defined length of time. Download OCCT here.

OCCT With Small Data Set

This option generates the highest load of the four available options, though it's still less taxing than Prime95 with AVX and Small FFTs. We can say, however, that it is more stressful on memory than Prime95’s.

OCCT With Big Data Set

The overall load drops yet again, making this test suitable for laptops that shouldn’t be pushed to their thermal limits.

Memory, specifically, is pushed harder than anything we saw under Prime95. In fact, this is the second-most-taxing memory workload in our suite.

OCCT With Linpack & AVX

OCCT’s AVX implementation generates a demanding overall load, though it doesn’t reach the numbers generated by this utility's own Small Data Set option. Load on the memory is still fairly high, too.

OCCT With Linpack & SSE

OCCT’s SSE implementation generates a low overall load, similar to the Big Data Set option. But while it doesn't set any new records, this one's a good alternative for lower-end PCs with stock coolers. It’s also a good option for prolonged testing.

MORE: Best CPUs For Gaming 

MORE: Intel and AMD Processor Benchmark Hierarchy Comparisons

MORE: All CPUs Content

CPU Only: AIDA64 With CPU, FPU, Cache, Or Memory

AIDA64’s CPU stress test offers several different configurable options. It offers a true memory stress test, a demanding cache load, and a taxing execution core workload that doesn't utilize memory extensively. In other words, everything, nothing, or anything in between can be tested. The free version has a time limit, so potential cost is the only real drawback we can think of. Download AIDA64's 30-day trial version here.

It's possible to log various sensor readings to disk, even as you monitor their measurements in real time. You can draw AIDA64's data to a curve, display instantaneous status on the Windows taskbar, or output its sensor information to a third-party application.

AIDA64 With CPU & FPU & Cache

If you thought that turning everything on would really push your hardware, then you were wrong. The individual tests might produce high numbers on their own, but combining them only gives us slightly above-average readings. Then again, that's fairly typical of what you'd see from a taxing real-world piece of software.

AIDA64 With Just CPU

All of these results are on the low end of what we've seen from other stress tests in our suite. In short, then, this may represent the maximum load generated by older apps, but certainly not the most demanding ones. Not surprisingly, we measure lower temperatures.

If you're looking to test an aging system by slowly increasing the load until hitting its limit, you have a good place to start here.

AIDA64 With Just FPU

The extreme FPU load results in very high package and CPU socket temperatures, yielding the highest core reading possible from our suite. Consequently, this test works well for determining the limits of potent cooling solutions. If throttling is ever going to be a problem for you, it should become apparent using AIDA64's FPU test.

AIDA64 With Just Cache

This test gets more interesting with increasing CPU cache size and surface area. System memory also endures a somewhat higher load. All together, AIDA64's cache workload is a good choice for long-term runs on overclocked systems with an emphasis on evaluating stability, as opposed to cooling performance.

AIDA64 With Just Memory

The software's memory test provides us with a more demanding load on system RAM than any other utility, as measured by power consumption and the warmest module's temperature. It’s a great choice for evaluating the stability of a memory overclock on its own, or as a supplemental test combined with other workloads.

MORE: Best CPUs For Gaming 

MORE: Intel and AMD Processor Benchmark Hierarchy Comparisons

MORE: All CPUs Content

CPU Only: powerMAX With AVX Or SSE & Heavy Load

CPUID powerMAX

This combined CPU and GPU stress test wasn’t our first choice when it came to evaluating graphics cards. However, it does do a better job loading down CPUs. Being able to switch easily between AVX and SSE code paths is also a definite plus. Download powerMAX here.

powerMAX With AVX

powerMAX tops out a bit below Prime95 with AVX, but it nearly matches the popular stress test's results. The loads powerMAX generates are well above average, making it a suitable tool for testing CPU coolers. 

Conversely, you won't want to use it for stability-testing overclocked PCs operating close to their limits unless the goal is to identify temperature issues.

powerMAX With SSE

On our test platform, utilizing the SSE code path resulted in temperatures that averaged ~10°C lower than before. System RAM was the exception; it actually heated up more than when we sent AVX instructions down the CPU's execution pipeline.

Heavy Load

Heavy Load works if you're trying to generate a realistic CPU load, and by realistic we mean the types of loads you'd encounter in a game applying average demands to your components. Running Heavy Load won’t tell you anything about your CPU cooler's limits. But it is a decent indicator of how a CPU might fare during everyday operation. It’s also safe to use on older or mobile hardware without worrying about breaking anything. In that sense, it's not really a conventional stress test. Download Heavy Load here.

What the CPU test actually does remains a developer secret. It doesn't seem to explicitly use modern instruction set extensions, though.

MORE: Best CPUs For Gaming 

MORE: Intel and AMD Processor Benchmark Hierarchy Comparisons

MORE: All CPUs Content

CPU Only: Measuring Power Consumption With Different Stress Tests

Regardless of whether you're talking about CPUs or graphics cards, the general principle remains the same: however much energy is put in reemerges as waste heat.

To that end, we start by observing the CPU's power consumption via its package. Differences between various architectures and models can be massive. But our Core i7-8700 does a fairly good job of representing the upper-mid-range, given its 65W TDP.

Prime95 with AVX and Small FFTs takes first place in our comparison of stress tests, followed by powerMAX with AVX and AIDA64’s FPU test.

The top three finishers land in their same positions, though the rest of the field shuffles around a bit. This is due to variances in certain loads, such as the memory tests. Moreover, there are slightly different voltage converter losses attributable to dissimilar load patterns.

We decided to present the deltas between overall power consumption and individual results as well, since some of them are pretty massive. This comparison allows for an estimate of how much additional waste heat is generated by the system that the CPU cooler can't take care of. Keep these numbers in mind if you're building a water-cooled system; you'll need proper ventilation to cope with the extra heat.

MORE: Best CPUs For Gaming 

MORE: Intel and AMD Processor Benchmark Hierarchy Comparisons

MORE: All CPUs Content

CPU Only: Measuring Temperatures With Different Stress Tests

Next up: temperatures. These are directly related to the waste heat dissipated during our stress tests. Various workloads push different parts of the CPU in unique ways, yielding more interesting findings than we were expecting.

Core temperatures are measured by the CPU’s integrated digital thermal sensor (DTS), which reacts to actual load levels on an almost real-time basis. Between this and our chiller-based cooling solution, which maintains its fluid at a near-constant 20°C, we get very accurate load measurements. Then, we calculate the average temperatures across all cores throughout our test run.

Package temperature, or, more precisely, the reading provided by the platform environment control interface (PECI), is typically similar to the average core temperature.

Once again, we calculate the average across the entire 30 to 45 minutes it takes for the temperature to stabilize. While the results usually come close to the respective core temperatures, they do exceed them in certain circumstances, depending on the type of test.

Next, we examine data from an additional sensor located directly underneath the socket. Since the CPU cooler does its job from the top, through a heat spreader, this second sensor's more sluggish result provides important information about stability and cooling performance. Using Prime95, the CPU hits a whopping 105°C without any overclocking, and in spite of the chiller!

This result demonstrates why you want to use Prime95 with AVX as a stability test for an extended period of time only if you're using monitoring software that can track all relevant temperatures.

In the end, knowing that the CPU (DTS) is cooled well doesn't really matter if the package underneath is burning up.

The memory load varies quite a bit between tests, as well. Here are all of the measurement results in one bar graph for direct comparison:

MORE: Best CPUs For Gaming 

MORE: Intel and AMD Processor Benchmark Hierarchy Comparisons

MORE: All CPUs Content

System: Prime95 & GPU

For our combined stress tests, we use Prime95 (with AVX or SSE, as well as Small FFTs) with MSI Kombustor and FurMark, our suite's two most challenging graphics workloads. For these combinations to produce their full load, Prime95 needs to be started first and set to low priority via Windows' Task Manager. All threads that are started during the stress test automatically inherit this priority.

MSI Kombustor and FurMark generate a processor load of just 1% to 5%. However, Prime95 at its default settings would still slow these applications down so much that their GPU loads would decrease significantly. We also start both GPU stress tests in windowed mode in order to not push Prime95 into the background too much, while making sure that the GPU workload's window is active. With all of this taken care of, we know our hardware is truly maxed out.

The warnings on page 1 apply again here, but even more urgently since we're being extra hard on our components. Be diligent in monitoring all relevant health parameters: Any misstep has destructive potential.

Prime95 With AVX & Small FFTs + MSI Kombustor GPU Core Burner

First, we combine the CPU and GPU stress tests that generated the highest loads during our previous measurements. It’s not surprising that this gets us the second-highest result for system stress testing. The highest result, achieved by using FurMark, is so close that we deem it within a margin of measurement error.

The loads that this combination generates would never be applied by a real-world application. In other words, they aren't realistic in any way. However, enthusiasts interested in examining the limits of their power supplies or cooling subsystems can do so with these two pieces of software.

Prime95 With SSE & Small FFTs + MSI Kombustor GPU Core Burner

The SSE code path is less intensive, making it safer for the CPU and other components involved in the stress test. However, it still generates enough load to push the system hard.

Prime95 With AVX & Small FFTs + FurMark

Once again, we generate an extremely high load. This time, though, it's the greatest level of stress we measured. Those using this combination should heed our warnings about possible system damage, though ideally your worst-case situation would be an automatic shutdown. Most motherboards don’t include any thermal protection for their voltage converters. Rather, these are reserved for high-end platforms (and they aren’t a given there, either).

MORE: Best CPUs For Gaming 

MORE: Intel and AMD Processor Benchmark Hierarchy Comparisons

MORE: All CPUs Content

System: MSI Kombustor & OCCT

MSI Kombustor Does It All

MSI’s Kombustor comes with a CPU stress test, but it doesn't use AVX instructions. As a result, Kombustor only generates about 110W, which is similar to the loads generated by demanding games. That's simply not good enough for a pure CPU test, so we're only trying it out in conjunction with Kombustor’s GPU stress test.

We like that the CPU and GPU test threads run simultaneously through the same application, which automatically gives them the same priority. Nevertheless, continue running the GPU stress test in windowed mode, unless you have a multi-GPU configuration.

MSI Kombustor GPU Core Burner

We start by generating the highest possible load for the GPU and running the CPU stress test concurrently. The results end up exactly as expected based on what we know about Kombustor.

MSI Kombustor Memory Burner

MSI’s Kombustor with Memory Burner shifts the focus towards extensive graphics memory utilization, which is why we recommend it for enthusiasts looking to see if their system can handle the power consumption and heat generated by today's most demanding games.

OCCT Power Supply

OCCT Power Supply generates CPU and GPU loads capable of taxing your PSU. At least that's the idea.

Tests like this one, which generate game-like power-consumption numbers, are used to optimize cooling solutions and fan curves. If everything in your PC works well up through this point, then OCCT Power Supply can be used to help minimize noise levels by guiding you in the direction of optimized ventilation. Any task that pushes a PC harder than this can be dealt with by setting fan curves that get more aggressive under higher loads. Establishing the system’s absolute limit comes first, after which you can fine-tune performance with less-demanding tests like this one.

MORE: Best CPUs For Gaming 

MORE: Intel and AMD Processor Benchmark Hierarchy Comparisons

MORE: All CPUs Content

System: AIDA64, powerMAX & Heavy Load

AIDA64 With FPU & GPU

The problem with AIDA64’s combined CPU and GPU stress test is that, even though it effectively isolates the performance of specific subsystems, the ratio of CPU and GPU load just isn’t right. This makes the test unrealistic. Adding up all of the individual loads to achieve a generally taxing scenario might provide some useful information, but we've already discussed better alternatives for the same type of task.

Even if the Stress CPU, Stress System Memory, and Stress GPU(s) settings are all activated at once, AIDA64’s efforts still fail to yield a realistic workload.

powerMAX With AVX Or SSE & GPU

powerMAX wasn’t one of our top contenders when it came to GPU stress testing. Can it redeem itself through the addition of an integrated CPU stress test? We run both versions again in order to find out what powerMAX does well. The GPU workload remains windowed in order to avoid pushing the CPU stress test too far into the background. We also ensure the GPU stress test window is active.

powerMAX With AVX & GPU

powerMAX has the same problems as AIDA64: Its CPU load is way too high, while the graphics load is a little too low. This test is simple and convenient to run, but it just doesn’t provide balance. It’s not suitable for pushing your system to its limits, and it doesn’t simulate normal operation, either.

powerMAX With SSE & GPU

Using the SSE code path shifts load from the CPU to the GPU. But that's not enough to simulate a realistic gaming scenario, and it’s not enough to push the system to its limits. Ultimately, our conclusions about this utility's AVX code path apply here as well.

Heavy Load With CPU & GPU

Heavy Load didn’t exactly inspire confidence in us when we tried its CPU and GPU tests on their own. As it turns out, combining two bad tests gets you another bad test. In fact, it’s completely unusable as either a stress or stability test, unless you want to be duped into believing an over-enthusiastic overclock will run stably under truly taxing loads.

MORE: Best CPUs For Gaming 

MORE: Intel and AMD Processor Benchmark Hierarchy Comparisons

MORE: All CPUs Content

System: Measuring Power Consumption With Different Stress Tests

Our power-consumption results tell us how much waste heat is generated by the different combinations of stress tests. For extreme cooling and power supply testing, there’s really only one option: Prime95 with AVX. There are two viable choices for simulating realistic gaming scenarios, and those are MSI Kombustor’s combined CPU and GPU test, and OCCT with its Power Supply test.

The same goes for graphics card stress testing. Prime95 with AVX, MSI Kombustor, and OCCT give you everything you need.

Looking at total system power consumption, we see that the top results are almost 100W above the numbers we'd expect from normal gaming. The difference is primarily due to the fact that Intel's Core i7-8700 draws a lot of power during AVX-based stress tests.

A look at how much of the remaining energy is converted into heat by the system paints the rest of our picture. The voltage-converter losses, in particular, increase quite a bit. Don't just focus on CPU or GPU cooling; also make sure the motherboard's VRMs (voltage regulator modules) receive plenty of airflow when you run punishing stress tests. These components can trigger throttling if they overheat. 

MORE: Best CPUs For Gaming 

MORE: Intel and AMD Processor Benchmark Hierarchy Comparisons

MORE: All CPUs Content

System: Measuring Temperature With Different Stress Tests

There’s a strong correlation between GPU temperatures and the graphics card's power consumption. A manufacturer temperature limit of 64°C means that results between 62 to 64°C represent a full load on our Radeon RX 560 card.

Both extreme and gaming-oriented loads should be tested if you're gunning for a stable and quiet platform.

Our package temperature results look a lot like the core temperatures we recorded above.

Once again, Prime95 hits the package hardest.

MORE: Best CPUs For Gaming 

MORE: Intel and AMD Processor Benchmark Hierarchy Comparisons

MORE: All CPUs Content

Conclusion: The Best Stress-Test Tools & Settings

Testing Maximum CPU Load

If your goal is to maximize power consumption and generate as much waste heat as possible to ensure your thermal solution is sufficient, then power viruses like Prime95 with AVX and Small FFTs, powerMAX, or AIDA64’s Stress FPU option are the way to go. However, these stress tests aren’t representative of the loads from commonly used applications. That's because most desktop software doesn't use AVX instructions the way artificially demanding workloads do. To simulate everyday scenarios, stress tests that tax your hardware using SSE code paths are the way to go.

Combine a tool like that with MSI Kombustor’s GPU Core Burner or Memory Burner for an excellent worst-case platform scenario. Even a game like The Witcher 3 can help you create similar or worse power consumption numbers than a lot of popular graphics stress tests.

Optimizing CPU Cooling

Your cooler should be set up so that it’s able to deal with these extreme loads, regardless of the code path responsible for them. And don't forget chassis ventilation. Many enthusiasts build great-looking custom water-cooling loops, but then forget that the motherboard needs active cooling in certain areas, else it'll overheat as well. Without at least some airflow, the board's VRMs might not last long.

Optimizing the Noise Level

Fortunately, airflow through your chassis can be temperature-controlled. Once the maximum temperature is known, a good combination of CPU and GPU stress testing can be used to simulate demanding gaming loads. MSI Kombustor and OCCT are particularly well-suited for this task if they’re configured as we described. The results of those tests may then be used to form the basis for fan-curve optimization by reserving the most aggressive fan speeds for the range approaching your ceiling.

Stability Testing & Its Limits

None of the pure stress tests guarantees 100% stable everyday operation. All they’re really good for is finding the system’s thermal and electrical limits. Everything else, especially when it comes to overclocked components, can be truly found out only by using suitable applications run over a long period of time. Prime95 and its contemporaries aren’t really designed to find errors, because they are fairly simple computational tools that serve one purpose: to consume a lot of power. (At least that's the case for Prime95's torture-test mode.)

Conclusion

With some thought and the right applications, most of which are free, you're able to test for cooling capacity and operational safety. These tests can (and should) be repeated periodically in order to suss out any changes due to aging components. In the beginning, testing every six months is sufficient, with the intervals between tests decreasing as the PC ages, especially if it’s used a lot or very intensively. It doesn’t matter if it’s been overclocked or not. Better safe than sorry!

MORE: Best CPUs For Gaming 

MORE: Intel and AMD Processor Benchmark Hierarchy Comparisons

MORE: All CPUs Content

When it comes to memory records, we're most used to seeing manufacturers or overclockers showing off their frequency peaks. However, another, perhaps a more practical reading are the read/write speed tests, of which Team Group has just set a new world record.

Team Group's record is based on a quad-channel DDR4 setup using the company's T-Force Xtreem ARGB DDR4 gaming memory, which Team Group introduced at the end of last year. The setup jotted down a write speed of a whopping 104,643 MB/s, and read that data back at a staggering 136,678 MB/s. The record was set by Taiwanese overclocker LeeGH.

The memory used for this record are the T-Force Xtreem ARGB 3200 MHz CL14 modules, but they were subjected to lofty overclocks. LeeGH pumped the units up to 4300 MHz while maintaining the short CL14 timings. This was accomplished using an Intel Core i9-9940X processor installed in an Asus ROG Rampage VI Extreme Encore motherboard.

Mind you, this is the quad-channel memory read speed record. If you look at the HWBot DDR4 AIDA64 memory read/write world records hall of fame, you'll find four other scores that surpass Team Group's record. Those scores are based on hexa-channel DDR4 setups, though, giving their record-setters a clear advantage. 

Nevertheless, for Team Group and LeeGH to reach 5th place overall on the DDR4 AIDA64 records using a quad-channel setup is nothing short of impressive.

Nvidia has outdone itself this time, even by the chipmaker's own standards. AIDA64 (via @momomo_us) has just added support for two new unreleased variants of the GeForce GTX 1650 graphics card, increasing the number of different versions to four.

The original GeForce GTX 1650, which landed last year, debuted with the TU117 silicon. The graphics card's 896 CUDA cores were complemented by 4GB of 8 Gbps GDDR5 memory across a 128-bit memory interface. 

A little less than a year later, Nvidia rolled out the GeForce GTX 1650 GDDR6, and, as the name implies, the graphics card came with faster 12Gbps GDDR6 memory. Although the GDDR6 variant features lower clock speeds, the swap to GDDR6 memory allowed the graphics card to deliver up to 50% higher memory bandwidth.  The upgrade contributed to a real-world performance improvement of up to 14% over the vanilla version. 

AIDA64's latest changelog shows that Nvidia isn't quite done with the GeForce GTX 1650 yet.

Nvidia GeForce GTX 1650 Specifications

*Specifications are unconfirmed.

AIDA64 refers to the two new variants as the GeForce GTX 1650 TU116 and GeForce GTX 1650 TU106. It's pretty much the same formula as the GDDR6 version that denotes the type of memory in the model name. On this occasion, the suffixes designate the silicon that's used inside each refreshed model.

In the case of the GeForce GTX 1650 TU116, the TU116 die is the same one that Nvidia employs in the the other GeForce GTX 16-series SKUs, including the GTX 1650 Super, GTX 1660, GTX 1660 Super and GTX 1660 Ti.

The TU106 die, on the other hand, is a very interesting selection, and the one that raises many questions. For context, the TU106 is present inside Nvidia's GeForce RTX 20-series models, such as the GeForce RTX 2060, RTX 2060 Super and RTX 2070. Unlike the TU116 and TU117 silicons, the TU106 does house Nvidia's Tensor and RT cores. Has Nvidia finally caved in and decided to bring ray tracing to a GTX 16-series graphics card? Unfortunately, we don't know for the time being. AIDA64 didn't give any hints on the specifications. Nonetheless, the base specifications should be the same for both new variants.

The new versions of the GeForce GTX 1650 probably aren't a coincidence. Nvidia is likely just reprocessing defective dies that don't meet the requirements for higher tier models and using them in the GeForce GTX 1650. Alternatively, the chipmaker might just be getting rid of leftover Turing silicon to make space for Ampere, which is heavily rumored to debut in September. In any event, it's easy to see why Nvidia chose the GeForce GTX 1650 in particular. 

According to the Steam Hardware Survey for May 2020, the GeForce GTX 1650 was the fifth most popular graphics cards among Steam gamers. If there was a fast way to dispose of unused silicon, the GeForce GTX 1650 would be the best candidate.

AIDA64, a popular system information and monitoring utility, has added support for AMD's upcoming Renoir APUs and preliminary support for fourth-generation Ryzen desktop CPUs. This may be a hint that early samples of both silicon are already out in the wild.

AMD has done a remarkable job of shifting the majority of its products on the new 7nm process node. Its APUs are currently the only parts still stuck on the previous 12nm process. However, Renoir should change that. AMD's next generation of APUs are likely to be based on the Zen 2 microarchitecture and 7nm process, just like the chipmaker's current processor lineup. As APUs, the chips will come with integrated graphics, and it seems AMD's Renoir APUs will cling to the Vega microachitecture.

AMD just launched its Ryzen 3000-series CPUs a few months ago, which are based on Zen 2 and 7nm manufacturing process. Its next generation of chips are expected to be based off a Zen 3 microarchitecture and an improved 7nm+ process. We're also expecting the fourth-generation Ryzen desktop processors to carry Ryzen 4000-series branding. Vermeer is the rumored codename for the upcoming chips.

Both Renoir and Vermeer are rumored to touch ground next year. In the last AMD earnings call, AMD CEO Dr. Lisa Su confirmed that new 7nm mobile chips will debut in early 2020. There's a high probability that AMD might announce Renoir at CES 2020 in January. As for Vermeer, we expect to see those chips later next year, probably around summer time.

AIDA64, a widely used system information, diagnostics, and benchmarking tool, has added information for a mysterious, unannounced Nvidia GeForce RTX T10-8 graphics card, which is apparently based on the TU102 die.

Currently, there are four Nvidia graphics card that use the Turing TU102 silicon. The GeForce Titan RTX and GeForce RTX 2080 Ti hail from Nvidia's mainstream product line while the Quadro RTX 8000 and Quadro RTX 6000 belong to the enterprise side. Thanks to AIDA64's latest changelog, we're almost certain that Nvidia is working on another TU102-based graphics card behind the scenes.

The GeForce RTX moniker implies that the unknown graphics card is most likely aimed towards the gaming market. That's practically the only clue we have at the moment. So, it could be something like a GeForce RTX 2080 Ti Super or GeForce Titan RTX Black. However, we're more inclined to the first since the GeForce Titan RTX already employs a maxed-out TU102 die. If Nvidia was to make a Black version, the chipmaker can either give it faster operating clocks or more memory. The latter wouldn't be a good decision since it'll probably cannibalize the sales of the Quadro RTX 6000. In any event, Nvidia would have a very hard time selling another Titan card.

From a performance standpoint, the GeForce Titan RTX and the GeForce RTX 2080 Ti are nearly in the same ballpark. Given the huge difference in price, there's definitely room for an in-between model. Hypothetically, Nvidia could pull another GeForce RTX 2080 Super on us. The chipmaker can grab the GeForce RTX 2080 Ti, enable a few more CUDA cores, slap some faster memory on it, and voilà – you have the GeForce RTX 2080 Ti Super.

During a secret E3 meeting outlining its Super cards, Nvidia stated that, at that time, the chipmaker had no short or long-term plans to introduce a GeForce RTX 2080 Ti Super. As you are well aware, plans are always changing. Now that AMD President and CEO Dr. Lisa Su has confirmed that high-end Navi, or commonly called Big Navi, are on the way, Nvidia could be gearing up new models to stand its ground.

FinalWire, the developer behind the system information and diagnostics tool AIDA64, released the latest beta version (V5.99.4983) over the weekend. It comes with preliminary support for AMD's upcoming trio of processor lines based on the Zen 2 microarchitecture. This could be a hint that we're closer to the next-gen CPUs' launch than we think.

Release notes:Fahrenheit option now applies to the Computer / Sensor pageStorage / SMART / disk health sub-window / Fahrenheit temperature unit supportComputer / Sensor / displaying primary and secondary NVMe SSD temperatures in a single lineimproved chipset information for AMD K17.3 IMC, K17.7 IMCIntel Processor Number detection for Xeon Platinum 9221, 9222physical CPU information for AMD Castle Peak, Romeupdated JEDEC memory module manufacturers databasefixed: physical CPU information for AMD Matissefixed: chipset information for AMD K17.5 IMC

For starters, Matisse is the codename for AMD's third-generation Ryzen desktop processors. The upcoming Ryzen 3000-series parts will be based on AMD's Zen 2 microarchitecture and built with TSMC's 7nm process node. There have been a lot of rumors and speculation surrounding the much-awaited chips, which are expected to challenge rival chipmaker Intel on all the processor tiers. AMD president and CEO Dr. Lisa Su is expected to discuss the aformentioned processors at Computex 2019, so AMD could spoils us with all the technical deets come May 27.

One thing's for sure, though, AMD certainly has a busy agenda ahead. A recent roadmap revealed the chipmaker plans to launch the third-generation Threadripper HEDT (high-end desktop) processors, codenamed Castle Peak, right after Matisse. As a matter of fact, AMD has been keeping a tight lip about the the forthcoming Threadripper 3000-series lineup. There are two possible outcomes here. AMD could either produce the multi-core monsters on the latest Zen 2 microarchitecture and TSMC's 7nm process node, or continue to employ the Zen+ microarchitecture and GlobalFoundries' 12nm process node like on the previous Threadripper 2000-series CPUs. The evolution from the first to second generation of Threadripper processors brought double the core count (from 16 to 32) for the flagship model. Let's cross our fingers that AMD surprises us once more.

On the enterprise side, AMD is diligently working on its next-generation EPYC, codenamed Rome, server processors. They will adhere to the Zen 2 microarchitecture and are co-produced with TSMC's 7nm process and some of its parts with GlobalFoundries' 14nm process. The new EPYC processors will feature up to 64 cores /128 threads and could potentially support up to 162 high-speed PCIe 4.0 lanes in a dual-socket setup. AMD has not yet shared a timeline for their release. 

Memory expert G.Skill continues to expand its "Trident Z RGB (for AMD)" series of memory kits, commonly known as the TZRX models, with the addition this week of a 32GB quad-channel kit that clocks in at 3,466MHz.

Previously, Ryzen Threadripper owners opting for a 32GB memory kit were limited to DDR4 speeds of 3,200MHz. With its new DDR4-3466 Trident Z RGB memory kit, G.Skill is once again raising the bar for high-speed memory on the AMD HEDT (high-end desktop) X399 platform. The Taiwanese memory manufacturer's latest offering consists of four DDR4 memory modules, each with 8GB capacity in size. The modules operate at a guaranteed frequency of 3,466MHz with an operating voltage of 1.35V. The memory kit's timings are configured to CL18-22-22-42. 

G.Skill validated the memory kit on Asus' latest ROG Zenith Extreme Alpha motherboard, along with AMD's Ryzen Threadripper 2950X 16-core processor. The memory kit passed over seven hours of MemTest and put in impressive AIDA64 benchmark results, such as 80,749 MB/s read, 98,747 MB/s write and 80,777 MB/s copy.

As always, G.Skill's Trident Z RGB memory kits come equipped with Intel XMP 2.0 profiles for an easy setup and are backed by a limited lifetime warranty.

The G.Skill DDR4-3466 Trident Z RGB for AMD memory kit will be available worldwide via G.Skill's distribution partners in the first quarter of this year. G.Skill hasn't revealed pricing yet.

Introduction to B450 and Product Description

AMD's Ryzen 2000 CPUs (like the Ryzen 7 2700X) might be old news by now, but AMD's new budget-focused B450 chipset comes out of the gate ready to inject more value into the AMD AM4 platform. Our success with the X470 Aorus Gaming 7 WIFI gives us high hopes for the lesser B450 Aorus Pro board, but Gigabyte does hold back on some features that could have made this board spectacular. Regardless, the B450 chipset is a sweet spot for builders and gamers who don’t need X470’s added feature set, and Gigabyte delivers a great product with the B450 Aorus Pro WIFI, at a surprisingly affordable $120 MSRP.

Specifications

Over the past year of testing Ryzen-based builds, we’ve had nothing but success with the company's boards, and their B450 takes its improvements from X470 and applies them to AMD's updated “mainstream” chipset.

AMD Socket AM4 Chipset Features

For those of you not keeping up with chipset changes, AMD’s increment from the 300 series to 400 series primarily includes improvements for board physical specifications to formally support XFR2 (an update of AMD's extended frequency range feature) and Precision Boost Override for what AMD calls its 2nd Generation Ryzen chips (Ryzen 2000 series). AMD provides specifications for layer counts and supported memory speeds, regulator design requirements, and other stipulations for these refreshed boards. Between X470 and B450, the same IO support dynamic that was found with last year's X370/B350 remains. In other words, B series boards support fewer USB3.0 Gen1, PCIe Gen3 x16 slots, and PCIe Gen2 lanes than their more expensive brothers. For more details, you can check the chart above.

On the surface, fewer slots and port support sounds like an instant “con,” but honestly the majority of builders don’t support for a second graphics card, larger SATA RAID configurations, or a plethora of 5Gbps USB. Also, with that same mindset, manufacturers don’t need to provide fully enabled lower-end boards that directly compete with their higher-tiered offerings. From there, it really is a thought experiment as to which chipset truly is “better,” and hopefully we’ll shine some light on that throughout the article.

Enter the Gigabyte B450 Aorus Pro WIFI. Gigabyte has traditionally kept its Aorus brand relegated to premium offerings, but for B450 they couldn’t resist unleashing the falcon that we have appreciate from previous reviews. Bringing a well-respected brand to a lower tier might appear to be a marketing gimmick, but it turns out this board is on-par with similar offerings from X470. From an aesthetics perspective, the B450 Aorus Pro WIFI looks shockingly similar to the X470 Aorus 5, with the omission of LED lit DIMM and PCIe slots. Comparing the B450 board against the X470 Aorus Ultra, the inclusion of Wi-Fi in the B450 Pro might be enough for some to negate the X470’s improved connectivity elsewhere.

The B450 Pro WIFI comes equipped with ample RGB lights, RGB power switches, and an addressable LED interface that helps mitigate these deficiencies for future upgrades for builders. For a budget board, we don’t expect the contents of the box to be substantial and Gigabyte doesn't overwhelm with accessories and extra parts. The company includes their badge sticker along with documentation, a couple SATA cables, and a Wi-Fi antenna for connecting the SMA plugs on the back panel.

The I/O panel reflects the B450 chipset’s decreased IO, with access to only four USB3.1 Gen1, two USB3.1 Gen2, HDMI 2.0, DVI-D, Gigabit Ethernet, five analog audio and one digital audio jack, and two Wi-Fi antenna ports. The integrated back panel is a nice aesthetic inclusion that's usually only found on pricier enthusiast-chipset boards. Builders with APUs like the Ryzen 5 2600 are primed for upgrading from older chipsets thanks to the video ports. And the lack of USB 2 means we aren’t left questioning which ports are high speed when reaching around behind the PC.

Audio coverage on the B450 Aorus Pro is provided through the Realtek ALC1220-VB codec boasting improved audio signal-to-noise ratios and impedance detection for use with different strength drivers on various grades of “head-worn audio devices.” As for wired networking, the tried and true Intel I211AT controller is used here and Gigabyte includes its cFosSpeed Internet Accelerator for prioritizing network traffic through the Ethernet port. We don’t have performance results with this tool, but it appears that more vendors are implementing features that mimic Rivet Network’s Killer-brand, which prioritizes packets for improved network performance. One benefit we like with this board is the inclusion of the 802.11ac controller, even if it is a 1x1 design using Intel’s AC3168. Even if you're planning to use wired Ethernet, Wi-Fi can be handy in a pinch when you're building, troubleshooting, or upgrading your PC and your work bench/kitchen table is nowhere near an Ethernet jack.

As we look along the outer edge of the board, we see two USB2.0 headers, one USB3.1 Gen1, and various front-panel headers on the bottom edge. RGB connectors are located in the bottom-left and top-right quadrants of the board, also accompanied by three of the board’s five 4-pin fan headers. The right edge hosts four angled SATA3 headers and the 2 vertical SATA headers which look lonely on this B450 chipset board. System power is provided through the standard 24-pin ATX header next to the four dual-latch DIMM slots, and an 8-pin EPS 12V plug is crammed next to a fan header in the top-left quadrant of the board. Lastly, a single 4-pin fan header is directly below the Vreg heatsink, for use with fans directly above the IO shield.

As simple as the PCIe configuration looks, there are some interesting things to keep in mind when picking parts. First, the M.2 slots are wired out for PCIe Gen3 x4 and x2 respectively, which is a pleasant departure from the Gen2 variants on other boards. However, the x16 slots are misleading, with only the top slot being wired out for PCIe Gen3 x16. The middle x16 and bottom x16 share bandwidth with the x1 slot, and the middle slot only works at x4 when neither of the others is populated. Otherwise, it runs at x2 mode.

Having looked at a few higher-end boards with X470, building with the B450 shows its limitations: The target audience steers component choices and our nitpicks. The fan header by the 8-pin EPS header really needs to be in a different location to be considered useful, and it would have been nice to have the 2 vertical SATA headers converted to angled headers and moved closer to the chipset. As for component recommendations, we suggest a dedicated GPU and either running a capture card or other device in the top or bottom slots. If extra bandwidth is needed, you can populate the x4 slot. But if you do, we see the risk of blocking the GPU’s airflow. We do not recommend running more than two IO cards in this motherboard.

MORE: Best Motherboards

MORE: How To Choose A Motherboard

MORE: All Motherboard Content

UEFI and Test Configuration

The Gigabyte B450 Aorus Pro WIFI’s UEFI, for the most part, is a copy what's found on the company's other AM4 products. So we'll focus on what's different from this board's X470 cousins. One thing we’ve never really covered is Gigabyte’s Easy Mode. Easy Mode is a pleasant departure from the default Classic mode and fills the screen with blocks of information covering system health, frequency settings, boot sequences, and various features enabled by the motherboard. Unfortunately, EZ OC is not enabled by default on this product, so we are stuck looking at the Eco mode icon which should really be swapped out for a “balanced” setting.

Flipping back to the classic menu, we are greeted with the M.I.T. menu by default, have access to the CPU clock ratio, and can engage available XMP profiles and modify the system memory multiplier. We’ll cover these settings in more detail later in the article but for this board and it’s limitations, this menu is all you really need to get a 4GHz overclock and DDR4-3200MHz (provided you have adequate cooling and a capable memory kit). Diving into the CPU core settings, we see similar options to the X470 boards, such as downcore control, SMT mode, and Cool ‘n Quiet.

Though not pictured, the advanced memory menu opens up the JEDEC knobs for DRAM, and users can fine-tune their memory timings until their hearts explode. For our testing, we just brute force XMP profiles and dial up the memory multiplier, but let us not get too far ahead of ourselves. Advanced voltage settings will be covered later on. But we should note that this option is completely different than the X470 boards' options. This B450 board is not wired for the higher level of tuning that X470 provides.

We’ve had mixed success with various vendors’ fan controller software, and Gigabyte’s Smart Fan 5 default fan control schemes are a little loud for our liking. Our Fractal Design fans can run very quiet, but this UEFI defaults the fans to max speed at lower temperature settings, so our provided fan curve proved to be quite effective in our testing. Your mileage--and preferences--may vary.. Another carry over from other AMD UEFIs, the board's RGB Fusion interface is effective in synchronizing the southbridge, audio isolation traces, and backplate shroud. The pulse mode setting is our light show of choice.

Though we often gloss over the Peripherals tab, it's important to know that Precision Boost Override and some Zen thermal control options are buried in the NBIO Common Options menu. In our testing, we do observe instances where the processor would throttle to a 5.5 multiplier, so the options are available if that is a setting you want to exploit. Another tab we primarily ignore is the Chipset tab, but that is where some APU, SATA, and RAID configurations are found.

Test System Configuration

Since our Gigabyte X470 Aorus Gaming 7 review, we’ve upgraded our memory configuration with a four-piece set of Corsair Dominator Platinum DDR4-3466, specifically the company's limited edition High Contrast SKU. While the aesthetics of this kit are nice, we hope the kit will helps us hit higher memory speeds in our overclocking section in subsequent reviews.

A top-end Ryzen 2700X is our CPU of choice for this testing, and it's cooled by the Corsair H110i 240mm AIO cooler. With the updated processor, it's time to upgrade our GPU from our trusty GTX 970 to a respectable GTX 1080 Gaming 8G from MSI. Our test monitor is a 4k LG 27UD58 that enables us to test FHD, WQHD, and UHD for all our gaming benchmarks. We still employ the Toshiba RD400 256GB M.2 NVMe drive for storage and various hard drives for Steam Libraries and miscellaneous software.

We have some goals to figure out while comparing data today. Our primary focus is how B450 compares against some of our previous X470 samples we have reviewed. Next is how does a “value” X470 board compares against a “premium” B450 board. Last is, of course, how well does the Gigabyte B450 Aorus Pro WIFI handle our test suite. It's a lot of ground to cover, so let’s get started.

Benchmark Suite

MORE: Best Motherboards

MORE: How To Choose A Motherboard

MORE: All Motherboard Content

Test Results: Synthetics, Games, Power and Thermals

Comparison Products

Synthetics and Applications

Finally, we have a full set of data where PCMark08 shows negligible differences between test samples, with the largest deviation coming from the creative benchmarks on the B450 board. Sandra’s Arithmetic shows sub-1 percentage point deltas across the board. Multimedia sees only a slightly larger 1.6 percentage point delta with the x16 Integer AVX workload. The Gigabyte B450 Aorus Pro WIFI draws a virtual tie with the Aorus Gaming 7 in Cryptography and Memory Bandwidth workloads.

Similar spreads are seen with Cinebench, but the Asus TUF X470 starts to creep closer to the B450. Compubench’s seven workloads don’t fit completely on this chart, but with all the data taken into account, the B450 Aorus Pro WIFI is the standard output for all metrics with the MSI X470 M7 AC finally taking a well deserved 2nd place finish.

Gaming synthetics are due for an update on our test bench, and 3DMark Skydiver proves the point with a 15 percentage point gap separating the combined scores. Physics scores don't fare much better with nearly 10 percentage points between 1st and last place. Fortunately, the metrics do start to tighten up with the more-demanding Firestrike at 1080p and Firestrike Extreme’s 1440p output resolution equalizes all test samples to within one percentage point.

Before we get to the gaming results, let’s see how the Ryzen 2700X performs in some application tasks. Handbrake encoding presents a 2.3-minute gap between the MSI board and the Gigabyte X470 Aorus Gaming 7 Wifi, and 7-Zip shows a similar trend but with a 1.17-minute gap. In both Blender workloads, it appears the test samples do start to tighten up, but the MSI board continues to struggle. Adobe finally pulls the MSI board back into the competition with impressive After Effects times, but the Aorus B450 board falls behind in Illustrator and After Effects. 

Gaming and Performance

Ashes of the Singularity: Escalation has been a great indicator of CPU performance in our test bench and we see the B450 Aorus Pro WIFI slump by roughly 4 frames per second (fps) at high settings at 1080p at all batch conditions. Bumping up the details and resolutions though do tighten up the spread across all review samples, but the B450 continues to lag behind the leaders by a few frames. In the end, this RTS game is easily enjoyable on this test bench across all batch conditions all the way up to 4k.

With F1 2018 closely approaching, we are eager to see how Codemasters leverages higher-end machines to increase our racing experience in this authentic simulator. For this Ryzen test bench and GTX 1080, 4k Ultra settings are easily playable with frame rates in the 70s and the Gigabyte B450 Aorus Pro WIFI only manages to slow down by up to 5 frames compared to the Aorus Gaming 7 WIFI. Unfortunately, that does translate into larger percentage gaps between the leaders, so we’ll have to keep testing F1 2015 to see how it stacks up.

Adjusting the CPU speed, GPU speed, and GPU memory settings across various resolutions in The Talos Principle is less than exciting. At Medium settings, we observe a 7.5 percentage point drop in frame rates on average, stepping up to 4k from 1080p. At Ultra settings, we more than double the delta to roughly 17 percentage points, thanks in part to the increased complexity in renders at the elevated resolutions. As for the Gigabyte B450 Aorus Pro WIFI, it manages to best the MSI X470 board in most scenarios, but does lag behind the X470 boards as a whole.

Similar to F1 2018, we look forward to Metro: Exodus in 1Q 2019. But until then Metro: Last Light Redux will have to keep us busy. All four samples show really stable 4k results at both detail levels here. But the Gigabyte X470 Aorus Gaming 7 WIFI manages to pull 10 fps ahead of the pack at 1080p and high details.

Taking all the average results for each metric into account, the Gigabyte B450 Aorus Pro WIFI only appears to take wins in the Synthetics category, and lags behind with gaming and applications for the most part. The higher-end Gigabyte X470 Aorus Gaming 7 WIFI is the clear performance winner in this chart, with the TUF X470 showing fairly strong results in areas that the B450 struggled in. 

Power, Temps, and Efficiency

This power draw graph is slightly confusing, but we do see some clear winners coming out of this test. For idle power, the B450 runs in the same ballpark between the Asus TUF and the X470 Aorus Gaming 7 WIFI. Full CPU load shows the MSI X470 M7 AC clobbering the B450 board by 14 percentage points, but full GPU load shows a 16W delta between those same boards favoring the B450. However, full system power virus shows how difficult it is to properly load the processor and GPU in a manner that effectively runs each component at TDP conditions in parallel.

The temperature story is very straightforward for a change, now that we aren’t changing our cooling solutions, and each board saw the exact same case and fan configurations. The 10C Tctl delta is apparent in our data, and CPU temperatures only vary by 5C under full Prime95 load. However, the real story is those voltage regulator temperatures. The TUF heatsink is severely under designed for its target market, but the Gigabyte B450 heatsink doesn’t fair much better but is properly suited for its target market.

In our efficiency calculations today, we removed the idle and max power data from our analysis to show a more realistic representation of what these systems are capable of running while in a typical use case. Each board shows interesting behavior: The MSI uses less power under load, but it’s performance suffers, and in-turn it becomes less efficient. The Asus TUF X470 has more performance but at the cost of slightly larger power draw, which hurts its efficiency score. Ultimately, the Gigabyte B450 Aorus Pro WIFI just draws too much power compared to its slight performance loss, thus dragging its efficiency rating below ideal scenarios.

MORE: Best Motherboards

MORE: How To Choose A Motherboard

MORE: All Motherboard Content

Overclocking, Value and Conclusion

Overclocking

We’ve been quite successful with our latest test bench in generating adequate overclocks, with Prime95 as our CPU workload and AIDA64 driving our memory tests. Having been spoiled by our X470 coverage, it’s time to see how these lower-tiered boards fair. Jumping into the UEFI, we kick the multiplier up to 40.0 for initial testing and both thermals and stability are in line with expectations. Multiplier increases in 0.25 steps get us all the way up to 4100MHz without adding any additional voltage. As we increment the voltage (with 6.25mV step sizes), we observe that this board does not support load line calibration changes. We confirmed with Gigabyte that the analog IC’s do not support loading changes, so we won’t be expecting a UEFI patch for that.

As we increase the voltage, we start to see a frequency overclock threshold of roughly 4175MHz, where the system becomes unstable in Prime95 with varying times to fail. Increasing the voltage up to a +204mV offset, we finally reach the regulator's adjustment limit through UEFI. Regulator temperatures are a toasty 113 C at those settings, which seems hot but is within the safety limits of the PowerPak MOSFETs. Turning down the clock multiplier in our search for stability, we find an effective maximum rate of 4125 MHz.

At this point, we have lots of voltage headroom to play with, so for this board, the regulator design is sufficient. The Gigabyte B450 Aorus Pro WIFI’s Vreg heatsinks certainly appear to be underdeveloped, and could use some additional slots in the slabs of alloy for an increased surface area. There are some videos out on the web questioning Gigabyte's claim that it utilizes eight “real” phases, but the implemented design appears adequate for modest overclocking. And AMD's 2nd Generation Ryzen chips don't tend to clock significantly higher than 4GHz anyway.

Gigabyte’s X470 board showed exquisite memory overclocking. So we were optimistic when firing up our DDR4-3466 Corsair Dominator Platinums in the lesser B450 board. Shockingly, the system is able to boot at the default XMP profile, but it only got through about two seconds of AIDA64 before the workload crashed. Without load line calibration, we have to dial down settings to gain stability, and we brute force it with the memory multiplier. In 66MHz steps, we have to dial down to DDR4-3266 before we see AIDA64 run for 30 minutes without erroring out. Looking back, this is huge victory compared to B350, and even rivals some of the X470 variants on the market. 

Value and Conclusion

As a high-end offering for its mid-range chipset, the Gigabyte B450 Aorus Pro Wifi draws on the Aorus brand for inspiration, but just barely misses the mark in terms of raw performance. However, this board is well equipped in certain areas that make it a much more straightforward purchase for anyone that is interested in good aesthetics.

Going back to the questions we posed earlier, does the B450 stand up to X470? For the most part, yes. From a raw performance perspective, the chipset can deliver performance that is in line with our 2700X expectations. But that comes at the cost of certain features and quality of life enhancements, namely improved regulator designs, voltage controls, and slight increases in IO.

Next, do we consider this “premium” mid-level board better than a “value” high-end board? Yes, the Gigabyte B450 has features that elevate it above the designs of previous value-oriented boards that make it a clear and easy pick above the X470 chipset. Given our overclocking success and reliability concerns of other designs, the B450 is better suited for a broader audience.

Coming in at a shockingly affordable $120 MSRP, it's hard to argue that Gigabyte doesn't deliver a heaping bucket of value with this board. Sure, some features that come with higher-range chipsets aren’t available, but there has to be some substantive feature delineation between the premium chipset and the one that's a step down.

Gigabyte's inclusion of Wi-Fi, appropriate overclocking for this market segment, and features ideal for a gaming build make the Gigabyte B450 Aorus Pro WIFI an excellent choice for someone that isn't willing to drop the coin to jump up to a higher-end package. So long as you aren't plugging in a second graphics card, a big bank of drives, or lots of bandwidth-craving USB devices, the $50 savings this option offers over some of the other X470 options is better spent on a new AAA title, a more spacious and/or speedy boot SSD, or a graphics card upgrade.

MORE: Best Motherboards

MORE: How To Choose A Motherboard

MORE: All Motherboard Content

Intel appears to be readying new sets of mobile and desktop processors, with Finalwire adding a plethora of new CPUs to a recent beta version of its popular AID64 benchmarking suite, including what appears to be the first mobile i9 processor.

The release notes of AID64 Extreme beta version 5.92.4397 are highlighted by several new entries for the program’s Intel Processor Number detection feature. All of the new CPUs are listed as 8^th and 9^th generation parts (8000 and 9000-series), ranging from Celeron and Core i3 to Core i7 and a new Core i9. The applicable release notes are listed below:

“Identification of Intel Core i7-8000H Series (aka Coffee Lake-H MB)Identification of Intel Core i9-8000H Series (aka Coffee Lake-H MB)Identification of Intel Xeon E-21xx(G) (aka Coffee Lake-S WS)Identification of Intel Xeon E-21xxM (aka Coffee Lake-H WS)Intel Processor Number detection for Celeron G4900, G4900T, G4920, G4930, G4930T, G4950Intel Processor Number detection for Core i3-8000, 8000T, 8020, 8020T, 8100T, 8120, 8120T, 8300T, 8320, 8320T, 9000, 9000T, 9100, 9100T, 9300, 9300TIntel Processor Number detection for Core i5-8300H, 8400B, 8400H, 8400T, 8420, 8420T, 8500, 8500B, 8500T, 8550, 8650, 8650K, 9400, 9400T, 9500, 9600, 9600KIntel Processor Number detection for Core i7-8670, 8670T, 8700B, 8700T, 8750H, 8850HIntel Processor Number detection for Core i9-8950HKIntel Processor Number detection for Pentium Gold G5400, G5400T, G5420, G5420T, G5500, G5500T, G5600, G5600T, G5620Intel Processor Number detection for Xeon E-2176M, 2186M”

In total, the AIDA64 beta release adds support for 57 new Intel processors. Although Coffee Lake mobile processors have been in the works for some time, this is the first solid source to disclose the specific models we’ll be seeing when the chips debut. New desktop processors are also inbound, with Intel releasing a mix of Core i3, i5, and i7 desktop models that appear to target lower power ceilings, clockrates, and pricing than its current line of CPUs, judging by the model numbers and suffixes.

The presence of 9000-series CPUs may also indicate that Intel is releasing a new processor generation, but whether or not these models are Coffee Lake refreshes (similar to the recent slew of Kaby Lake refresh CPUs) or 10nm Cannon Lake chips is anyone's guess at this point. The timing seems somewhat premature for the anticipated architectural die shrink, but with CES just around the corner and CPU competition higher than it's been in decades, anything is possible.

Most notably, it seems the mobile segment is about to receive its first Core i9 CPU, the -8950HK. We can only speculate what this new CPU will bring to the table in terms of core count and clock speeds, but it seems to indicate that Intel plans to offer six-core (or possibly eight-core) mobile processors for the high-end mobile market.

The Ryzen Debut

The conundrum AMD currently faces started when it launched Bulldozer to lackluster reviews back in 2011. The following years found it trying to right the ship with Piledriver, Steamroller, and Excavator. But it’s safe to say the company's host processing portfolio never regained its lost luster. Meanwhile, Intel dominated the mobile, desktop, and server markets with a seemingly insurmountable performance built on the excellent Sandy Bridge design and an unrelenting cadence of incremental improvements.

Heading into today's review, nobody can argue the fact that AMD is far behind by comparison. Wouldn't that make a comeback all the more impressive, though?

We started seeing Zen micro-architecture teasers last year. Company representatives told us its next generation would usher in incredible performance gains, matching or surpassing Intel's best efforts on multiple fronts. Zen also promised to revitalize an aging platform through Socket AM4 and new core logic. And AMD says it has a clear path forward planned for future versions of Zen.

At some point, though, the rubber has to meet the road. A good first step was taking aim at a competitor. Intel's $1000+ Core i7-6900K seemed like an ambitious choice, but early hand-picked benchmark results made AMD's eight-core engineering samples look formidable. Then, announcing that the flagship model would sell for less than half of of the -6900K's price sent the masses into a frenzy. Most online vendors even sold out of their Ryzen 7 1800X allocation during pre-sales based on little more than AMD's own endorsement.

Now it's time for Ryzen to stand up on its own accord and show us what it can do in the real-world. We have several Ryzen SKUs in-house, spread across multiple Tom's Hardware labs. We've identified a number of unexpected results that bear continued investigation. We'll continue updating our coverage as answers materialize. But we want to start putting our findings in front of you so enthusiasts can make more informed buying decisions in the face of general availability, which begins today.

Finding Zen

Four years ago, AMD began its work on the Zen core, which is its first clean-sheet architecture since Bulldozer. AMD’s initial objective was to transition from the 28nm process used for its modern APUs to GloablFoundries’ 14nm FinFET node, which offers increased performance and density within a similar power envelope. The company also set an ambitious goal to increase instruction-per-clock throughput by 40% over Excavator through a series of design choices that significantly boost performance. Notably, AMD deployed a new architecture and a lithography shrink simultaneously, which is a daunting challenge.

Last year, we published Everything Zen: AMD Presents New Microarchitecture At HotChips. In that story, we stepped through the composition of Zen, from front to back, right up to describing the CPU complex (CCX) responsible for housing four execution cores, each core's 512KB L2 cache, and 8MB of shared L3 cache. If you aren't already familiar with Zen and how it differentiates from prior-gen designs, check that piece out.

Moving forward, you need to know that the Zen core is Ryzen's fundamental building block. All three SKUs we're introducing today employ two quad-core CCXes, adding up to 4.8 billion transistors across the entire die. The company says its Infinity Fabric connects the CPU complexes, but remains shy about how that's a quantifiable benefit.

As we established in our architectural deep-dive, AMD also arms Zen with simultaneous multi-threading support, allowing each physical core to operate on two threads in parallel. In theory, this improves the utilization of available hardware resources. A lot of our real-world benchmarks bear that out with phenomenal performance gains. But other workloads expose teething pains we're still trying to diagnose.

How about the 40% IPC improvement goal AMD set for itself? Well, after factoring in the new micro-op cache (bypassing the L1 and L2 for frequently-accessed micro-ops), the better branch prediction engine, the 1.75x-larger instruction scheduler window, and faster caches, the company cites a +52% final tally compared to Excavator. Naturally, we have our own single-threaded workloads to run and will gladly make comparisons using CPUs from our lab.

The Ryzen 7 Line-Up

AMD is splitting its newest CPUs into the eight-core Ryzen 7 family, a six-core Ryzen 5 series, and the quad-core Ryzen 3 line-up. Only the Ryzen 7 SKUs are shipping today, but it's easy to see that AMD is targeting Intel's Core i7, i5, and i3 portfolios with a similar naming scheme.

Aside from Intel's eight- and 10-core i7s, the Ryzen 7s deliver higher core counts across the board. The AMD CPUs also blow Intel's Broadwell-E prices out of the water, though four-core/eight-thread Kaby Lake is generally cheaper (albeit with half as many cores).

It's not entirely clear what features AMD plans to roll out across the Ryzen 5 and 3 CPUs, but we do know 7s sport the SenseMI suite. We'll go into more depth on SenseMI shortly. What's important here, though, is that SKUs with an X suffix include the eXtended Frequency Range capability. XFR automatically increases clock rate beyond the factory-set Precision Boost frequency if you provide additional thermal headroom with an aggressive cooler. This extra bit of speed applies to two of the chip's cores.

Ryzen 7 is solely a host processor, devoid of integrated graphics. All three models debuting today drop into the Socket AM4 interface, include eight physical cores, and boast 16MB of shared L3 cache. They also sport unlocked ratio multipliers, though you'll need a motherboard based on the X370, B350, or X300 chipsets to overclock.

The Ryzen 7 1800X features a 3.6 GHz base frequency able to hit 4 GHz under lightly-threaded workloads via Precision Boost technology. Both of those specifications are higher than Intel's eight-core Core i7-6900K. With Precision Boost enabled, all of the 1800X’s cores can operate at 3.7 GHz. And with enough thermal headroom, two cores jump as high as 4.1 GHz.

Perhaps surprisingly, given the comparisons to Intel's 140W Broadwell-E behemoths, 1800X bears a 95W TDP. If that's not enough to make you believe AMD has a new lease on life, the $500 price tag should excite professional content creators especially. Of course, if you don't regularly find yourself running heavily-threaded tasks, Ryzen 7's value isn't as pronounced. After all, Intel's Kaby Lake-based Core i5s and i7s offer solid performance and generally sell for less than the top-end AMD chips. Ryzen 7's performance in our benchmark suite will have to justify the premium.

The 95W Ryzen 7 1700X's clock rates drop to 3.4 GHz base and 3.8 GHz under Precision Boost. Those frequencies compare favorably against the 140W Core i7-6800K, which tops out at 3.6 GHz in lightly-threaded tasks and only comes equipped with six cores. Worse, Intel charges $425 for the -6800K, while AMD is introducing Ryzen 7 1700X at $400. The Core i7-7700K also becomes relevant at this point, with its $350 price tag.

AMD's Ryzen 7 1700 has a 65W TDP, making it the lowest-power eight-core desktop CPU available. A 3 GHz base clock rate and 3.7 Precision Boost ceiling are significantly lower than Intel's 91W Core i7-7700K. However, the company compensates with twice as many physical cores and a comparable price tag.

The six-core/12-thread Ryzen 5 family should surface in Q2, and include at least two models. The Ryzen 5 1600X will feature a 3.6 GHz base and 4 GHz Precision Boost ceiling, while the 1500X is expected to start at 3.5 GHz and ramp up to 3.7 GHz in lightly-threaded workloads. AMD hasn't shared cache configurations yet for those models. Ryzen 3s are also in the queue, though those aren't expected until the second half of 2017.

AMD geared its pricing structure to target the 99% of enthusiasts it says buy CPUs priced under $500. If Ryzen 7 is successful, the stage is set for even more disruption in the mid-range and low-end segments as well. So, does Ryzen begin its life on stronger footing than Bulldozer? Let's find out.

MORE: Best CPUs

MORE: Intel & AMD CPU Comparisons 

MORE: Everything Zen: AMD Presents New Microarchitecture At HotChips

MORE: Intel Kaby Lake Core i7-7700K, i7-7700, i5-7600K, i5-7600 Review

MORE: Broadwell-E: Intel Core i7-6950X, 6900K, 6850K & 6800K Review

AMD SenseMI Suite & XFR

AMD's SenseMI feature suite consists of five key technologies that allow Ryzen 7 processors to adjust performance and power consumption parameters in real time.

Pure Power

According to AMD, its Zen architecture employs an array of 1000 sensors accurate to 1mA, 1mV, and 1°C. The Pure Power feature monitors these temperatures, voltages, and currents, enabling real-time adjustments based on decisions made by what company representatives describe as learning algorithms.

Those sensors feed telemetry data across the Infinity Fabric loop to the Infinity System Management Unit at 1ms intervals. The management unit analyzes that data and issues commands across the fabric to adjust voltage and frequency settings for optimal performance. AMD also notes this functionality helps manage its speculative cache features and AI-based branch prediction.

Each piece of silicon is unique, and AMD points out that its algorithms allow the processor to optimize itself based upon its own characteristics. Notably, other semiconductor vendors employ a similar technique to control the power consumption of their processors dynamically.

Precision Boost

Precision Boost adjusts the power/performance curve to optimal settings for the operating environment, much like Intel's Turbo Boost, based on information derived from Pure Power. The algorithms that control Precision Boost facilitate changes in 25 MHz steps, which is of course more granular than Turbo Boost's 100 MHz increments.

On the Ryzen 7 1800X, for example, Precision Boost increases the 3.6 GHz base frequency to 3.7 GHz across all cores, and can push two cores up to 4 GHz. This is important: whereas Intel's Turbo Boost technology varies clock rate based on the number of active cores, Precision Boost draws that distinction between two active cores and anything in excess of two cores, at which point maximum frequency drops to the all-cores number. What you end up with is a nice speed-up in lightly-threaded tasks, but less benefit than Turbo Boost as soon as a third core spins up.

XFR (eXtended Frequency Range)

AMD's eXtended Frequency Range feature allows the processor to dynamically adjust its clock rate above the stock and Precision Boost clocks based on available thermal headroom.

XFR automatically increases the Ryzen 7 1800X's Precision Boost ceiling by 100 MHz if your cooling solution can keep the CPU running below a certain threshold. We tested with Noctua's NH-U112S SE-AM4 air cooler and Corsair's H100i v2 closed-loop liquid cooler, and both perform well enough to engage XFR, allowing a 4.1 GHz peak clock rate. Unfortunately, we don't have AMD's stock heat sink/fan, so we don't know if it's beefy enough to make XFR work.

AMD claims this feature scales with air, water, and LN2 cooling, but doesn't specify if there's a maximum frequency with LN2.

Interestingly, the Precision Boost and XFR features are intertwined on our Asus Crosshair VI Hero motherboard. If you enable the "Core Performance Boost" setting, the 1800X triggers both Precision Boost and XFR frequencies based on your workload. But you cannot disable either feature independent of the other.

Neural Net Prediction & Smart Prefetch

AMD claims that its Neural Net Prediction capability describes a built-in neural network able to learn application behavior and pre-load instructions before they're needed, while Smart Prefetch learns access patterns to pre-load data into local cache.

Both features appear to be marketing-speak for AMD's perceptron branch predictor, which debuted in the Jaguar core. Of course, the company likely improved and optimized this technology for Zen, specifically, but it's not magic.

The AM4 Platform

AMD announced its AM4 interface and corresponding chipsets during the Bristol Ridge launch. Previously, core logic was a weak point, as the 9-series chipsets lacked a lot of modern features available from Intel's platforms. Moving forward, all Ryzen CPUs share the same socket. This gives you a number of options for building a Ryzen-, Bristol Ridge-, or future Zen+-based system.

Similar to Intel's platform controller hub architecture, AMD pushes integration with the move to Socket AM4 and tasks its chipset with functions typically associated with southbridges. Meanwhile, it adds more capabilities to the processor die you wouldn't expect to find there. For instance, Ryzen 7 1800X provides four USB 3.1 Gen 1 ports. It also offers 16 lanes of PCIe 3.0 we're guessing most enthusiasts will use for graphics, along with four lanes of second-gen PCIe for SATA 6Gb/s and NVMe storage. In time, we may even find board vendors building platforms based on just the CPU's I/O.

Intel's Core i7-6900K doesn't expose any USB connectivity directly. Rather, it relies on X99 and attached controllers to enable those ports. On the other hand, Broadwell-E packs up to 40 lanes of third-gen PCIe for a lot more flexibility in what you attach. Ryzen's ability to split its 16-lane PCIe 3.0 link into two x8 slots looks a lot more like Intel's mainstream chipset line-up in comparison.

AMD breaks Socket AM4 down into five different chipsets. Enthusiasts will want to pick up X370. B350 targets the mainstream crowd, whereas A320 goes the locked-down route for "essential" builds. For those of you who anticipate deploying APUs in HTPCs and LAN boxes, X300 and A/B300 should address those markets (we're still waiting on their specifications).

Given Tom's Hardware's focus on the enthusiast and mainstream markets, our motherboard reviews will emphasize X370 and B350. The two platforms are differentiated in much the same way AMD positioned the old 990FX and 970: a higher-end chipset gets you extra I/O, mostly. X370 includes four extra USB 3.1 Gen 1 ports, two additional SATA 6Gb/s ports, and two more second-gen PCIe lanes. Moreover, B350 doesn't allow you to split the 16-lane PCIe 3.0 link into two slots like X370 does.

Both Socket AM4 platforms support overclocking out of the box though, in addition to RAID 0/1/10 and two SATAe ports. AMD's documentation notes that SATAe ports can be broken up and re-purposed for additional SATA ports or second-gen PCIe lanes, potentially opening the door to larger storage arrays or additional M.2 ports.

According to AMD, the Socket AM4 interface will carry it through 2020. By then, technologies like DDR5 and fourth-gen PCIe should be prevalent, paving the way for newer platforms.

Overclocking & Test Setup

AMD Ryzen Master

AMD's Ryzen Master software allows you to tune a number of variables, such as the CPU's ratio multiplier, voltage, and memory speeds, from inside the operating system. When you open the utility, it warns you of the perils associated with overclocking and disavows AMD of responsibility should you inadvertently nuke your processor. When you adjust parameters, the CPU automatically switches into OC mode and disables all of the normal thermal and voltage restrictions.

You're able to disable cores in groups of two, which could help you hit more aggressive frequencies. Individual cores cannot be sped up or slowed down. Rather, adjustments apply to all cores.

It's worth noting that the older AMD Overdrive tool is not compatible with Ryzen CPUs. 

We prefer overclocking from the motherboard firmware, though. Using simple multiplier and voltage adjustments, we achieved a Prime95-stable 4 GHz clock rate at 1.425V using Asus' Crosshair V Hero (that was with load-line calibration set to Auto). The highest temperature we saw was 82°C during our stress test. 

Of course, if you're using a cooling solution less capable than our Corsair H100i v2 at its maximum fan/pump settings, overheating could become a problem. AMD predicts that most customers should see somewhere between 3.9 and 4.1 GHz across all cores, and suggests you stick with a 1.35V ceiling if you want your chip to last. Although core voltages in excess of 1.45V are considered sustainable, they'll have a more pronounced effect on longevity.

Asus notes that pushing memory transfer rates in excess of 2933 MT/s on its Crosshair VI Hero requires memory binned at 3200 MT/s or higher with Samsung A-die ICs. AMD hasn't opened up all of Ryzen's memory sub-timings yet, but we expect the company to open this up soon. Currently, Asus' testing with Hynix A-die ICs tops out at 3000 MT/s, but that may improve with microcode updates.

Measurement System and Methodologies

Splitting our tests between two locations proved difficult this time around, since AMD sent our German lab samples two days before the launch. We wanted to benchmark with retail hardware though, not downclocked engineering samples, so we raced the clock to crank out results.

Our challenge was complicated by AMD's decision to send our labs two different motherboards. Germany received MSI's X370 XPower Gaming Titanium, so all of our workstation, HPC, and power consumption tests were run on that platform. Meanwhile, the U.S. team used Asus' Crosshair VI Hero and the EVGA GeForce GTX 1080 FE for graphics testing.

MSI’s X370 XPower Gaming Titanium, just like Asus’ model, is a flagship motherboard based on AMD's X370 chipset. We stuck with AMD’s recommended presets on both machines, minimizing issues attributable to our dissimilar boards. Our German office used two 8GB Corsair Vengeance LPX DDR4-2666 memory sticks and a custom water-cooling loop.

If you want to know more about how the Tom's Hardware DE system looks and is controlled, check out How We Test Graphics Cards.

To keep things simple, we present the highlights in the following table. The Intel systems are identical to those for the gaming tests.

Cache Testing

The Cache Testing Dilemma

The Ryzen reviewer's guide contained an interesting note regarding L1, L2, and L3 cache measurement tools. AMD indicates that AIDA64 and SiSoftware Sandra, both commonly-used tools for cache testing, are "not yet equipped to accurately measure cache performance of the Zen architecture." AMD provided its own internally-measured reference values and noted that it is working with the FinalWire (AIDA) and SiSoftware teams to facilitate accurate Zen cache measurement methodology in the future.

We measured performance with the utilities and achieved similar results for Intel's Core i7-6900K, but we also noticed a large gap between the AMD-provided Ryzen measurements and our test results. Ryzen’s L3 cache latency measured 20 ~ 23ns, which is double the provided value. Due to some of the performance characteristics we noted during our game testing, we also tested with SMT enabled and disabled, but the results fell within expected variation. We also measured a ~10ns memory latency gap in favor of the Intel processor.

Many common utilities write zeros to the cache to measure performance. AMD responded to our inquiries and stated that Intel coalesces incoming "zero" write traffic before it passes it to the cache, which could yield artificially high cache throughput measurements, particularly because those patterns don't exist in real-world usage. In our opinion, changing the access pattern would result in reduced performance measurements for the Intel processors, but it certainly shouldn't boost AMD's cache measurements. AMD responded that the current utilities are also not optimized for Zen's unique architecture, and optimizing the utilities' code paths will expose more performance.

We contacted both SiSoftware and FinalWire to request any updated or beta versions of their utilities that can facilitate accurate testing. There is still collaboration between the company's and AMD to resolve the issue, but unfortunately, we aren't at liberty to discuss the details of those conversations. Until that time, and until the vendors and AMD agree on what's happening, we've determined that drawing any conclusions from these results would be irresponsible, and ultimately misleading. When we have final, verified, and real numbers, we'll provide them.

Ashes of the Singularity & Battlefield 4

A Note On Testing

As mentioned, we encountered lower-than-expected results in many of our game benchmarks. Concerned that the explanation could be related to Windows' handling of P-states, we ran tests on Ryzen under the High Performance and Balanced power profiles as a diagnostic measure. The former is denoted with an HP in our charts. Intel's chips are tested using Windows' Balanced profile, representing the way most of us configure our desktops.

The Zen architecture is AMD's first with simultaneous multi-threading, so we also tested the Ryzen 7 1800X with SMT disabled to flesh out any performance deltas attributable to this feature. Indeed, we observed higher performance with SMT turned off in some titles.

Finally, we normalized the clock rates of our test samples to compare per-clock performance. Game engines and DX11 do not always scale linearly, so our 3.8 GHz numbers merely serve as a reference.

In light of the unexpected results we do present, we have to warn you that some of our data could be attributable to factors like motherboard firmware or early microcode. Then again, these are products AMD and its partners are selling, so that's a risk early adopters assume. In the meantime, we continue running tests and asking questions. Expect a follow-up soon.

Ashes of the Singularity

Ashes of the Singularity scales well with additional cores and frequency. This rewards the Core i7-6900K and its broadside of eight cores and 16 threads.

Surprisingly, Ryzen 7 1800X does not reap the same benefits. We expected much higher performance in this CPU-intensive title from the 1800X due to its similar core configuration and higher clock rates. In the end, Core i7-6900K outperforms the stock Ryzen setup, despite AMD's seemingly favorable specs.

We disabled the 1800X's SMT feature and its average frame rate increased to 70.9 FPS. By comparison, we disabled Hyper-Threading (Intel's equivalent technology) and re-tested the -6900K, yielding a 7 FPS performance loss.

Surprisingly, even the 4C/8T Core i7-7700K outpaced AMD's Ryzen 7 1800X by a significant margin (and that was after trying to provide the 1800X with the most favorable settings possible). We also noted minimal uplift from overclocking the 1800X to an all-core 3.8 GHz. Due to the automatic XFR function, and the fact that we couldn't disable XFR without losing all Precision Boost capabilities, we couldn't test at AMD's base 3.6 GHz.

AMD provided its own benchmark numbers at 4K, creating a graphics-bound situation. We tested at 2560x1440 to reduce this bottleneck, exposing more difference between CPUs. The same strangeness happens at QHD too though, and 1800X trails the field by a quantifiable margin. The 1800X's behavior suggests the game isn't optimized for Zen at this point; Oxide Games released this statement:

“Oxide games is incredibly excited with what we are seeing from the Ryzen CPU. Using our Nitrous game engine, we are working to scale our existing and future game title performance to take full advantage of Ryzen and its eight-core, 16-thread architecture, and the results thus far are impressive. These optimizations are not yet available for Ryzen benchmarking. However, expect updates soon to enhance the performance of games like Ashes of the Singularity on Ryzen CPUs, as well as our future game releases.” - Brad Wardell, CEO Stardock and Oxide

Battlefield 4

We traipse into graphics-bound territory during our 1920x1080-based Battlefield 4 testing. Ryzen 7 1800X in its stock configuration (with the Balanced power profile) is the only modern processor that doesn't average 160 FPS or more. It fares better with the High Performance profile activated, pulling in with the rest of the field if you're willing to leave Windows in this mode.

The FX-8350 trails at its 3.8 GHz setting, though that's a down-clock compared to its 4 GHz base frequency. The similarly-clocked Ryzen 7 1800X provides a notable speed-up compared to the old -8350.

The Ryzen 7 1800X provides the same performance as Intel's Core i7-6900K, pushing the EVGA GeForce GTX 1080 FE into graphics-bound territory. If your favorite titles tend to be GPU-limited, you can save $500 on an eight-core processor that offers a lot of heft in more heavily-threaded applications (or save an additional $175 on a quad-core chip that games really well).

Hitman, Project CARS & Metro: Last Light

Hitman (2016)

Hitman scales well during our benchmark, obviously responding to clock rate and core count. The Ryzen 7 1800X receives a boost when we disable SMT, but it still lags the field under the influence of a slight overclock to 3.8 GHz.

The FX-8350, which experiences severe frame time variance, highlights how far AMD has come in a few short generations on the performance front.

The Core i7-7700K takes the lead with a fixed 3.8 GHz frequency, but Ryzen closes the gap as resolution increases. The 1800X at its stock clock rate shows up well behind Intel's Core i7-6900K, but to keep things in perspective, you get 87% of Broadwell-E's performance for less than half of its price with Ryzen 7 1800X. Conversely, you can opt for the $350 Core i7-7700K and enjoy more performance than AMD's 1800X in many popular titles. 

The gap narrows even more between Ryzen and Intel's processors as we shift to 2560x1440. The Core i7-6900K and -7700K are still faster, but the utility question surfaces again: if you specifically need an eight-core processor for content creation, you can go with the 1800X and give up some gaming alacrity, or, if high frame rates are top priority, buy the Core i7-7700K knowing it won't be as fast elsewhere.

Project CARS

Intel's Core i7-7700 takes a big lead during our Project CARS testing, followed by the Core i7-6900K. AMD's Ryzen 7 1800X with a fixed 3.8 GHz frequency slides past the SMT-disabled results, suggesting the game works well with AMD's SMT implementation.

Project CARS delivers similar maximum frame rates, even at a higher resolution, which illustrates just how CPU-constrained the game is. The 1800X without SMT and the 1800X under Microsoft's High Performance power profile swap positions on the chart, but are only separated by 0.1 FPS.

The FX-8350 suffers significant performance variation, while Ryzen 7 1800X provides a smoother experience using the High Performance profile. Intel's Core i7-7700K continues to impress with a beastly lead over the other processors, including the -6900K overclocked slightly.

Metro: Last Light Redux

The Ryzen 7 1800X averages 91+ FPS during the benchmark, and only lags the Core i7-7700K by 2.8 FPS.

The FX-8350 falls to the bottom of our chart, and though it trails the rest of the field by a marginal average frame rate, it experiences much more frame time variance.

In all four configurations, the Intel processors are separated by a scant 0.2 FPS during the graphics-intensive workload. The Ryzen 7 1800X offers nearly the same performance as the leading CPUs.

Results: Desktop & Office

It takes more than 18 hours to run the office, workstation and HPC (High Performance Computing) benchmarks once. Consequently, we limited our tests to a total of four CPUs and two runs each (i.e., at stock clock frequency and at 3.8GHz). We used Microsoft Windows’ performance profile for all of the tests. AMD’s older generation is represented by the 5GH FX-9590, which necessitates water cooling.

We get going with a good old friend: the GDI/GDI+ benchmark. We decided to start with it, because its results will provide context for what’s to come later. This benchmark is used to test two different ways to output 2D objects. These are primarily found in older applications, print output, and, in a modified form, in today’s GUI display. We’re using an Nvidia Quadro P6000, since it represents the best available graphics performance that money can buy. This means that there’s no GPU bottleneck to worry about whatsoever.

We start with directly drawing to an output device. The graphics driver uses the CPU quite heavily for this task, but, for the most part, doesn’t use that many threads. This is due to the demise of true 2D hardware acceleration with the introduction of the unified shader architecture. The Microsoft Windows driver model’s also not exactly conducive to it. It’s interesting to see that, in spite of all of this, AMD’s new Ryzen 7 1800X manages to beat Intel’s Core i7-6900K. Since each individual action has to make its way through the entire system, these results aren’t just representative of processor performance, but also include chipset performance.

Next, we add a memory task by using the single 2D function that’s still hardware-based: copying the graphics output from the place in memory where it was built to the output device. We’re executing the exact same steps, but draw a virtual bitmap and not on the monitor. Only the completed picture is sent to the output device. The result is that the CPU load increases significantly, which, in turn, has a surprisingly large impact on the benchmark results. Without the rest of the system providing a bottleneck, the older FX-9590 manages to beat AMD’s new offerings. Only if its clock frequency is lowered to 3.8GHz does it have to admit defeat. The same goes for Intel’s Core i7-6900K.

Update (this is in reference to some reader questions about these results): Drawing into DIB is a pure 2D workload, and the stock i7-6900K can't reach its highest possible clock rate in this low-priority multi-threaded workload. To be specific, the i7-6900K spreads the workload over its 16 threads, so the processor runs at close-to-idle clock speed, because it only utilizes 20% of many of its cores. The distributed workload also strips the processor of its Turbo Boost capabilities.

By comparison, the -7700K spreads the workload among half the threads, which yields a higher load per core, thus boosting its clock speed. The -6900K at 3.8GHz also experiences a boost because it has 50% more clock speed than its stock configuration.

This test further implies that AMD is able to transition between power states faster than the Intel processors, which means it is tuned well for light workloads.

These findings are of great significance for AutoCAD, which we get to on the next page. Ultimately, AutoCAD draws all of its output in memory first, and then sends it to the monitor.

DTP & Presentation

Adobe’s Creative Cloud makes for an exemplary benchmark, and it certainly makes more sense to use a real-world application than a synthetic benchmark. It tests both single- and multicore performance.

Encoding & Multimedia

AMD’s Ryzen 7 1800X can live up to its full potential with our Handbrake benchmark. No matter what the quality is set to, AMD’s new processor just crunches away.

Compression & Decompression

Crunching lots of numbers in short order is also very useful for file compression. AMD’s Ryzen 7 1800X manages to draw even with Intel’s Core i7-6900K at the same clock frequency. Intel’s Core i7-7700K takes the lead when it comes to decompression due to its higher clock frequency. In the end, packing’s harder than unpacking.

Programming in Python

This application sits right on the boundary to the workstation space, and thus leads us seamlessly to the next page. Apart from the visual task library (e.g., graphically complex charts), which pose problems due to the usual reasons (i.e., difficulties with parallelization), the scientific and engineering fields are where AMD’s Ryzen 7 1800X shines.

Ryzen’s a well-suited CPU for both developers and users due to the fact that many libraries don’t provide an explicit advantage for Intel processors.

Bottom Line

With Ryzen, AMD has produced a processor that’s competitive in many application areas. It certainly doesn’t have to hide from Intel’s current octa-core offerings. The price/performance ratio makes it an interesting choice for commercial applications, unless the application in question specifically demands an Intel CPU.

Results: Workstation

Let's take a look at several common workstation productivity applications.

The AutoCAD results remind us of what we reported on the previous page when we looked at output devices. AMD’s Ryzen 7 1800X ends up exactly where you’d think, based on those findings. There’s no real parallelization, but there’s plenty of system memory and cache usage, which turns out to be a combination that puts its performance squarely in the middling category.

Ryzen’s 3D performance is okay, especially in light of the number of cores not being the priority here. Consequently, the results aren’t surprising.

The Cinebench R15 OpenGL benchmark gives us a preview of what we’re in for if an application has not been optimized for Ryzen. It brings up the rear in spite of noticeable multithreading. Then again, it’s also often more than 30% faster than its predecessor, which is running at 5GHz.

Important NoteMany of the professional applications in the development field have been optimized and compiled for Intel CPUs. This is reflected in the performance numbers. Still, we want to provide a complete performance picture, so we won’t skip them. The results might motivate the application developers to focus more of their efforts on AMD’s Ryzen and Naples so that their users have two options.

One of these professional applications is Solidworks by Dassault Systemès. It’s clear to see across the two composites’ performance numbers that both single- and multithreaded scenarios are affected to varying degrees, in spite of the rendering results having an influence on the overall results as well.

The same general principle applies to Creo 3.0 by PTC. Just like Solidworks, this application’s an important and de-facto standard tool for professional development.

Catia’s older, and it’s not optimized for multithreaded performance. This means that the CPUs are left with using just one single core, which results in a lack of performance for AMD’s newest offering.

Autodesk’s Maya 2013 has the same problem, since graphics output in real time isn’t among the tasks that lend themselves well to parallelization.

Parallelization vs. Single-Core Performance

We start with AMD’s very own favorite example that the company kept showing during the presentations: Cinebench R15. First, we need to note that the only way to get the following numbers is by disabling the AVX instruction set completely. This is to say that Cinebench might be a good example, but it can’t necessarily stand in for all that many other applications when it comes to the AMD Ryzen 7 1800X’s real-world performance. Nevertheless, these results provide us with a good basis of comparison for what’s to come later. Note: For all 3.8 GHz benchmarks we turned XFR off (AMD) and Turbo Core off (Intel). For the single core benchmark we also disabled SMT.

Using just one core at stock clocks, AMD’s Ryzen 7 1800X leads Intel’s Core i7-6900K. Looking at them at the same clock frequency, which we set at 3.8GHz for all tests, provides the best comparison, and here Ryzen 7 1800X @3.8 GHz is a tick slower than the Core i7-6900K @3.8 GHz.

Rendering

Next up, we have two different Blender benchmarks. First, all we do is generate a picture output with a sample size of 200 pixels. AMD’s Ryzen 7 1800X beats Intel’s Core i7-6900K at the same clock frequency, but falls behind at stock clock frequencies.

Then again, this type of rendering is just the final step in a daily workload, so we run the preceding steps in addition to the final rendering for our second benchmark. This includes preview pictures and content creation-related processes.

AMD’s Ryzen is unbeatable at pure rendering tasks, and it does a solid job otherwise.

It’s a neck-and-neck race between AMD’s Ryzen 7 1800X and Intel’s Core i7-6900K once again. At the same clock frequency, Ryzen pulls ahead when using the console version, which doesn’t include any workflow-related tasks, but just the graphics output.

As was the case with Blender, the 3ds Max benchmark distinguishes between several different areas, which don’t just include CPU rendering and, consequently, provide a good idea of what the work processes would feel like to the user. AMD’s Ryzen 7 is at a disadvantage when there’s real-time graphics output via the GPU or when only a few cores are used.

Bottom Line

There might have been a few hiccups here and there, which were usually due to the specific software, but, overall, AMD’s Ryzen 7 1800X seems surprisingly mature and offers solid performance. It’s questionable if it’s a good investment if the application in question has been heavily optimized for Intel processors, but there’s no real drawback to its use as a general-purpose CPU.

Results: Scientific & Engineering Computations And HPC

We’re using the SPECwpc benchmark suite, since it provides a good variety of workstation tasks that use very different types of mathematical computations. These are both highly parallelized and also make heavy use of memory bandwidth, cache, and latencies.

The pre-Euler3D CFD (Computational Fluid Dynamics Benchmark) test did work, but the rest of Rodinia’s benchmarks (i.e., Accelerating Compute-Intensive Applications) didn’t. We’re still showing all of the results below, since the remaining CPU’s results might still be of interest.

LAMMPS stands for Large-scale Atomic/Molecular Massively Parallel Simulator, and it didn’t just run flawlessly, but also really showed off Ryzen’s capabilities. In the resulting head-to-head race, AMD got there first more often than not.

NAMD is a benchmark for high-performance simulation of large biomolecular systems. All individual tests went Ryzen 7 1800X’s way.

The Fastest Fourier Transform in the West, or FFTW, is a popular open-source solution to compute one-, two- and three-dimensional DFTs (Discrete Fourier Transforms). The C library makes heavy use of single-precision AVX these days, which proves to be a terrible thing for the Ryzen 7 1800X.

Convolution is a benchmark for an area of functional analysis. Convolution stands for a mathematical operation on two functions (i.e., f and g) that produces a third function.

The CalculiX benchmark is based on a three-dimensional structural finite element program.

Poisson's Equation is a partial differential equation that allows the modeling of the number of incidences that occur at a constant medium rate independent of each other within a fixed time interval or area. It doesn’t prove to be a problem for AMD’s Ryzen 7 1800X, but the older AMD FX-9590 doesn’t fare well.

Sequential Reweighted Message Passing, or SRMP for short, is an algorithm used to solve discrete energy minimization problems. AMD’s Ryzen 7 1800X runs into trouble, which might be due to the specific benchmark application. It uses a total of eight threads.

The earth’s underground structure can be ascertained based on seismic processing. One of the four basic steps to do so is the Kirchhoff Migration, which is used to build a graphical representation out of the available data. AMD’s Ryzen 7 1800X does well with it.

GNU Octave is a scientific programming language. Octave Forge is a toolbox that’s an important part of it, since it allows additional functions to be added and the application’s functionality to be extended. AMD’s Ryzen 7 1800X does a solid job, but we do have to note that not all of the functions worked without problems and that we had to skip one chart graph altogether.

Bottom Line

Unfortunately, we just didn’t have time to try to find possible solutions to the problems that occurred with GNU Octave and some of the other benchmarks. Otherwise, we would have searched for the root of the problems or tested the software using a different operating system. Alas, two days just aren’t enough time to perform a comprehensive CPU test.

This is even more regrettable in light of the fact that AMD’s Ryzen generally does well with these types of tasks. It would have been nice to know the reasons behind the few times when it didn’t. Overall, we’re left with the impression that AMD’s boiled down a server CPU to the desktop level. This doesn’t have to be a bad thing, though, and it certainly makes us excited for Naples.

Results: Power Consumption And Temperatures

Power Consumption

We measure voltages and currents directly on the motherboard using the existing sensors and calculate the power consumption based on them. To achieve valid results, we take the average of the two-minute measurements for each of our scenarios and use our low-pass filter and analysis software to get rid of any extreme peaks or valleys.

These aggregated results are a lot more telling than extremely short peaks due to the latter’s very brief nature. We took a look at a total of eight different scenarios.

Some of the applications consume more power over time. This is due to the CPU heating up. We found that the difference between a cold and fully warmed up CPU can be up to 3W. If these 3W are interpreted as leakage currents, then this is actually a very good result.

We took the average of the power consumption curves after the CPU had reached its full operating temperature and put the results in a bar graph to provide a summary.

Temperatures

Keeping the processor cool is more important than ever due to AMD’s XFR and related technologies. Lower temperatures have the potential to translate to higher clock frequencies. This is why we used a custom water cooling loop, which was very much needed during the rendering benchmarks with all cores running at 3.8GHz.

The CPU diode that AMD uses should be closest to the Tpackage value, which is the part of the CPU that has to endure the most heat in the long run. Tcore doesn’t matter as much anymore due to the massive areas that the cache takes up these days.

Infrared Video and CPU Temperatures

We decided to make an infrared video and spare no expense. Shooting it also required a degree of bravery, since we had to take off the CPU cooler to record it. We installed the camera so that it was pointed downward directly facing the CPU’s heat spreader. The camera’s distance from the processor was a function of the camera’s focal length and efforts to avoid any reflections caused by the camera’s lens and the metallic surface of the heat spreader.

We also applied a total of six thin layers of special lacquer used in board manufacturing to the heat spreader. This is necessary because the pure metal’s temperatures can’t be measured accurately due to its emissivity changing along with its temperature. After the usual calibration, we recorded the following video.

It’s easy to locate the processor’s die this way. It appears to be long and narrow in shape. The heat spreader’s uniform heat distribution indicates that there’s sufficient contact between the die and the heat spreader, especially in light of the fact that we weren’t able to assert any pressure on the heat spreader while we were recording the video. We recently discovered that the die is soldered to the Integrated Heat Spreader (IHS), which is typically much more efficient than Intel's technique of using thermal paste.

AM4 Kit Incompatibility

One of the hardware components that’s seen some changes is the AM4 motherboard’s backplate. It’s different in two very significant ways: The distance between the holes has changed and the threaded pins are longer. Even though AMD seems to have informed manufacturers of the first change, they apparently forgot about the second one.

We asked the manufacturers, and they confirmed that this is what happened. This means that if the original motherboard backplate’s used in conjunction with longer screws, then the cooler might not be held against the processor tightly enough.

The solution involves adding the needed millimeter by using either thick ring washers or suitable nuts. We’ll keep in touch with the manufacturers, since the problem doesn’t seem to affect all of the kits.

Conclusion

AMD's Ryzen 7 launch represents more than just a new CPU family. For most of our readers, it signals the return of competition to the enthusiast-oriented processor market. And considering the flagship 1800X’s potent cost advantage compared to Intel's Core i7-6900K, the competitor AMD singled out months ago, Ryzen 7 does deliver. It's just not as universally superior as the company wanted everyone to believe.

We come away from today's coverage with a number of questions that couldn't be answered in time for the launch. For instance, we discovered Ryzen's tendency to perform better in games with SMT disabled. Could this be a scheduling issue that might be fixed later? AMD did respond to our concerns, reminding us that Ryzen's implementation is unique, meaning most game engines don't use if efficiently yet. Importantly, the company told us that it doesn’t believe the SMT hiccup occurs at the operating system level, so a software fix could fix performance issues in many titles. At least one game developer (Oxide) stepped forward to back those claims. However, you run the risk that other devs don't spend time updating existing titles.

The evening before launch, AMD sent us a list of games that it says should perform well with Ryzen, including Sniper Elite 4, Battlefield 1, Star Wars: Battlefront, and Overwatch, among others. Many of the titles tend to be heavily threaded, which would lend itself well to Ryzen's high core count. We plan on revisiting some of those. Further, AMD suggests adjusting several different parameters for games that suffer from low performance. It recommends using Windows' High Performance power profile (which also helps Intel CPUs). It also says to disable the HPET (High Precision Event Timer), either in your BIOS or operating system, to gain a 5-8% advantage. Our results already reflect HPET disabled, though. Interestingly, AMD's Ryzen Master software requires HPET to “provide accurate measurements,” so you may find yourself toggling back and forth for the best experience.

It’s hard to recommend the Ryzen 7 1800X over Intel's lower-cost quad-core chips for gaming, especially given the Core i7-7700K's impressive performance. That's not a knock against AMD, specifically. After all, we say the same thing about Intel's own Broadwell-E CPUs. High-end Kaby Lake processors constantly challenge pricier competitors, and the flagship -7700K sells for $350. Even after down-clocking the -7700K to 3.8 GHz, it still beats Ryzen 7 1800X in nearly every game in our suite. Those issues would only be exacerbated on a Ryzen 7 1700X, which operates at lower clock rates.

Conversely, the Ryzen 7 1800X is in its element when you throw professional and scientific workloads at it. It isn't the fastest in every high-end benchmark, but any calculation that factors in value almost assuredly goes AMD's way. For years, Intel has operated with impunity, charging inflated prices for incremental speed-ups. The 1800X’s $500 price tag and competitive performance will no doubt excite power users on a budget. To that end, when we weigh the 1800X’s strong showing in workstation and HPC workloads against its issues with games, we can't help but believe that AMD designed this specific configuration with a datacenter-driven mindset and didn’t optimize it thoroughly for desktops. Much like Intel and Broadwell-E, in fact.

AMD’s Precision Boost technology yields a nice dual-core boost during lightly threaded workloads, but it isn’t as advanced as Intel’s sophisticated multi-core Turbo Boost functionality. XFR is a nice feature that automatically offers improved performance with robust cooling solutions, but most of us only get 100 MHz out of it, so it's hard to call it a compelling advantage. Achieving a 4 GHz overclock was straightforward enough through multiplier and voltage adjustments, and there are plenty of AMD-specific firmware settings we need to explore. More headroom could certainly be available (though the Core i7-7700K is honestly more exciting to overclock if all you care about is higher numbers). On the memory overclocking side, AMD hasn’t opened all of the sub-timings yet, and the Core i7-6900K has a throughput advantage with its quad-channel controller.

Ryzen 7 1800X's aggressive price might help put enough pressure on Intel to compel price cuts on Broadwell-E, but the bigger battle is going to happen when Ryzen 5 and 3 emerge to challenge the competition's more affordable (and difficult to usurp) models. AMD is also bringing its Naples server CPUs forward soon, and with what we’ve seen from the Zen core, that should be an exciting launch.

It's a bummer the Ryzen launch was so clearly rushed. We expected AMD to have a better explanation for its gaming performance, but all of the feedback we received from the company came very last-minute. It's hard to imagine these shortcomings weren't discovered previously and diagnosed more thoroughly. We're happy to put in the time and effort, though. Expect more information as it becomes available.

In the meantime, we would recommend Ryzen 7 1800X for heavily-threaded workloads like rendering and content creation. And while we won't judge a processor on its gaming performance alone, current indications suggest AMD's $500 flagship doesn't beat Core i7-7700K for value in that specific segment.

MORE: Best CPUs

MORE: Intel & AMD Processor Hierarchy

MORE: Everything Zen: AMD Presents New Microarchitecture At HotChips

MORE: Intel Kaby Lake Core i7-7700K, i7-7700, i5-7600K, i5-7600 Review

MORE: Broadwell-E: Intel Core i7-6950X, 6900K, 6850K & 6800K Review

PresentMon And Our Proprietary Software

The introduction of DirectX 12, Microsoft's Universal Windows Platform, and Vulkan presented us with a problem: we no longer had a reliable tool for measuring performance. Fraps only works up through DirectX 11, and it's not always as reliable as some enthusiasts assume.

Consequently, the timing of Andrew Lauritzen's PresentMon tool, freely available on GitHub, was ideal. It monitors Windows' event tracing stack for present commands and records a bunch of information about them to a CSV file for analysis.

But PresentMon is subject to its own technical limitations, since it operates at the same place in the graphics pipeline as Fraps. Thus, it doesn't completely replace tools like FCAT in our repertoire. But what makes PresentMon so interesting to us is its compatibility with not only DirectX 11, but also DX 12, OpenGL, and Vulkan.

The tool's most significant disadvantage is its command-line interface. It can be time-consuming to figure out the right combination of switches to use in order to generate the information you need. By default, it's not set up to simply spit out the data you'd necessarily expect. And that's what we set out to solve.

Detection Made Easy: The PresentMon GUI

We made the decision to develop an app that would start PresentMon, control its switches, and collect sensor data at the same time as performance was being recorded.

In the screenshot above we have a number of frequently benchmarked games, each with its own profile stored in our tool. PresentMon uses the chosen parameters for each, depending on how it was set up. One test might run for three minutes and then stop, for example, while another is started and stopped with a hotkey.

Data Collection Made Easy

Of course, PresentMon isn't set up to collect sensor data as it records performance information. That capability comes from FinalWire's AIDA64. We tap into the engineering version, synchronously recording a number of host processing and graphics statistics. Do you want to correlate GPU utilization with frame rate at a certain point in a test? That's possible. Latency and overhead are kept to a minimum, as AIDA64's approach is different from tools like HWiNFO that use a DLL for their interface.

In order to minimize the impact of storage I/O, we write this second log file to memory first (it's small enough, after all), and then to disk once PresentMon's log file closes.

Since PresentMon doesn't use a time stamp, unfortunately, we put the beginning of our own records first, then PresentMon's log, and track time between them in parallel. This is necessary for controlling external measurements (like our oscilloscopes).

Analyzing The Data And Processing It For Presentation

All of our recordings result in an enormous pile of data for every benchmark run, which has to be sorted through. To help with this, we use another proprietary piece of software that combines the two log files and applies all of the math needed for our charts. It simply wouldn't be possible to do this manually in Excel due to the complexity and number of steps involved.

With our log file interpreter now in its third generation, we can run the calculations we need quickly, and even add new charts/graphs as needed. We'll get more in-depth on what we evaluate directly with an example using two different graphics cards and a DirectX 12 game.

Our purpose here is to help explain the information we're generating. Without a thorough understanding of what the charts are saying, it's sometimes possible to draw the wrong conclusion. Thus, you'll see us point to this piece in future reviews to ensure our processes and procedures are fully detailed.

The bar chart above conveys minimum, maximum, and average FPS. These are important, but they tell us nothing about the perceived smoothness of a gaming experience, nor do they give us a more granular look into frame consistency. In order to present that information, we need more data.

Two Cards, Two Test Systems, And One Benchmark

This example isn't intended to compare MSI's GeForce GTX 1060 Gaming X 6G to its Radeon RX 480 Gaming X 8G. Rather, we're using results from both cards to explain how we measure and evaluate.

Using a single benchmark simplifies the analysis. So we focus on Hitman, tested on both graphics cards using DirectX 11 and 12, and on two very different platforms.

MORE: Best Graphics Cards

MORE: Desktop GPU Performance Hierarchy Table

MORE: All Graphics Content

Frames Per Second: Bars Or Curves?

Frames Per Second

In the end, time-based averages only tell you how many frames were rendered in a given second. There's no way for you to know how well-paced those frames were, or if a long pause interrupted the action at one point, negatively affecting the experience. After all, a one-second interval with lots of fast frames and a single frame that took 100 ms to display is going to "feel" less smooth than the same interval with slower, consistently-rendered frames, even if their averages appear similar.

The following two bar charts correspond to our enthusiast and mainstream PCs. Each includes minimum, average, and maximum frame rates. They're all rough indicators of performance, mostly useful for comparing many cards at a glance. But even then, all three figures are commonly maligned for over-simplifying the real-world gaming experience.

Charting frame rate over time shows instantaneous performance at any point in our benchmark run. These charts are more interesting than the simple bars, though they don't quantify a test's outcome as succinctly. Still, it's pretty obvious which configurations fare better in this type of comparison. 

For 108 seconds of our test run, we have 108 points on our X axis with the corresponding frame rate at each second on the Y axis. But we still know nothing about dropped frames or micro-stutter within each one-second interval. These are still rough averages.

Properly representing and interpreting the rendering times is so important because the two FPS indicators already discussed, whether they're bars or lines, tell us nothing about how a game actually "feels." Of course, we want to retain those averages, peaks, and floors. But it's important to add that sub-one-second dimension as well.

Frame Times And Rendered Frames

Presenting the render time of each individual frame as a line chart looks simple enough. However, it gets a lot more complicated when you start comparing cards at different performance levels, which generate different numbers of frames during a defined benchmark interval. For instance, in our DirectX 11 test on the enthusiast system, MSI's Radeon RX 480 Gaming X 8G creates 8090 frames. In the DirectX 12 benchmark, it outputs 8446 frames. Meanwhile, the GeForce GTX 1060 Gaming X 6G pushes 7362 and 7332 frames, respectively.

Ideally, we'd want to compare the render times of individual frames in order to better understand the graphical flow of each card. But we can't simply overlap variable-length recordings on a common horizontal axis like we do in the FPS graphs, which are based on 108 data points in our example.

For the two test systems, starting with the faster one, we end up with this:

On both graphs, the Y axes scale based on the render time results we measured. We do this to show as much chart detail as possible; simply keep it in mind to avoid comparing the left and right diagrams at a glance. Though it'd certainly be possible to extend the left graph's Y axis up to a 160-second maximum, doing so would sacrifice resolution.

So now we know what it looks like when you compare the frame-by-frame histories of two GPUs on a single graph's X axis, even when their outputs differ. Excel can't do this on its own though, so we use our software to create optimized curves of equal length for each card. But how do you get 8446 or 7362 individual data points onto an axis able to fit, say, 1000 values?

We preserve the benchmark run's peaks and values exactly as they'd appear in a graph with however many thousands of points were actually captured as losslessly as possible. Everything is interpolated cleanly, so that the visual representation matches exactly.

MORE: Best Graphics Cards

MORE: Desktop GPU Performance Hierarchy Table

MORE: All Graphics Content

Performance Versus Smoothness

Quantifying Experience In The Game World

Moving beyond the idea of average performance, or even performance over time, we break apart the smallest unit of measure, the second, to evaluate frame pacing, consistency, and stutter, all of which affect your perception of smoothness. Obviously, we want to see all of the frames that go into a frame-per-second average to render at even intervals, yielding the most pleasant pacing possible for a given performance level.

But that's not always what we observe. Sometimes high frame rates are accompanied by problematic stuttering. And that's why we have to evaluate performance separately from smoothness, even though they're inherently related.

A Simple Performance Analysis

First, let's compare two cards under DirectX 11 and 12. In the following charts, frame times are on the X axis and percentage points are on the Y axis. Ideally, you want to see the numbers to the left (shorter frame times) represented by the highest percentage possible. Anything above 33 ms drops below the 30 FPS mark. That's just ugly, so the far-right line should be as short as possible.

Interestingly, we see DirectX 12 shift the Radeon's line to the left on the enthusiast and mainstream PCs, while Nvidia isn't affected as profoundly.

A Simple Smoothness Analysis

To achieve suspension of disbelief and really draw you in to the gaming experience, individual frames must be delivered as smoothly as possible with minimal differences between render times. Swings as small as 10 to 20 milliseconds are perceived by our brains as micro-stutter, and these interruptions negatively affect our gaming experience. Of course, some of us are more sensitive to this phenomenon than others.

Again, we're looking at percentage on the Y axis. But instead of raw frame times on the X axis, we're looking at frame to frame differences up to 10 ms. Above that threshold, we'll have to zoom out for more detail. For now, though, those spikes at the end of the chart are large enough to tell us we aren't getting as smooth of an experience as we'd want on either system, using DirectX 11 or 12. Ideally, we'd want to see the whole line shifted as far left as possible. The RX 480 in our high-end config comes closest to achieving this under DirectX 11. Its behavior under DX 12 isn't as compelling.

The preceding charts alluded to what gets spelled out more explicitly in these graphics: despite overall higher performance (especially from the mainstream FX-based machine), our DirectX 12 runs suffer from greater frame time differences than DX 11, which appeared smoother, at least according to our benchmark results. It's time to go deeper...

Frame Rate Versus Frame Time Difference

As they say, many roads lead to Rome. First, we interpolate the frame rate curve to match the frame time output, allowing us to compare them directly. But the differences in render times are not a simple subtraction problem between frames. Rather, we perform a more complex calculation that reveals values most likely to affect your level of immersion.

Let us first consider the faster Core i7-based machine:

Talk about an illustrative example. We see very clearly that higher frame rates don't necessarily correspond to a more immersive experience. Particularly on the Radeon, spikes up above 100 ms aren't just micro-stutter; those are significant hitches in the action, and you're definitely going to notice them. Nvidia isn't immune either. Its GTX 1060 also suffers a higher frequency of pauses shifting from DX 11 to 12.

Frame rates on the slower system improve more dramatically under DX 12, particularly with AMD's Radeon RX 480 installed. There's an obvious trade-off in the form of disturbing frame time variance, though.

DirectX 12 only really helps the GeForce GTX 1060 on our lower-end platform. The Nvidia card's peaks and valleys aren't as pronounced, and its frame time variance over time isn't as prone to wild swings.

The Incorruptible "Stuttering Index"

At first, we split the run into one-second intervals and calculate the frame rate of each one. This is the basis for our index listings. Each interval gets a value based on its FPS result. Anything under 30 FPS is unplayable, between 30 and 60 FPS ranges from playable, good, and very good. The index goes lower the faster the frame is rendered.

This is still too rough, though. A brief burst of slow frames may show up within a smooth 70 FPS interval. You wouldn't know it from the frame rate, but you'll definitely see it during real-world gaming. Therefore, we explore the render times of individual frames and the differences between respective frames, too.

To prevent accidental misinterpretations, we use an intelligent filter that catches transitions between the cut scenes you often see in built-in benchmarks. If one sequence is not complex and runs faster, and the second is more challenging, slowing performance, this artifact is filtered out (frame preview, block-wise comparison); it's not real stuttering, but rather a scene change.

In this way, a fairly accurate forecast can be made whether stuttering or dropped frames are visually perceptible to the gamer. If the score for a single frame is higher (that means worse) than the base FPS value, the whole interval is marked with the higher/worse index value.

With all of those calculations complete, we end up with a subjective integer-based "Stuttering/Uneveness index," free from fractional values. This rating ranges from a score of zero (perfect, no interfering influences) to a score of five (the limit of acceptance) up to level 10 (real stuttering and dropped frames). The most sensitive enthusiasts will perceive micro-stuttering at a score of three or four.

In the chart above, a faster CPU helps maintain fairly smooth playback, though the Radeon RX 480 struggles under DirectX 12 with some visible stuttering.

This becomes even clearer under the power of a lower-end CPU. Although the GeForce isn't as fast, it does facilitate a smoother-looking picture.

MORE: Best Graphics Cards

MORE: Desktop GPU Performance Hierarchy Table

MORE: All Graphics Content

Even More Data To Evaluate

FPS Vs. CPU Utilization

It can be difficult to identify bottlenecks or single-out the advantages/disadvantages of different APIs using a Fraps or PresentMon capture on its own. Adding the sensor data our software captures currently changes that, though.

First, let's consider the utilization of two CPUs, starting with the faster one. We can clearly see performance isn't being held back by any kind of host processing bottleneck.

The picture changes dramatically as we drop our cards into a lower-end platform. It's easy to see when and where graphics performance is limited by a CPU bottleneck. In addition, you can see how the load drops under DirectX 12 while using the Radeon RX 480, while the opposite appears true with Nvidia's GeForce GTX 1060.

FPS Vs. Power Consumption

A drop in CPU load is expressed partly by lower power consumption. We can scrutinize this more closely by testing DirectX 11 against DX 12 on our two systems, each with two graphics cards.

What we find is that the CPU's power consumption corresponds more or less to the CPU load.

This effect is also observable on the weaker CPU, though it's not quite as pronounced.

Attentive readers may have noticed that our recently-updated Nvidia Titan X Pascal 12GB Review featured improved charts, including temperature history. The data for this came from our new tool, replacing third-party software we were using previously.

Conclusion

We have seen that pure bar graphs are well and good, but they're certainly not the last word when it comes to evaluating graphics performance. There's a ton of additional analysis that can go into reviewing video cards, and we're elbow-deep. The emphasis at this point is presenting the most valuable data, without overwhelming anyone, in a way that makes sense.

Our advantage as editors is that we always see the full data dump. So as it's sorted through, potential problem areas become clear and we can prioritize certain performance passages or sensor read-outs. Moving forward, we'll handle all of this on a case-by-case basis to ensure you get the information you need to guide your purchasing decision.

A special thanks to all of the readers who helped shape our direction. Whether it was emails, comment section posts, or group discussions on other forums, we listened every step of the way. This is actually where the idea for our stuttering index came from.

In the end, it's all a neat balancing act between information that enthusiasts can use and illustrations easier for the masses to digest. One thing's for sure, though: we're sure there's something here for everyone.

MORE: Best Deals

MORE: Hot Bargains @PurchDeals

Follow us on Facebook, Google+, RSS, Twitter and YouTube.

Chinese forum Benyouhui.it168 has posted some initial benchmarks of Intel’s upcoming flagship mobile processor, the 22 nm quad-core Core i7-4930MX.

The CPU features a base clock of 3.0 GHz, a boost clock of 3.7 GHz, an 8 MB L3 Cache, 57 W TDP and supports DDR3 and DDR3-L memory clocked at up to 1600 MHz. The i7-4930MX also includes the Intel HD 4600 iGP which features a 400 MHz core clock, 1350 MHz memory clock, 20 execution units with two ROPs, four TMUs and a 128-bit DDR3 memory interface. 

In addition to evidently allowing games such as Battlefield 3, Need for Speed: Most Wanted (2012) to be playable at 1366 x 768, the i7-4930MX also scored the following in AIDA 64 and Futuremark’s 3DMark 11:

AIDA 64 Extreme Edition

• CPU AES - 16,492 MB/s
• CPU ZLiB - 309.1 MB/s
• CPU Hash - 3014 MB/s
• CPU PhotoWorxx - 14,315 Mpixel/s

3DMark 11

• Performance Preset: P1418
• Extreme Preset: X374 

It’s worth noting that Intel’s upcoming “Iris Pro” graphics seems set to provide even more processing power that certainly has the potential to allow systems to offer respectable gaming performance without needing a discrete graphics card.