AMD's Kabini: Jaguar And GCN Come Together In A 15 W APU showed us what the company's Jaguar and GCN architectures could accomplish between 4 and 25 W TDPs. But, on the desktop, AMD isn't quite ready to make the leap to a next-gen design. It just introduced its desktop-oriented Richland APUs, which aren't really new at all. Rather, you can think of them as power-optimized Trinity parts, sporting the same Piledriver-based x86 cores and VLIW4 graphics configuration. Moreover, Richland-based APUs have been available to mobile device makers for months. The only real revelation is that we're getting this update in the desktop and low-voltage mobile spaces now.
| Model | Radeon | Package | TDP | CPU Cores | Base/Max CPU Clock | L2 Cache | Radeon Cores | Base GPU Clock |
|---|---|---|---|---|---|---|---|---|
| A-Series Low-Voltage and Ultra Low-Voltage APUs | ||||||||
| A10-5745M | HD 8610G | FP2 | 25 W | 4 | 2.1/2.9 GHz | 4 MB | 384 | 533 MHz |
| A8-5545M | HD 8510G | FP2 | 19 W | 4 | 1.7/2.7 GHz | 4 MB | 384 | 450 MHz |
| A6-5345M | HD 8410G | FP2 | 17 W | 2 | 2.2/2.8 GHz | 1 MB | 192 | 450 MHz |
| A4-5145M | HD 8310G | FP2 | 17 W | 2 | 2.0/2.6 GHz | 1 MB | 128 | 424 MHz |
In the table above, we see the new mobile-oriented options spanning 17 to 25 W TDPs. Richland isn’t much different from Trinity, but it's more efficient thanks to specific Turbo Core optimizations that include a greater number of P-states to facilitate more granular power and performance levels.
| Model | Radeon | TDP | CPU Cores | Base/Max CPU Clock | Total Cache | Radeon Cores | GPU Clock | Unlock | Price |
|---|---|---|---|---|---|---|---|---|---|
| A10-6800K | HD 8670D | 100 W | 4 | 4.1/4.4 GHz | 4 MB | 384 | 844 MHz | Yes | $149 |
| A10-6700 | HD 8670D | 65 W | 4 | 3.7/4.3 GHz | 4 MB | 384 | 844 MHz | No | $149 |
| A8-6600K | HD 8570D | 100 W | 4 | 3.9/4.2 GHz | 4 MB | 256 | 844 MHz | Yes | $119 |
| A8-6500 | HD 8570D | 65 W | 4 | 3.5/4.1 GHz | 4 MB | 256 | 800 MHz | No | $119 |
| A6-6400K | HD 8470D | 65 W | 2 | 3.9/4.1 GHz | 1 MB | 192 | 800 MHz | Yes | $77 |
And then we have the desktop-specific Richland parts. The very fastest model enjoys a 300 MHz base clock rate bump compared to the A10-5800K, along with official support for 2133 MT/s DDR3 memory (the other SKUs top out at 1866 MT/s memory). Also, its GPU is 44 MHz faster than the prior-gen version. And yet it fits within the same 100 W TDP.
On the other hand, the A10-6700 looks a lot like the -5800K, aside from a 100 MHz-lower base clock, a 100 MHz-higher Turbo Core ceiling, and a slightly quicker GPU. That one drops to a 65 W thermal limit.
Like the Trinity-based APUs before them, these Richland designs plug into a Socket FM2 interface. A BIOS update should be all that you need for compatibility with existing A55, A75, and A85 platforms.

And what about those shiny new Radeon model numbers? That's marketing being bad. We were already dealing with Radeon HD 7000-series naming on APUs, which was confusing because AMD's desktop 7000-series GPUs are commonly associated with the GCN architecture. Now we have 8000-series nomenclature. And yet, we're still working with the VLIW4 configuration that was around back when AMD was shipping Radeon HD 6900-series cards. Architecturally, the only difference between Richland's Radeon HD 8000 graphics and Trinity's Radeon HD 7000 graphics is the name.
Again, the top-end A10-6800K gets 2133 MT/s memory support, which is particularly meaningful for its on-die graphics engine given a lack of shared L3 cache. The theoretical 34 GB/s of DDR3 bandwidth should go a long way to improve frame rates in the games we'll be testing. With that said, let's move on to why we aren't able to test Richland's Dual Graphics feature today...
AMD’s Dual Graphics technology, once referred to as Hybrid CrossFire, allows the APU to work cooperatively with a discrete graphics card to deliver higher game frame rates than either component flying solo. At first glance, this seems like a great way to extract value, offering an upgrade path not available on any competing platform.

Unfortunately, there are limitations. First, this is a software-based capability that only works with DirectX 10 and 11 game engines. Second, the APU is quickly outclassed by most discrete cards. So it really only works with Radeon HD 6450, 6570, and 6670 boards, maintaining balance between the two parts.
We've long since wanted to dig deeper into Dual Graphics; after all, as you can see in the screen capture above, AMD claims it serves up a serious performance increase. But in our subjective experience, this feature does not necessarily appear to yield smoother game play. For the company's Richland introduction, we were excited to finally test Dual Graphics using our FCAT tools, capturing the raw display output and analyzing the stream to determine if frames were being dropped entirely or rendered in a series of full and tiny (runt) frames.

Unfortunately, our video-based analysis turned up an unexpected issue that prevents us from reporting the performance of Dual Graphics. Intermittently, we'd see a frame rendered, followed by a piece of the following frame, a piece of the original frame, and the rest of the following frame. This artifact is accompanied by a tear across the screen as Dual Graphics puts the two frames together.
It's consequently impossible to run our FCAT analysis on the output, since the frame sequence can't be measured. Fraps-based testing in this case would clearly be inaccurate. So, it's better to hold off on trying to quantify the performance of Dual Graphics until AMD can provide a solution that composites the frames free from artifacts or tears. The company is aware of our findings and is working to address them. As of yet, though, we don't have an explanation of why this is happening.
We're testing the Richland-based A10-6700 and A10-6800K APUs against their predecessor, AMD's Trinity-based A10-5800K. We have an Ivy Bridge-based Core i3-3220 in our Canadian lab with HD Graphics 2500. Unfortunately, we didn't have an opportunity to snag a Core i3-3225 before leaving for Computex, so the comparison to HD Graphics 4000 will have to wait.
The good news is our testing facility in Hillsboro, Oregon already ran the numbers comparing both Core i3 chips and A10-5800K. The pertinent data can be found in Gaming At 1920x1080: AMD's Trinity Takes On Intel HD Graphics. For the time being, Core i3-3220 reflects the x86 performance of both Intel chips in today's tests, and in the same $130-150 price range these new Richland-based APUs are expected to sell for.
In order to maximize performance in our game tests, we used DDR3-1866 and -2133 data rates, which are the highest officially-supported settings on AMD's new APUs. In addition, we included results with a discrete Radeon HD 6670 DDR3, a baseline mainstream gaming card.
| Socket FM2 | LGA 1155 | ||||
|---|---|---|---|---|---|
| CPU | AMD A10-6800K (Richland) 4.1 GHz Base, 4.4 GHz Turbo Core w/ Radeon HD 8670D (844 MHz) AMD A10-6700 (Richland) 3.7 GHz Base, 4.3 GHz Turbo Core w/ Radeon HD 8670D (844 MHz) AMD A10-5800K (Trinity) 3.8 GHz Base, 4.2 GHz Turbo Core w/ Radeon HD 7660D (800 MHz) | Intel Core i3-3220 (Ivy Bridge), 3.3 GHz, Hyper-Threading enabled w/ Intel HD 2500 | |||
| Motherboard | ASRock FM2A85X Socket FM2, Chipset: AMD A85 | Asus P8Z77-V LX LGA 1155, Chipset: Intel Z77M | |||
| Networking | On-Board Gigabit LAN controller | ||||
| Memory | AMD Gamer Series Memory, 2 x 4 GB, 1866 MT/s, CL 9-9-9-34-2T Overclocked: 2133 MT/s, CL 10-11-11-28 | ||||
| Graphics | AMD Radeon HD 6670 DDR3 800 MHz GPU, 1 GB GDDR5 at 800 MHz (1600 MHz effective) | ||||
| Hard Drive | Western Digital Caviar Black 750 GB 7,200 RPM, 32 MB Cache, SATA 3Gb/s | ||||
| Power | ePower EP-1200E10-T2 1,200 W ATX12V, EPS12V | ||||
| Software and Drivers | |||||
| Operating System | Microsoft Windows 8 Pro x64 | ||||
| DirectX | DirectX 11 | ||||
| Graphics Drivers | AMD Catalyst 13.6 Beta; Intel HD Graphics Driver 15.31.3.64.3071 | ||||
And here are the benchmark details:
| Benchmark Configuration | |
|---|---|
| 3D Games | |
| Metro: Last Light | Version 1.0.0.0, DirectX 10, Built-in Benchmark |
| The Elder Scrolls V: Skyrim | Version 1.6.89.06, Version 1.5.26.05, 25-Sec. Fraps |
| Tomb Raider | Version 1.04, Custom THG Benchmark, 60-Sec. Fraps |
| F1 2012 | Version 1.2, Direct X 11, Built-in Benchmark, 60-Sec. Fraps |
| Audio/Video Encoding | |
| HandBrake CLI | Version: 0.98, Video: Video from Canon EOS 7D (1920x1080, 25 frames) 1 Minutes 22 Seconds, Audio: PCM-S16, 48,000 Hz, Two-Channel, to Video: AVC1 Audio: AAC (High Profile) |
| iTunes | Version 10.4.1.10 x64: Audio CD (Terminator II SE), 53 minutes, default AAC format |
| Lame MP3 | Version 3.98.3: Audio CD "Terminator II SE", 53 min, convert WAV to MP3 audio format, Command: -b 160 --nores (160 Kb/s) |
| TotalCode Studio 2.5 | Version: 2.5.0.10677, MPEG-2 to H.264, MainConcept H.264/AVC Codec, 28 sec HDTV 1920x1080 (MPEG-2), Audio:MPEG2 (44.1 kHz, Two-Channel, 16-Bit, 224 Kb/s) Codec: H.264 Pro, Mode: PAL 50i (25 FPS), Profile: H.264 BD HDMV |
| Productivity | |
| ABBYY FineReader | Version 10.0.102.82: Read PDF save to Doc, Source: Political Economy (J. Broadhurst 1842) 111 Pages |
| Adobe Photoshop CS6 | Version 13 x64: Filter 15.7 MB TIF Image: Radial Blur, Shape Blur, Median, Polar Coordinates |
| Autodesk 3ds Max 2013 | Version 14.0 x64: Space Flyby Mentalray, 248 Frames, 1440x1080 |
| 7-Zip | Version 9.28, LZMA2, Syntax "a -t7z -r -m0=LZMA2 -mx=5" Benchmark: THG-Workload-2012 |
| WinRAR | Version 4.2, RAR, Syntax "winrar a -r -m3" Benchmark: THG-Workload-2012 |
| WinZip | Version 17.0 Pro, Best Method, ZIPX Benchmark: THG-Workload-2012 |
| Synthetic Benchmarks and Settings | |
| 3DMark 11 | Version: 1.0.1, Entry, Performance, Extreme Suite |
| PCMark 7 | Version: 1.0.4, System, Productivity, Hard Disk Drive benchmarks |
| SiSoftware Sandra 2012 | Version: 2012 SP5c-1872, CPU Test = CPU Arithmetic / MultiMedia, Memory Test = Bandwidth Benchmark |

Sporting the same number of Radeon cores (384), it's hardly surprising that all of the A10s score fairly similarly. A slight 44 MHz clock rate increase gives the Richland-based parts an imperceptible speed-up. The Core i3-3220's HD Graphics 2500 engine is ill-equipped to fight off any of AMD's APUs, though we know full-well that HD Graphics 4000 and 4600 do narrow the gap.
The Physics module is a pure measure of x86 processing performance though, and in this one, the dual-core Core i3 finishes in first place (albeit just barely).

Although PCMark 8 launched a couple of days ago, AMD wasn't able to get us the test before we needed to board planes and head to Taipei for Computex. At least for one last processor launch, PCMark 7 will have to do. We don't see this as a problem; the benchmark is build using components of Windows 7, which many enthusiasts continues to use, and is representative of many common desktop workloads.
It's not much of a surprise why AMD doesn't care for PCMark 7, though. The Core i3's two Hyper-Threaded cores outperform AMD's two Piledriver modules in the Overall, Creativity, and Productivity subtests.

Intel's IPC advantage gives Core i3 the lead in Cinebench's single-core test. AMD comes surging back in the threaded component of this test, though, as its four integer pipelines outperform SMT technology.

Intel's Core i3-3220 returns higher integer and floating-point results in Sandra's Arithmetic module, though the A10-6800K isn't far behind in either metric.

Intel has a bad habit of using important features to differentiate its processors. The Core i3s, for example, arbitrarily lose AES-NI support. So, the Core i3 performs dismally in this measurement. Meanwhile, AMD's APUs process instructions as fast as they can be fed from memory, resulting in the great AES256 numbers.

This is an interesting result. Intel's two x86 cores deliver the top OpenCL-based result in LuxMark. All three Radeon-based graphics engines fare roughly the same when we limit processing to the GPUs. And the A10-6800K scores a first-place finish with both the CPU and GPU working cooperatively.
Remember, though that we're only testing Intel's HD Graphics 2500 engine. If you want some interesting data from HD Graphics 4000, check out this page in The Core i7-4770K Review: Haswell Is Faster; Desktop Enthusiasts Yawn. The A10-5800K's numbers match almost identically. But HD Graphics 4000 and 4600 kick performance up significantly, suggesting that a Core i3-3225 would make a big difference in this test.

As far as average frame rates go, All of the APUs are playable at 1920x1080 using Medium quality settings, maintaining at least 30 FPS. As we might have guessed, using system memory running at 2133 MT/s makes an appreciable difference, slightly outpacing the same APU complemented by a discrete Radeon HD 6670.
We knew going in that HD Graphics 2500 would have been unplayable. Intel figured the same thing out quite a while after introducing its modest Core i3-3220; the -3225 didn't become available until quite a while after. The company redeems itself, however, when we drop in the same Radeon HD 6670, though. Alleviating the A10's processor bottleneck allows AMD's add-in card to achieve much higher frame rates.

Charting frame rate over time gives us a clearer look at the peaks and valleys of each solution, though we don't derive any additional insight.

All of our tested configurations exhibit fairly low variance in the time it takes to render one frame to the next. Even when frame rates get slow, then, at least you have this consistency to look forward to.

No surprise; Intel's HD Graphics 2500 implementation isn't cut out for gaming. For that, you'd want to look to HD Graphics 4000, at least.
But even then, data from Chris Angelini's review of Core i7-4770K shows that Intel cannot keep up with AMD's APUs in The Elder Scrolls V: Skyrim. It's not even close. HD Graphics 4600 doesn't even do the trick.
Faster memory isn't a huge deal for the AMD parts in this title, though we will note that the 2133 MT/s kit helps A10-6800K outperform Intel's Core i3 complemented by discrete graphics.

The frame rate over time chart shows how 2133 MT/s memory helps push the A10-6800K to the top during this benchmark run.

Unfortunately, playable frame rates don't guarantee smooth performance. When we look at worst-case scenarios, each one of these solutions incurs more than 10 ms of variance between frames. That's definitely a difference you're going to feel as you're playing. And indeed, our experience in Skyrim reflected relatively moderate consistency in how frames are delivered.
While we're able to run Tomb Raider at 1920x1080 using on-die graphics, we had to drop the quality settings as low as they go (except for keeping FXAA and 8x AF turned on).

Again, we see that HD Graphics 2500 aren't viable in a 3D workload like this. Also again, there's plenty of data in Chris' Core i7-4770K review to show that neither HD Graphics 4000 or HD Graphics 4600 can overtake A10-5800K. So, A10-6800K is probably safe unless Intel decides to enable its GT3/GT3e configuration on the desktop. The company currently has no plans to do this.

Mapping frame rates out over time simply gives us a little extra detail.

Although the variance between subsequent frames appears low in general, we found this title to suffer from noticeable stuttering. Perhaps it was simply a result of lower overall frame rates giving the perception of less-smooth performance.
Metro: Last Light is easily the most demanding title in our suite, requiring us to drop the graphics details to their lowest settings and dial the resolution back to 1280x720. Fortunately, the game still looks good at those settings.

A demanding benchmark sequence pushes frame rates under 20 at times, despite decent averages from the A10 APUs. The discrete Radeon HD 6670 doesn't do any better, and Intel's HD Graphics 2500 isn't viable at all.

As you can see, the APUs manage to maintain frame rates in excess of 30 for much of the test, but are pushed under 20 FPS in several cases. It's tempting to say that you'll see better real-world performance from this title, though we know taxing sequences like the built-in benchmark are when gamers most commonly decry the inability of their hardware to perform.

Variance between subsequent frames isn't bad, but we simply cannot get around the fact that the experience in this title suffers from frame rates that are simply low.
It's clear from the pages of benchmarks we just ran that Richland's graphics subsystem has little trouble making quick work of Intel's HD Graphics 2500 engine. And we can be fairly certain from Angelini's evaluation of the Core i7-4770K that AMD's A10-6800K is going to be faster than HD Graphics 4000 and even HD Graphics 4600, too. But what happens when we shift away from the 3D applications and focus more intently on x86 apps? We'll start with a couple of single-threaded audio encoding titles.

iTunes is notoriously single-threaded, giving Intel's Ivy Bridge architecture a notable advantage. The A10-6800K manages to improve on AMD's A10-5800K, but it can't come close to the Core i3.

The same applies to LAME, another single-threaded audio encoding title.

HandBrake exploits multi-core architectures well, yielding an advantage to AMD's Piledriver design. Two Hyper-Threaded Ivy Bridge cores simply cannot keep up.

As the Lame results mirror iTunes, so does TotalCode Studio emulate the outcome of HandBrake. In general, software-based video encoding workloads favor the CPUs with more cores to work in parallel. It's true that AMD's Piledriver modules do share certain resources. However, Hyper-Threading isn't enough to keep Intel's efficient dual-core configuration ahead.
We should also note that these are software-based video encoders. A build of HandBrake recently emerged with optimizations for OpenCL and Intel's Quick Sync technology. Factoring in hardware acceleration affects performance and quality, so keep that in mind when you start looking at encode jobs that exploit other resources.

After Effects tends to demonstrate sensitivity to available memory per core, so perhaps we should have expected the dual-core Core i3 to establish a lead early on.

Premiere Pro CS6 scales much more reliably according to core count. Although AMD's cores don't get as much done per clock cycle, the fact that there are four of them earn the A10-6800K a first-place finish in this benchmark.

Remember, we have two distinct Photoshop benchmarks. The CPU test consists of several threaded filters that tax each CPU's x86 cores. The OpenCL benchmark uses filters that leverage CPU and GPU resources.
In the CPU-heavy test, Intel's two Hyper-Threaded cores trail the three AMD APUs we're testing, all of which have a pair of Piledriver modules. Conversely, the combination of HD Graphics 2500 and x86 cores puts the Core i3-3220 ahead in our OpenCL-based metric. AMD's processors aren't far behind, though.
Our Acrobat benchmark involves exporting a PowerPoint presentation to PDF format. It's the last single-threaded test in our suite.

Not surprisingly, the Core i3 ends up on top. The APUs aren't far behind, at least.

On the other hand, a well-threaded application like 3ds Max gives each A10 APU an opportunity to stretch its legs. The four integer cores you get from two Piledriver modules outmode the best efforts of two Hyper-Threaded Ivy Bridge cores.

Blender puts the Core i3-3220 and A10-6800K on equal footing, with the other APUs slightly behind.

FineReader is also well-threaded, which is why AMD's quad-core APUs manage to eke out a win, despite lower per-core performance.

Despite the threaded nature of our Visual Studio test, Intel's Core i3-3220 scores a first-place finish. The two APUs finish a couple of minutes behind. Clearly, something else is bottlenecking AMD's processors.

Given the differences between platforms up until now, it's unexpected to see the A10s separated by almost 30 seconds. However, it appears that AMD's latest APU matches the Core i3-3220.

Only a few seconds of difference distinguish the contenders in 7-Zip, though the A10-6800K claims a technical victory in this benchmark.

Intel's Core i3-3220 wins in our CPU-oriented compression test. Enabling OpenCL hands the win to AMD's A10-6800K. And the EZ test, which maximizes compression, also favors the Richland-based chip.
Although power consumption has less of an effect in the desktop space compared to notebooks, it remains an important point of comparison between Intel's Core i3 and AMD's top-end APUs. Can Richland's efficiency-oriented optimizations help close in on the efficient Ivy Bridge architecture?


The A10-6700 exhibits an impressive 25 W drop compared to AMD's A10-5800K in our Metro: Last Light benchmark. But that's hardly an achievement next to the 61 W Intel's Core i3 uses in the same test.
Even still, that's a tough comparison to make. AMD averages 30 FPS, while the Core i3 achieves less than half of that. It'd be a lot more interesting to substitute in a more capable CPU, though Intel's Core i3-3225 includes HD Graphics 4000 and is also rated for a 55 W thermal ceiling. Looks like the A10s are destined to be more power-hungry given their higher TDPs. The 65 W A10-6700 just can't come anywhere close to the 55 W Core i3.


When you aren't taxing their graphics components, AMD's APUs naturally use a lot less power. Of course, so does Intel's Core i3. The difference narrows in our Web browsing workload, though. Notably, the Richland configurations reduce power consumption by about 10 W compared to Trinity.


Again, the Richland-based parts are almost 10 W under the Trinity-based A10-5800K in our video playback test. Meanwhile, Intel's Core i3 is decisively in the lead here.
The following chart reflects aggregate performance of the four CPUs we tested. The red bar is the average of all benchmark performance categories, and the aquamarine bar gives us average power efficiency.

If you took the time to flip through our individual benchmark results, none of these results should surprise you. The A10-6700 offers similar performance as a stock A10-5800K, but offers greater efficiency. Unfortunately, it's also multiplier-locked. I have to believe that if you're willing to spend $150 on a 65 W A10 that can't be overclocked easily, then you're probably better off with a 55 W Core i3 that's also stuck in place for $10 less.
How about the unlocked models? AMD's A10-6800K is slightly faster than the -5800K it succeeds, roughly matching the Core i3-3220 in applications, while killing its HD Graphics 2500 engine in games. Based on performance data generated in The Core i7-4770K Review: Haswell Is Faster; Desktop Enthusiasts Yawn, we also know that an A10-5800K's graphics component is faster than HD Graphics 4000 and HD Graphics 4600. Unless Intel finds a way to get its higher-end graphics configurations on desktop processors, AMD's APUs maintain their top position for mainstream gaming.
The Intel chip's performance in single-threaded apps is exceptional. It holds its own in more parallelized workloads. And it offers the best efficiency, without question.
At the end of the day, AMD's Richland design is an iterative evolution of the Trinity architecture. It's not a game-changer. The only real surprise in this story is that, complemented by 2133 MT/s memory, the A10-6800K manages to elbow past the same processor with a discrete Radeon HD 6670 DDR3 installed. That's what we consider an entry-level gaming card, so the APU's achievement is a significant one. A quick price check shows that dual-channel, 8 GB, 1866 and 2133 MT/s kits start around $70. Armed with that, you can get low-end game performance without spending more on a graphics card. Consider this a landmark of sorts; a number of games are playable at reduced details using Richland-based parts, just as they were with Trinity.
Of course, when you have a little more money to spend, we also see that the Core i3's x86 cores allow even cheap graphics cards to performance closer to their peak potential without imposing a bottleneck. As far as gaming goes, then, the APUs fill a fairly narrow niche, above which you're better off with an add-in GPU and fewer restrictions on the settings you can use.
Calling Richland a stopgap between Trinity and the next-gen Kaveri would be just about right. From what we hear, Kaveri should still land sometime late this year. So, although Intel took a big step in catching up to AMD with its HD Graphics 4600 engine, the integration of GCN should help AMD widen that gap again. Steamroller-derived x86 should help as well, as should the implementation of hUMA (heterogeneous unified memory architecture). Unfortunately, Kaveri requires a new socket interface, so it won't be an upgrade path for anything with Trinity or considering Richland.