Performance overview of AMD A10- and A8-Series Trinity CPUs
Author: Debbie Wiles
AMD recently launched their Trinity APUs for the desktop, with claims that the Piledriver CPU architecture would offer a 10-15% performance boost over Bulldozer, and the IGP will outperform Llano by a similar amount. We looked at several reviews to see how the performance of the top A10 and A8 APUs compare against the Llano A8 and A6 APUs, and against a similarly priced Intel Core i3-3220. Figures we provide in this article are a summary of benchmark results from around a dozen reviews.
To see how Piledriver performs against Bulldozer, we can look at single-threaded benchmarks with an FX 8150 and an A10-5800K, which will run at the same speed. In the Cinebench 11.5 single-threaded test, the A10-5800K performs approx 6% better than the FX8150. It should be noted we have only seen one single-threaded test, so this may not be representative of all single-threaded workloads. In the same test, the A10-5800K shows a 21% boost over the A8-3850, the top end Llano at launch. The A8-5600K shows a 25% gain over the similarly positioned A6-3650, and even 12% over the top-end Llano. This is more down to the clock speed increase than the architecture at this time. In the single-threaded test, the i3-3220 shines, with a score 27% higher than even the A10-5800K.
More important for most people is the multi-threaded CPU performance. As the Bulldozer/Piledriver architecture uses modules with only one floating point unit to every two integer units, it's important to see how this affects overall performance. Once again, many of the performance gains seen over Llano are due to the increased clock speed the new architecture allows. Compared to the A8-3850, the A10-5800K shows a performance ranging from a loss of 25% in Cyberlink MediaShow to a gain of 41% in WinRAR 3.93. Over the full range of results we analyzed, the A10-5800K performs around 8% better than the A8-3850, losing ground only in apps that are heavy on floating point instructions. Even the A8-5600K performs on average 3% faster than the A8-3850, and 12% faster than the A6-3650.
Looking at the results for the two core, four thread i3-3220, the picture is more interesting, and dare I say surprising. At one extreme, the A10 is 12% worse than the i3 in an Excel test. At the other end, it outperforms the i3 by around 42% in TrueCrypt AES Encryption. Overall, the A10 shows a 4.4% performance boost over Intel's i3-3220, a figure that is close enough to call the two equal on a diverse CPU workload. This surprised me due to the fact the i3 showed very good single-threaded performance and the A10/A8 only have 2 FP units.
Where we expected to see the best results is in the integrated GPU performance. We looked at results for a Radeon HD 7660D in an A10-5800K, an HD 7560D in an A8-5600K and an HD 6550D in an A8-3870K. From Intel we looked at the HD 4000 iGPU, their top-end integrated GPU, in an i7-3770K and an HD 2500 in an i3-3220T, priced similarly to an A10-5800K. Finally, for an idea how Trinity fares against a discrete GPU, we looked at the results from an HD 5570, currently available new on eBay for $55 as it's two generations out of date, and a GT 640, from approx $90 on Amazon, as they have a similar number of shaders. Each set of results was divided into three sets of data - gaming, GPGPU and Synthetic benchmarks.
Across a selection of games, played at differing resolutions and quality/performance settings, the Radeon HD 7660D/A10-5800K has a performance boost of around 19% over the Radeon HD 6550D/A8-3870K, with the GPU compute results averaging out around 17%. This figure climbs to 24% when we look at the synthetic benchmark results. When we look at the Radeon HD 7560D/A8-5600K, with the shader count reduced by one third compared to the HD 7660D/A10-5800K, it seems fair to assume we will see a similar performance drop. However, both the games and synthetic benchmarks paint a different picture.
In most game tests, the HD 7560D has a small performance gain over the HD 6550D, with the OpenGL heavy Minecraft being the only test to show a large performance drop over the HD 7560D. On average, the HD7560D outperforms the HD 6550D by around 6%. The synthetic results show the same thing, with the HD 7560D outperforming the HD 6550D by around 8%. Llano GPU only outperforming the A8-5600K by 1%, close enough to call equal. We see a different picture with GPU compute workloads, with all the tests showing the HD 7560D performs around 30% worse than the HD 6550D, giving the Llano around 25% advantage.
When we look to the Intel CPUs/GPUs, we see what the APUs are really good for. Compared to the $320 Intel HD 4000/i7-3770K, the Radeon HD 7660D/A10-5800K is between 21% and 278% faster in the gaming tests, and produces 93% more throughput overall. Even the Radeon HD 7560D/A8-5600K is 71% faster overall than the HD 4000, and is at least 8% faster in all tests. A similar picture is painted by the synthetic results, where the HD 7660D is almost twice as fast as the HD 4000, and the HD 7560D is 65% faster. The GPU compute results are a little different, and the HD 7660D only outperforms the HD 4000 by 47%, with results ranging from 11% to 108% performance boost. The HD 4000 is favored over the HD7660D for some GPU compute tasks, with values ranging from the HD 4000 being 33% faster to the HD 7560D being 50% faster.
The GeForce GT 640 outperforms the HD 7660D in almost all the tests. In the gaming tests, it is 22% faster, with the OpenGL tests in Minecraft being the only place the APU wins out. The lead is stretched to 40% in the synthetic results. The GPU Compute results tell another story. In one test, the GT 640 scores double what the HD 7660D scores, and in another test those results are reversed, showing clearly how different applications might make better use of different architectures.
Finally, despite it being two generations old, we are looking at the Radeon HD 5570 because it has a similar shader count to the HD 7660D and we have seen more results for it than other, later cards. The HD 7660D outperforms the HD5570 in almost every test, and overall in the gaming and synthetic benchmarks that gain is around 10%. Only one of the GPU compute tests was carried out on the HD 5570, and in that test the HD 7660D is 36% faster. Comparing the HD 7560D with the HD 5570, things are a lot closer. In the gaming tests, the relative scores range from the HD 5570 being 24% faster to the HD 7560 being 23% faster, and overall the performance roughly equals out. In the synthetic benchmarks, the HD 5570 consistently outperforms the HD 7560D, and is overall around 6% faster. The only available GPU compute result for the HD 5570 has it on level footing with the HD 7560D.
Jump to CPU / Family