Articles: CPU

Bookmark and Share

Pages: [ 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 ]

Heterogeneous Performance

Even when we benchmark their conventional computing performance, we shouldn’t forget that AMD’s Socket FM2 processors are hybrid devices that combine x86 and graphics cores. The Devastator graphics core employed in the Richland and Trinity series supports the OpenCL framework which allows using its shader pipelines for general computing. The use of heterogeneous resources for a single task is the key idea of the APU concept as promoted by AMD. The company wants to influence the software market so that OpenCL were used everywhere.

There has been some progress in this area indeed. The number of software applications that can utilize the graphics core’s computing resources is on the rise, so AMD can proudly post a rather long list of OpenCL-compatible programs.

Unfortunately, not all of these programs offer full support. Some of them only use the graphics core for certain tasks, which is due to the specifics of the graphics core design. It is only with simple parallel operations that a graphics core can really be efficient, so the APU concept can’t really push the performance bar of AMD’s hybrid processors to a whole new level across a variety of applications. On the other hand, there are quite a lot of scenarios, such as image or video processing, where the graphics core can be of huge help to the conventional x86 cores.

Ideally, we wouldn’t use special tests to benchmark OpenCL performance. It would be better if the popular applications we have on our ordinary test program supported OpenCL and heterogeneous computing. But since they do not, we have to check out such processors in specific tasks.

The first of them is Luxmark 2.0, which is based on the LuxRender rendering engine. We use the average-complexity Sala scene, rendering it on both graphics and x86 cores of the tested processor.

As you can see, things hardly get different when we make use of the graphics cores’ computing resources. Like AMD’s APUs, modern Intel processors offer full support for OpenCL. So they remain in the lead, even though their graphics core is less advanced. The superior integrated graphics of the Richland and Trinity processors doesn’t help them win even at heterogeneous loads. It is the Core i5-4430 with Intel HD Graphics 4600 and the Core i3-3225 with Intel HD Graphics 4000 that occupy the top places. The AMD FX-4350 is on top, too, but mostly due to our using it together with a discrete graphics card Radeon HD 6670.

Like before, the A10 and A8 series APUs are only comparable to the Core i3 series whereas the A6 series is still inferior to the Intel Pentium. And again, we don’t see any big difference between the Richland and the Trinity: the A10-6800K is a mere 4% ahead of the A10-5800K.

The introduction of OpenCL support into the popular archiver WinZIP shows that the APU concept is embraced by the software market, so we can’t help testing our processors in WinZIP 17.5. We compress a folder with files with a total size of 2.53 GB.

Besides their OpenCL compatibility, the latest versions of WinZIP work well on multicore processors, so the benefits of a specific graphics core are masked by the conventional x86 performance. In other words, WinZIP doesn’t differ much from OpenCL-incompatible archivers despite its using graphics core resources.

Image-editing and video content processing applications are a more traditional type of software that support OpenCL acceleration. Corel AfterShot Pro, a popular tool for batch-processing of digital photos, is an example of that. We use a scenario in which two hundred 12-megapixel RAW-format images are post-processed and exported as JPEG files.

AMD’s hybrid processors seem to be more efficient here. The A10 and A8 models beat the Core i3 series irrespective of what graphics cores they have and are very close to the junior Core i5. The AMD A6 series is brilliant as well, outperforming the Intel Pentium.

Another example of a popular OpenCL-compatible application is the professional video editing tool Sony Vegas Pro 12. When rendering video, it can distribute the load among all the computing resources of hybrid processors.

Intel CPUs with HD Graphics 4600 and HD Graphics 4000 can offer better performance than the various Richland and Trinity products from AMD. So, promoting the heterogeneous computing concept, AMD actually helps its archrival as well because Intel also implements OpenCL in its products and equips them with fast graphics cores. As a result, senior Socket FM2 solutions don’t look superior even if we use APU-optimized applications with OpenCL support. The Richland is only 5% faster than the Trinity, so we don’t see any breakthroughs here. Unfortunately, the modern GCN graphics architecture is not implemented in the Richland.

The second OpenCL benchmark we used was SVPMark 3. It is a specialized performance benchmark for the SmoothVideo Project software which improves video playback smoothness by inserting new intermediary frames into the video stream. This software makes active use of GPU resources via OpenCL.

The Richland-based APUs fail to compete with the junior Core i5 but look good enough in comparison with the Core i3 that have the older Ivy Bridge microarchitecture. Even the Core i3-3225 with HD Graphics 4000 falls behind all of the A10 and A8 series APUs. The A6-6400K beats not only the Pentium but also the Core i3 models with the junior version of the integrated graphics core.

So we can see that OpenCL optimizations can produce miraculous results in some situations, but there are too few such examples. AMD’s much-touted concept of heterogeneous computing on Socket FM2 processors doesn’t look like a killer feature. We’ve never seen A10 series APUs beat the junior Core i5 even under the most favorable conditions. The desktop Richland-based products can only compete with Intel’s Core i3 series.

Pages: [ 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 ]


Comments currently: 81
Discussion started: 09/01/13 07:48:19 PM
Latest comment: 11/29/16 07:59:48 AM

View comments

Add your Comment