AMD Radeon HD 7950 Review Feat. Sapphire & XFX: Sewing Up The High-End Market

Compute Performance

Moving on from our look at gaming performance, we have our customary look at compute performance. Since compute performance is by definition shader bound, the 7950 is at a bit of a disadvantage here compared to gaming performance. Whereas ROP performance scales with the core clock, shader performance is hit by both the reduction in the core clock and the disabled CU array.

Our first compute benchmark comes from Civilization V, which uses DirectCompute to decompress textures on the fly. Civ V includes a sub-benchmark that exclusively tests the speed of their texture decompression algorithm by repeatedly decompressing the textures required for one of the game’s leader scenes. Note that this is a DX11 DirectCompute benchmark.

AMD’s greatly improved compute performance continues to shine here, though in the case of Civilization V it’s largely consumed by just closing the previously large gap between the GTX 500 series and the Radeon HD 6000 series. As a result the 7950 falls ever so short of the GTX 580, while the factory overclocked Sapphire and XFX cards give the 7950 enough of a push to come within 5% of the 7970.

Our next benchmark is SmallLuxGPU, the GPU ray tracing branch of the open source LuxRender renderer. We’re now using a development build from the version 2.0 branch, and we’ve moved on to a more complex scene that hopefully will provide a greater challenge to our GPUs.

Under SmallLuxGPU the 7970 enjoyed a large lead over the GTX 580, and this continues with the 7950. Even though the 7950 is well behind the 7970—to the tune of 24%—it’s still 33% ahead of the GTX 580 and the lead only grows from there. Meanwhile the XFX and Sapphire cards can catch up to the 7970 somewhat, but as this is truly a shader-bound test, you can’t make up for the lack of shaders units on the 7950.

For our next benchmark we’re looking at AESEncryptDecrypt, an OpenCL AES encryption routine that AES encrypts/decrypts an 8K x 8K pixel square image file. The results of this benchmark are the average time to encrypt the image over a number of iterations of the AES cypher.

In spite of being a compute benchmark, AESEncryptDecrypt is not particularly sensitive to GPU performance, showcasing the impact that setup times can have. The 7950 trails the 7970 by 10%, and overclocking doesn’t change this much. Unfortunately for AMD NVIDIA is still the leader here, showing that AMD’s compute performance still has room to grow.

Finally, our last benchmark is once again looking at compute shader performance, this time through the Fluid simulation sample in the DirectX SDK. This program simulates the motion and interactions of a 16k particle fluid using a compute shader, with a choice of several different algorithms. In this case we’re using an (O)n^2 nearest neighbor method that is optimized by using shared memory to cache data.

With the compute shader fluid simulation we once again shift back into a compute task that’s much more shader-bound. The 7950 only reaches 80% of the performance of the 7970, once more proving the real impact of losing a CU array. This is still enough to handily surpass the GTX 580 however, with the 7950 taking a 15% lead.