OpenCL should herald a brave new world of vendor-agnostic GPU compute applications... when those applications start coming out, that is. For now, SiSoftware has released a new version of its Sandra analysis, diagnostic, and benchmarking suite that includes a specially crafted OpenCL benchmark.
According to SiSoftware, Sandra 2010's OpenCL synthetic benchmark is a port of the Nvidia CUDA and ATI Stream benchmarks from Sandra 2009, the previous version of the suite. The new test probes GPU compute performance from two angles:
- Computational performance: in simple terms how fast it can crunch numbers. It follows the same style as the CPU Multi-Media benchmark using fractal generation as its workload. This allows the user to see the power of the GPGPU in solving a workload thus far exclusively performed on a CPU.
- Memory performance: this analyses how fast data can be transferred to and from the GPGPU. No matter how fast the processing, ultimately the end result will be affected by memory performance.
SiSoftware notes that its support for AMD's OpenCL 1.0 implementation is "available as Beta at this time," and performance "cannot be guaranteed." No such warning applies to the benchmark's Nvidia OpenCL support. Nevertheless, AMD has put out a press release saying it worked with SiSoftware to "[optimize] the performance of the OpenCL benchmarks for its GPU implementations." Purportedly, some tests run 2.7 times faster on the Radeon HD 5870 than on the dual-GPU GeForce GTX 295.
If you'd like to try the benchmark out for yourself, you can either download the "Lite" version or purchase commercial versions of Sandra 2010 from this page.