SiSoftware releases OpenCL benchmark

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.

Comments closed
    • Hattig
    • 13 years ago

    It’s only an API, it just sounds like the DirectCompute drivers are more efficient than the OpenCL/CUDA ones at utilising the hardware. It’s still running on the same hardware.

    Next development cycle and the team that did the DirectCompute drivers will share their optimisations with the other teams and improvements will happen all around.

    It could be the other way around on the ATI cards (from that other gpGPU benchmark I’ve seen floating around) – does that mean that DC is crap?

    • Meadows
    • 13 years ago

    8800 GT, GeForce driver 195.62.

    CUDA and OpenCL scores were nigh identical, while DirectCompute scored far above them. To my understanding, all 3 were tasked to do the same thing, which would mean DC is superior.

    • codedivine
    • 13 years ago

    Well, maybe they just wrote better code for DirectCompute and did not pay much attention to optimising OpenCL code. Or maybe your video card (whatever it is) has better drivers for DirectCompute than for OpenCL. So 35% more does not say anything about the APIs involved.

    • UberGerbil
    • 13 years ago

    Savvy developers are usually willing to make platform-specific optimizations when a particular platform represents the majority of their “market.” Resources permitting, of course.

    • thecoldanddarkone
    • 13 years ago

    No, they go with whatever is easiest for them, it has nothing to do with what you said.

    • wingless
    • 13 years ago

    OpenCL works on any platform. Stanford likes to make a client easily portable to MacOS, Linux, Windows, and all the different variants of each. DirectCompute may be higher performing (with this generation of GPUs/drivers/software), but it would be limited to Microsoft products only.

    • Deanjo
    • 13 years ago

    Actually had both GPU’s been utilized in the GTX 295 the speed difference should have been closed to 1.35 times the GTX 295. Unlike SLI, crossfire, etc scaling is near linear when it comes to GPGPU applications that can take advantage of multiple GPU’s.

    • BabelHuber
    • 13 years ago

    Sandra always has been a benchmark not suited for comparisons of different CPUs or GPUs.

    However, it is a cool tool to check if the PC you just have built performs as expected.

    Since I can do little wrong when installing a video card, the GPU test is not very important, OTOH.

    • wingless
    • 13 years ago

    OpenCL works on any platform. Stanford likes to make a client easily portable to MacOS, Linux, Windows, and all the different variants of each. DirectCompute may be higher performing (with this generation of GPUs/drivers/software), but it would be limited to Microsoft products only.

    • khands
    • 13 years ago

    So it’d be closer to 1.7x, still a pretty good gap.

    • Deanjo
    • 13 years ago

    l[

    • Meadows
    • 13 years ago

    According to comment #15, you might want to fold using DirectCompute instead.

    • anotherengineer
    • 13 years ago

    oooo a benchmark big deal *sarcasm*

    How about some [email protected] program designed for OpenCL or something.

    A benchmark, like that does anything useful ………sigh

    • Meadows
    • 13 years ago

    No.

    • flip-mode
    • 13 years ago

    #1, #2, #3, #5, #7, #9, #10, #8, #13, #14:

    STFU.

    • Meadows
    • 13 years ago

    All of this doesn’t matter anyway, because I got 35% higher scores using DirectCompute than when using /[

    • lycium
    • 13 years ago

    but he’s good about using american english on tr. however in this case, being a british site, its use of “analyses” is correct and meadows’ point is valid.

    • BobbinThreadbare
    • 13 years ago

    Cyril lives in France…

    • lycium
    • 13 years ago

    hmm, it should reflect real-world practices; optimising gpu kernels for specific hardware (i’m pretty sure they’re referring to use of SIMD on AMD, which is crucial for realising its great potential) is realistic.

    • lycium
    • 13 years ago

    i have one that could be used for this purpose, it renders the recently-famous 3d “mandelbulb” fractal.

    • lycium
    • 13 years ago

    get back under the bridge

    • Meadows
    • 13 years ago

    Go back to school.

    • Meadows
    • 13 years ago

    I know that. But “sic” means that this was the “original, unaltered spelling”, usually to avoid personal liability when a quote has original errors. But to what end? It’s not spelled /[

    • lycium
    • 13 years ago

    “this analyses” is incorrect in american english, should be “analyzes”. in british english “analyses” is the plural of “analysis”, which would be incorrect.

    edit: nope, it’s obviously also the present continuous form of analyse, so i was definitely wrong.

    • PRIME1
    • 13 years ago

    TR benchmark?

    • Hattig
    • 13 years ago

    No problem at all. sisoftware.co.uk -> UK company therefore “analyses”

    • ew
    • 13 years ago

    *[

    • Meadows
    • 13 years ago

    TR don’t use auto-correct. Trust me.

    • WillBach
    • 13 years ago

    Maybe he meant to copy over an error, which was corrected by autocorrect?

    • Meadows
    • 13 years ago

    g{

Pin It on Pinterest

Share This

Share this post with your friends!