ATI Stream: Finally, CUDA Has Competition

AMD On Stream’s Prospects

TH: Enough with the Inquisition! Let’s look forward. Do you anticipate Stream and GPGPU technology being applied in the consumer space beyond today’s video player/editor/transcoder apps?

AMD: Yes, very much so. Applications like physics simulation, AI, and video processing are natural to accelerate using Stream/GPGPU/parallel technologies. The key challenge is to get programmers to start using the hardware and programming languages/APIs that are available to them to properly model the compute problem. OpenCL and DirectX 11 Compute make major advances in how a programmer can harness the processing power of a modern GPU.

TH: Some have proposed that the GPU will someday overtake the CPU in prominence, and certainly the GPGPU benefits we’ve witnessed make that seem more plausible. Given the impending arrival of Fusion and the continuing evolution of Stream, what is AMD’s stance on the role of graphics processing and the GPU going forward?

AMD: It is simply wrong and, quite frankly, self-serving when companies say that a GPU is the most important processor in the PC. It is equally wrong to say x86 CPUs are the most efficient processors to solve all compute problems. The answer is “it depends.” The workload or compute problem determines what is the best processor to handle that particular workload or compute problem. CPUs are great for linear/out-of-order/unpredictable compute problems while GPUs are great for parallel workloads like video processing, physics simulation, or AI. Trying to modify or adapt a compute problem to suit the hardware that a company happens to sell is not the way forward.

It is also easy to forget a third class of processor—fixed function. Fixed function blocks are often overlooked these days because they are no longer in vogue. People mistakenly believe that general purpose devices cure all. Rubbish. Fixed function blocks for things like video decoding or tessellation are not only great from a power perspective but they also scale very well from the top of a product stack to the bottom. Those products that only have general purpose processing will be less power efficient and will not scale better than processors that balance general purpose and fixed function. Finally, fixed function blocks are best for compute problems that are neither efficient for CPUs or GPUs, like the reverse entropy portion of H.264 decode.

The best solution is a good balance of x86 CPU, modern GPU, and fixed function. A balanced platform for the average consumer will be the best solution as it will have all the compute elements. AMD’s Fusion products will balance all three processor types. Having a fast CPU or GPU alone won’t cut it today.

TH: Given the recent hotfix and Stream update, how would AMD now characterize Stream versus CUDA in terms of quality and performance?

AMD: Well, I usually prefer to provide the hardware and software so that you can test and build your own opinion, but objectively I believe that the ATI Stream transcoding framework has leapfrogged CUDA transcoding with this update both on performance and quality. I say “objectively” because it’s first time we have an application (MediaShow Espresso) that enables both technologies and thus direct comparison.

  • radiowars
    So..... TBH they both work pretty well, I hope that we don't start a whole competition over this.
  • falchard
    Did someone necro an old topic? I think ATI has been talking about ATI Stream for a while. I know atleast a year since FireStream.
  • cl_spdhax1
    arcsoft simhd plugin is currently only enabled for n- cuda graphic cards.
  • Andraxxus
    They're good but hopefully they will manage to improve them more. Competition is good for business.
  • DjEaZy
    ... why just now talk about? I use it sins Catalyst 8.12...
  • IzzyCraft
    Stream is old but not nearly as old and compatible as CUDA I'd get it a year or two more when more capable cards circulate the market and trickle down to the people before i would call it competition.

    Well it's good to see more then just 1 app that supports it.
  • ThisIsMe
    Just for the sake of it, and the fact that many pros would like to know the result, it would be nice to see comparisons like this using nVidia's Quadro cards vs. ATI's FirePro cards.
  • ohim
    why use 185.85 since those drivers are a total wreck

    13 pages with ppl having different problems with that driver
  • I think the second graph on the "Mixed Messages" page isn't the right graph.

    It's the same graph from the following "Heavier Lifting" page instead of a graph for the 298MB VOB file that should be shown?
  • Spanky Deluxe
    Stream and CUDA are likely to go the way of the dodo soon though. OpenCL's where its at. Unfortunately its a tad hard to get programming with it right now since you need to be a registered developer on nVidia's Early Access Program or you have to be a registered developer with Apple's developer program with access to pre-release copies of Snow Leopard.
    Virtually no one will bother using CUDA or Steam after OpenCL's out - why limit yourself to one hardware base after all? It'd be like writing Windows software that only ran on AMD processors and not Intel. Developers will not bother writing for both when they can just use one language that can run on both hardware platforms.