Sign in with
Sign up | Sign in

The “Balanced Platform”

ATI Stream: Finally, CUDA Has Competition
By

Despite that the phrase “balanced platform” just aches of being something that emerged from a marketing round table, it’s worth pausing here to explain some good thinking on AMD’s part. If you harken back to our CUDA coverage, you’ll recall that one of our greatest praises for Badaboom was that it exhibited some of the lowest CPU usage from the field of competing CUDA encoders—“low” being utilization in the 60% instead of 95% range. Transcoding is incredibly compute-intensive, and the de facto reasoning among software vendors seems to be that users want their jobs done as quickly as possible. If that means sacking the CPU and GPU concurrently to the exclusion of practically all other tasks, so be it. We have yet to see a Stream or CUDA application with a resource utilization slider.

But that doesn’t mean hardware vendors have to take the same approach. AMD’s “balanced platform” concept tries to leverage several major components of the system and spread the load as evenly as possible, doing the same work in the same time but leaving enough overhead for other applications to still function normally. That sounds pretty, but how well does it pan out in real life?

I hit on an excellent example of the balanced platform at work in our very first Espresso test, which involved taking a YouTube HD video (MPEG-4, 1280x720) and transcoding it into the iPhone 640x360 profile, which is also H.264 MPEG-4. The two images below show performance with the HD 4890. On the left, you see the test run without GPU acceleration, and on the right you have Stream enabled. You can see that with CPU-only encoding, all four cores of our Phenom II are practically maxed and GPU-Z registers the GPU load at a fairly steady and minimal 6%, which probably represents some element of the UVD pipeline being exercised during transcoding. In the GPU-enabled capture, you see a typical pattern emerge. Core 2 stays hammered (we don’t know why so many encoder utilities lean on this particular core) but cores 1, 3, and 4 drop down to sub-50% levels while the GPU load jumps.

The obvious next question is whether Nvidia, without a so-called balanced platform at its back, delivers similar results under CUDA. The question is a bit tricky to answer given our test setup because GPU-Z fails to display the GPU Load readout with the Nvidia card installed. Still, we can infer some things by CPU results and final performance. You’ll note in the left shot that CPU-only utilization looks topped out—very similar to AMD. When we add CUDA into the mix (right image), CPU utilization remains almost unchanged.

So is CUDA doing anything? You bet. There’s a 35% drop in output time with GPU acceleration enabled, so CUDA is plainly throwing everything it can at the job.  But here’s the interesting part: AMD and Nvidia show essentially the same encode time in CPU-only mode, but Stream yields a 108% performance gain, easily trouncing the CUDA result while averaging 40% less CPU utilization than CUDA.

As we’ll see, this is not a universal result. There are times when CUDA shines brighter. But tests like this show that AMD’s balanced platform statements are based on real benefits, not empty marketing meant to sell processors.

Display all 59 comments.
This thread is closed for comments
Top Comments
  • 12 Hide
    Spanky Deluxe , June 15, 2009 11:21 AM
    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.
Other Comments
  • 0 Hide
    radiowars , June 15, 2009 6:11 AM
    So..... TBH they both work pretty well, I hope that we don't start a whole competition over this.
  • -7 Hide
    falchard , June 15, 2009 6:16 AM
    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.
  • 0 Hide
    cl_spdhax1 , June 15, 2009 6:46 AM
    arcsoft simhd plugin is currently only enabled for n- cuda graphic cards.
  • 2 Hide
    Andraxxus , June 15, 2009 7:34 AM
    They're good but hopefully they will manage to improve them more. Competition is good for business.
  • -7 Hide
    DjEaZy , June 15, 2009 8:38 AM
    ... why just now talk about? I use it sins Catalyst 8.12...
  • 0 Hide
    IzzyCraft , June 15, 2009 8:43 AM
    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.
  • 7 Hide
    ThisIsMe , June 15, 2009 9:11 AM
    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.
  • -4 Hide
    ohim , June 15, 2009 9:25 AM
    why use 185.85 since those drivers are a total wreck

    http://forums.nvidia.com/index.php?showtopic=96665&st=0&start=0

    13 pages with ppl having different problems with that driver
  • 0 Hide
    Anonymous , June 15, 2009 11:09 AM
    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?
  • 12 Hide
    Spanky Deluxe , June 15, 2009 11:21 AM
    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.
  • -2 Hide
    wasteoftime , June 15, 2009 11:28 AM
    How come Folding is not something you guys ever cover in your CUDA comparisons. The main reason I left an ati card and bought an nvidia was the huge increase in my ppd.
  • 5 Hide
    armistitiu , June 15, 2009 11:52 AM
    wasteoftimeHow come Folding is not something you guys ever cover in your CUDA comparisons. The main reason I left an ati card and bought an nvidia was the huge increase in my ppd.

    Yeah...that's just what i want from a GPU : Folding@Home. I find video transcoding to be a more 'useful' way of using you GPU.
    Nice article. Haven't seen one in a long time.
  • -2 Hide
    thejerk , June 15, 2009 12:52 PM
    why did i feel the expresso "overview" page was an embedded advertisement?

  • 0 Hide
    sailer , June 15, 2009 12:54 PM
    wasteoftimeHow come Folding is not something you guys ever cover in your CUDA comparisons. The main reason I left an ati card and bought an nvidia was the huge increase in my ppd.


    I agree. The last three cards that I bought were Nvidia cards, based solely on their folding performance. When gaming, I prefer an ATI card. Oh yeah, I have four computers, three using Nvidia cards for folding and one with an ATI card for gaming. I think it would be great if the reviews included Folding@Home performance. It might also encourage ATI to make cards that did better for folding.
  • 0 Hide
    astrotrain1000 , June 15, 2009 12:58 PM
    Anandtech was saying that the output quality from ATI wasn't very good, but I didn't see any mention of it here. Can anyone confirm or deny, I have a 4870 but I'd like to know how the output quality is before I buy Expresso.
  • 0 Hide
    nickcardwell , June 15, 2009 12:59 PM
    I have used newer cards from both Nvidia and ATI, I also love my Mac. I would really like to see OpenCL take off. It's cross platform with Nvidia and ATI on both Mac and PC. That to me would end the CUDA / Stream war and bring it back to simply who builds the better GPU.
  • 0 Hide
    thejerk , June 15, 2009 1:01 PM
    "Final score: AMD 3, Nvidia 4. However, I’m going to call this a tie..."

    It actually IS a tie. You awarded NVidia a point for not offering an option for WMV encoding,
    even though performance showed a very slight, but measurable, increase with Stream enabled.

    You didn't give credit where credit was due. Do it right the next time.
  • 0 Hide
    lire210 , June 15, 2009 1:28 PM
    the gpu looks a bit weak on the ati side. when was the gtx 280 ever ment to go head to head with a hd4870. i would like to suggest a hd4890 for da job bit more even. that is if we are going on who drivers are better which seems to be more of the focus considering the lack of hardware evenness.
  • -1 Hide
    JPForums , June 15, 2009 1:57 PM
    There is no mention of how good the output quality is. How does it compare to the source? How does it compare to nVidia?
  • 3 Hide
    williamvw , June 15, 2009 4:37 PM
    Hey, all. Many thanks for the initial feedback. A few notes:

    1. I would have loved to test with the Folding@home app. I actually tried to when doing the former CUDA-on-a-budget article. However, I quickly discovered that the results were meaningless because the work loads varied too much. NVIDIA helped solve this problem by creating a series of batch files for SETI@home that used a common work load, and that's what you see in the article. However, there is no such tool that I know of for Folding@home and AMD/ATI has not released an equivalent set of testing tools for SETI@home.

    2. I count seven charts -- 4 to 3. I did give the better coding point to NVIDIA on page 7. NVIDIA has 0 points on page 6 and two points by the end of page 7. :-)

    3. The side-by-side captures you see in the later article pages show samples of Stream vs. CUDA output. These are taken from GPU-accelerated output files. To my eye, they look almost identical, but I offer them up for you to make your own judgments. I would say that the output quality issues that plagued Stream's initial launch have been remedied.

    4. Yes, I agree that, ultimately, OpenCL and DirectX 11 will lay the entire Stream/CUDA issue to rest. But that's someday. For now, this article's purpose was to take a look at today's technology.

    5. I tested with an HD 4890, not a 4870. Apologies if there are any typos to the contrary.

    6. There is no behind-the-scenes money changing hands that resulted in my page detailing CyberLink Espresso. I developed that page for two reasons. First, as I mentioned, Espresso is the ONLY application today with even support for both Stream and CUDA, so it made sense to me that many people might want to buy it because of its agnostic support -- and it's a great tool. Second, in part because of this agnosticism, CyberLink has been immensely helpful to me in writing this article in a fair, even-minded, and accurate manner. The company helped me through many nights, often maintaining email dialogues well past midnight. So forgive me for being enthusiastic about the product. If CyberLink's customer support is even half its press support, I think you'll be pleased.
Display more comments