Forgot your password?
typodupeerror
AMD Graphics Games Hardware

Frame Latency Spikes Plague Radeon Graphics Cards 158

Posted by Soulskill
from the have-you-tried-turning-it-off-and-on-again dept.
crookedvulture writes "AMD is bundling a stack of the latest games with graphics cards like its Radeon HD 7950. One might expect the Radeon to perform well in those games, and it does. Sort of. The Radeon posts high FPS numbers, the metric commonly used to measure graphics performance. However, it doesn't feel quite as smooth as the competing Nvidia solution, which actually scores lower on the FPS scale. This comparison of the Radeon HD 7950 and GeForce 660 Ti takes a closer look at individual frame latencies to explain why. Turns out the Radeon suffers from frequent, measurable latency spikes that noticeably disrupt the smoothness of animation without lowering the FPS average substantially. This trait spans multiple games, cards, and operating systems, and it's 'raised some alarms' internally at AMD. Looks like Radeons may have problems with smooth frame delivery in new games despite boasting competitive FPS averages."
This discussion has been archived. No new comments can be posted.

Frame Latency Spikes Plague Radeon Graphics Cards

Comments Filter:
  • by earlzdotnet (2788729) on Wednesday December 12, 2012 @05:53PM (#42265931)
    It seems like everyone always wants a single measurement to judge how good something is. Graphics cards have FPS, CPUs have GHz, ISPs have MB/s. What's not shown in these single number measurements are things like lag, or overheating problems, or random spikes of instability.

    Sigh. Maybe one day we'll learn that every product needs more than a single number to judge how good it is performance-wise.
    • Re: (Score:2, Informative)

      by Anonymous Coward

      The new measurement for graphics cards is ms/frame, and it's been used on some review sites for more than a year now.

      • by The Last Gunslinger (827632) on Wednesday December 12, 2012 @06:10PM (#42266115)
        So they just flipped the fraction and multiplied by 1000...brilliant! </sarcasm>
        • by TheLink (130905) on Thursday December 13, 2012 @05:10AM (#42270443) Journal
          There's a difference between measuring milliseconds per frame and frames per second.
          With the former your minimum resolution is one frame.
          With the latter your minimum resolution is one second.

          Because of that even the minimum FPS rate doesn't necessarily tell you how jerky or smooth the rendering is - since it's averaged out over one second.

          Taken to the extreme it could be rendering 119 frames in 1 millisecond and get stuck on one frame for 999 milliseconds, look like a frigging slide show but still show up as 120FPS. Whereas that sort of thing will stick out like a sore thumb on a milliseconds per frame graph. Hence measuring milliseconds per frame is better.

          The extreme case shouldn't happen in practice, but as the article shows (and from personal experience) the max/high latency stuff does happen. I've personally experienced this on my old ATI and Nvidia cards - my Nvidia 9800GT was slower but smoother than my current Radeon. I went ATI because the Nvidia cards were dying too often (they had a manufacturing/process issue back then). But my next card is probably going to be Nvidia. Even with Guild Wars 1 my ATI card can feel annoyingly "rough" when turning in the game - you see the FPS stay high but it's rough to the eyes if you get 60 fps by getting a few frames fast then a very short pause then a few frames fast then pause repeat ad nauseum.

          On a related note it's good to see that at least some benchmark sites are also starting to take latency/consistency into account for stuff like SSDs. A maximum latency that is too high and occurs too often will result in worse user experience, even if the overall throughput is high, and even for storage/drives.
      • That is fundamentally the same thing as FPS, unless there are some unstated assumptions that come along with this.

    • by Anonymous Coward

      A person who knows all the different measurements usually doesn't just look at FPS. The single numbers are there for people who don't understand all the other things. It helps them decide on what card is supposed to be better.

      Hell even cars use a MPG sticker on each window, they don't go into how they get that MPG number, and most are not reflected in real world tests of the car.

    • Re: (Score:3, Informative)

      by Anonymous Coward

      How bout next time you RTFA. For the past year or so, that site has scatter plots showing frame render times in milliseconds; they show time and time again that despite similar FPS rates, performance isn't the same.

    • by Anonymous Coward

      You didn't read the article, did you...the whole point of it was that despite the Radeon's often better FPS performance, it suffers from spikes that make it feel less smooth.

    • Re: (Score:2, Informative)

      by Anaerin (905998)
      Oh, we have learned this. Which is why decent review sites don't just publish a "single number" representation of speed. They post a complete FPS graph [hardocp.com] for similar runs through each game, so you can compare side-by-side
      • by TheLink (130905)
        That's not the same as milliseconds per frame.

        Frames per second graphs will be showing numbers that are AVERAGED over one second. So the minimum number still might not reflect how crappy the rendering is.

        To take an extreme example, you could have a card that does 120 FPS by drawing 119 frames in 1 millisecond and the last frame in 999 milliseconds.

        Then you have a card that does 50FPS by drawing each and every frame in 20 milliseconds. The 50FPS card will appear smoother, whereas the 120FPS card will look li
    • by Anonymous Coward

      Sigh. Maybe one day we'll learn that every product needs more than a single number to judge how good it is performance-wise.

      I understand what you're saying, but it's hard to take a comment seriously from someone with such a high uid...

  • by UnknownSoldier (67820) on Wednesday December 12, 2012 @05:57PM (#42265975)

    The fact that it effects single GPU in addition to SLI and Crossfire is worry some.

    Micro-Stuttering And GPU Scaling In CrossFire And SLI
    http://www.tomshardware.com/reviews/radeon-geforce-stutter-crossfire,2995.html [tomshardware.com]

    • by ultranova (717540)

      Micro-Stuttering And GPU Scaling In CrossFire And SLI

      Personally, I'm more annoyed at macro-stuttering: the tendency of AMD drivers to repeatedly crash and recover in some games, resulting in screen freezing for several seconds.

  • CPU Backup (Score:5, Funny)

    by dohzer (867770) on Wednesday December 12, 2012 @06:02PM (#42266025) Homepage

    So half of AMD's business isn't at its peak. Big deal.
    They can just rely on their CPU business for a while.

    • Indeed. If AMD doesn't fix their x86 stuff with something that makes me salute, they're going to be out of the next round of upgrades. The FX-8350 is an improvement over the FX-8150, enough for my to upgrade to it. However, it's still a ways behind where we should be. That Intel processor with the six-cores, with six hyper-threaded units, is calling out to me. Given that I have already purchased the necessary liquid coolant system to keep the FX processor happy, switching over to that i7 wouldn't be much of

  • I will reserve judgement and see if other review sites have similar issues and if it is a problem with just the 7k series or not.
  • by Ratchet (79516) on Wednesday December 12, 2012 @06:16PM (#42266195) Homepage

    Anyone that actually knows anything about the GPU industry knows that both AMD and NVIDIA graphics suffer from these latency spikes, but it's not with all their SKUs. NVIDIA's 660 Ti works well in this case, but their 670 and 680 has more latency spikes than the competitive AMD cards do. The 7850 demonstrated here is an anomaly for AMD. None of their other cards do this. Look at past reviews from Techreport and you will see what I mean.

    • by AvitarX (172628)

      hah, the last GPU I purchased was a 6600, it's 10 times better (GTX though, I think the Ti is better)

    • by edxwelch (600979)

      > The 7850 demonstrated here is an anomaly for AMD. None of their other cards do this.

      Just shows that you haven't RTFA. The article compairs 7950 versus 660 Ti.

    • by enderwig (261458) on Wednesday December 12, 2012 @10:41PM (#42268717)
      However, if you actually read the Tech Report's review of the GTX 670, you will find they say the exact opposite. The GTX 670 has ridiculously low latency compared to the Radeon 6990 and just a bit lower than the 7950 and 7990.

      As clearly seen on page 3 [techreport.com] of the 670's review.
      • 7970, or the 7990? The 7970 has been the top part from ATI for the last however many months, the 7990 is something of a unicorn, as we have heard tales of the 7970x2, but no one has really seen it.

        • by jakobX (132504)

          7990 exists but its not exactly a high volume product. You can of course buy it if you are rich enough.

    • by Ash Vince (602485) *

      Anyone that actually knows anything about the GPU industry knows that both AMD and NVIDIA graphics suffer from these latency spikes, but it's not with all their SKUs. NVIDIA's 660 Ti works well in this case, but their 670 and 680 has more latency spikes than the competitive AMD cards do. The 7850 demonstrated here is an anomaly for AMD. None of their other cards do this. Look at past reviews from Techreport and you will see what I mean.

      Sorry, but ATI suffer from this far more than Nvidia as apparently it is something nvidia actively try to improve:

      http://techreport.com/review/21516/inside-the-second-a-new-look-at-game-benchmarking [techreport.com]

      You have to read the full article, but even though it is old now it sound like it is still relevant.

  • ...and I just ordered a 7850. Well, here's hoping it's overblown or fixable in software.
    • I wouldn't get too worried. I took advantage of the free games offer to pick up a HD 7970. Running Windows 8 with the latest AMD beta drivers and I have yet to notice this issue.

    • by Jetra (2622687)
      I got a Sapphire 6450. It's nice and works for what I use it for *coughMinecraftcoughcough*
  • Frequency scaling (Score:5, Interesting)

    by fa2k (881632) <pmbjornstad@gmailDEBIAN.com minus distro> on Wednesday December 12, 2012 @06:44PM (#42266497)

    The article didn't mention power settings. I'm quite skeptical of all the new tech which overclocks on demand and then clocks down when it gets too hot (or too idle). They should definitely try this test with the standard frequency: pinned at the nominal frequency (if there is such a thing at all).

  • by Spy Handler (822350) on Wednesday December 12, 2012 @06:55PM (#42266587) Homepage Journal

    Diamond Stealth 3D card with Cirrus Logic chip. It doesn't suffer from latency spikes.

    • Diamond Stealth 3D card with Cirrus Logic chip. It doesn't suffer from latency spikes.

      I know what you mean. I was annoyed my car had a flat spot when accelerating, so I ditched it and went and bought a Cutlass with the Olds diesel engine. Problem solved.

    • by dywolf (2673597)

      I miss Diamond.
      Ah nostalgia.

  • by slacka (713188) on Wednesday December 12, 2012 @06:57PM (#42266601)

    "This trait spans multiple games, cards, and operating systems, "
    First of all the article only tests 2 cards accross Win7 and Win8. Considering that Win8 is basically just Win7 SP2, it's hardly fair to make that statement. Micro-stuttering an issue that mainly affects multi-GPU cards. Both Nvidia and ATI have had issues with this in their SLI and Crossfire cards. You can read more about it here:
    http://hardforum.com/showthread.php?t=1317582 [hardforum.com]

    • As far as 3D graphics is concerned, there are noticeable differences between Win7 and Win8 - the former is WDDM 1.1, the latter is WDDM 1.2, which covers quite a bit of new stuff (basically all the new features in Direct3D 11.1).

    • by Ash Vince (602485) *

      "This trait spans multiple games, cards, and operating systems, "
      First of all the article only tests 2 cards accross Win7 and Win8. Considering that Win8 is basically just Win7 SP2, it's hardly fair to make that statement. Micro-stuttering an issue that mainly affects multi-GPU cards. Both Nvidia and ATI have had issues with this in their SLI and Crossfire cards. You can read more about it here:
      http://hardforum.com/showthread.php?t=1317582 [hardforum.com]

      But if you avoid SLI then nvidia cards are fine, only ATI really suffers from this issue badly. See the link I posted above.

  • Games on the Radeon 6850 would generally perform fine but seem 'jittery'. Usually didn't notice it, but sometimes it was quite obvious that you were losing frames here and there even when in non-complex situations. Of course you wouldn't notice a single one, but when it's happening every 2-3 seconds it starts being noticable when you're playing for hours. Some games were much worse than others, though I couldn't say it was directly related to how shiny the game looked. It was never big enough a deal that I

  • A few months ago I decided to do a complete replay of the entire Mass Effect trilogy with my 6900 series card, and I am seeing the occasional lag that didn't used to be there. I also revisited Skyrim when Dawnguard came out, and I'm seeing it there too. This machine didn't used to do this, and since I can't find anything else running that could cause the CPU to spike, I have been working on the assumption that some driver update (perhaps as far back as six months ago) has been to blame.

    It's nice to see th

  • I'm shocked, simply shocked.

  • by dkegel (904729) on Wednesday December 12, 2012 @08:02PM (#42267211) Homepage

    Look at the bottom Skyrim graph in
    http://techreport.com/review/24022/does-the-radeon-hd-7950-stumble-in-windows-8/8 [techreport.com]

    The slow frames always follow extra-fast ones. Looks like some work is being deferred past a frame boundary?!

  • I notice that like 2 months ago, after upgrading from a Radeon 5850 to a 7850. Since my CPU is an old Core Duo 2.4GHz I didn't expect that much of a performance boost, but expected a noticeable change. When comparing the Haven benchmark results with the previous card, the higher frame rate went up as expected (15-25 fps not remember now), but the lower frame rate went down too from 18 fps to 6 fps on new card. Tried with some driver revisions, being 10.10 the last one tested having same behavior on the 7
  • Well it's a good thing they just finished laying off a slew of engineers now they can afford to absorb the bad press from this one.
  • I have some serious doubts about this kind of testing.

    - he is using different thresholds in different games.
    - he is using fraps to measure it, but fraps only records part of the process
    - he is testing with vsync off, which is weird considering he is trying to measure smoothness
    - his average FPS numbers for AMD card are much lower than in reviews from other sites

    Are his tests always reproducable (spikes always occur at the same spot)?
    Why such a small selection of games?

    The way this test is constructed means

  • Well, I had seen this problem for very very long time with the ATIs. First with Rage 128 Pro (with Intel CPU) and later with the 4850 (with AMD CPU). That's why during the last upgrade I went with the nVidia instead.

    Though of course I have never thought about it as a problem(*) nor investigated it thoroughly: I have simply seen that unless one keeps graphics at low settings, ATIs tend to occasionally drop to 10-20 FPS from the usual 40-60. Last game - and very dated at that - I have tested with ATI 4850 w

  • For years I was part of the cult that would invest $1000's upgrading or building my own PC game box, buying the best video card, researching the best CPU and RAM that could be overclocked, even matching the exact spec's of a system bench marked on an enthusiasts sight.

    After the many hours of building and tweaking and break in I finally would get a hold of the most popular game at the time and then try to run it will all the graphic's features cranked to the max because every website I reviewed said I could.

Machines certainly can solve problems, store information, correlate, and play games -- but not with pleasure. -- Leo Rosten

Working...