1. #1

    GPU performance by frame latency & Radeon 7900 problems

    Found this interesting article of testing the graphics card performance not only on raw FPS but the latency between frames which can show as stuttering.

    TL;DR version would be something like:

    Even if card A has higher average FPS than card B, card A might frequently have big lagspikes which makes the game look shit outside benchmarking and in real world use.

    Apparently Radeon 7950 with latest few driver versions has suffered from this problem pretty badly, and makes me wonder about the generally accepted best value for money cards list.
    Never going to log into this garbage forum again as long as calling obvious troll obvious troll is the easiest way to get banned.
    Trolling should be.

  2. #2
    Quote Originally Posted by vesseblah View Post
    Found this interesting article of testing the graphics card performance not only on raw FPS but the latency between frames which can show as stuttering.

    TL;DR version would be something like:

    Even if card A has higher average FPS than card B, card A might frequently have big lagspikes which makes the game look shit outside benchmarking and in real world use.

    Apparently Radeon 7950 with latest few driver versions has suffered from this problem pretty badly, and makes me wonder about the generally accepted best value for money cards list.
    This is been common knowledge for a while, it is one of reasons Vsync is used. By staying at a divisible you get a better viewing pleasure than w/o, screen tearing is one thing that it fixes but it also solves incomplete rendering of frames, 30 FPS w/ 30 completed frames looks a lot better than 60 Fps w/ 1-5 out of 10 frames being incomplete or plain dropped.

    Issue is tho was it that card, multiple card etc. Article doesnt seem to be very deep rather 1v1. In this type of scenario I'd like a few more reports, as one guy gets perfection the next guy gets a big pile of doa poop.
    Last edited by Milkshake86; 2012-12-13 at 02:10 AM.

  3. #3
    That's why I felt my gtx 680 was running a game much smoother than 2x 6970's (which are performing better than my 680@stock)

  4. #4
    Quote Originally Posted by Milkshake86 View Post
    This is been common knowledge for a while, it is one of reasons Vsync is used. By staying at a divisible you get a better viewing pleasure than w/o, screen tearing is one thing that it fixes but it also solves incomplete rendering of frames, 30 FPS w/ 30 completed frames looks a lot better than 60 Fps w/ 1-5 out of 10 frames being incomplete or plain dropped.
    No it's not about screen tearing at all, but straight out bugs in the rendering pipeline that causes the GPU to hang every 15-20 frames or so. Screen tearing is caused by the rendering not being synced to screen refresh rate. Two entirely different things.

    Quote Originally Posted by Milkshake86 View Post
    Issue is tho was it that card, multiple card etc. Article doesnt seem to be very deep rather 1v1. In this type of scenario I'd like a few more reports, as one guy gets perfection the next guy gets a big pile of doa poop.
    Because reading is obviously hard: from the first page you might notice that the article was a follow-up to Win8 only test (because critics thought the Radeon 7900 bugs were caused by bad Win8 drivers) and last page has some comments on the previous results with the different cards tested in the previous article so it's not 1v1.
    Last edited by vesseblah; 2012-12-13 at 10:28 AM.
    Never going to log into this garbage forum again as long as calling obvious troll obvious troll is the easiest way to get banned.
    Trolling should be.

  5. #5
    This is actually quite funny. I have a 7970GHZ edition card as well. But I ended up using my GTX680 instead cause I feel it runs games smoother despite having marginally lower frames.

    For HTPC videocard though, the 7970GHZ is a a beast and the GTX680 doesn't even come close. Too bad this has nothing to say for gaming cards....

    And before I get attacked by someone who is obviously biased towards amd: I have bought an AMD card for every generation, this is the first nvidia card I have personally owned since the GTX280. My 7970 is an ASUS Matrix platinum card, which is supposed to be handpicked and it is clearly the best gpu on the market when it comes to frame and in multi screen the GTX680 looks like a 10 year old card compared to it.

    My point here being: So many people say the GTX680 is a waste of money bla bla bla. Iirc the 7970 was a good 120$ more than the GTX680, delivering on average 4-5 frames more per second. It looks better but runs hotter, despite it being a handpicked ROG card.

    The GTX680 is, how I see it, after comparing the two in many games the superior video card if given these criterias: 1080p-1440p 1 screen(Anything more than this and the two cards can no longer be compared). Well optimized engine that doesn't require extreme amounts of video memory (metro2033 is not such a game, don't know why it's tested), not the highest AA(When a game utilizes heavy AA, the GTX680 falls behind by alot.)

    Now, you can say what you will about pricing, but I feel the GTX680 is the first top card from nvidia that has been sold at a reasonable price. GTX580 was way more expensive when released. At least in Norway. Now it is very possible to pick up a GTX680 for around 400 euros, in comparison the Matrix will set you back around 550 euros. At this time, I feel like the Matrix was a dumb investment despite being "best gpu on the market"

  6. #6
    Quote Originally Posted by vesseblah View Post
    No it's not about screen tearing at all, but straight out bugs in the rendering pipeline that causes the GPU to hang every 15-20 frames or so. Screen tearing is caused by the rendering not being synced to screen refresh rate. Two entirely different things.



    Because reading is obviously hard: from the first page you might notice that the article was a follow-up to Win8 only test (because critics thought the Radeon 7900 bugs were caused by bad Win8 drivers) and last page has some comments on the previous results with the different cards tested in the previous article so it's not 1v1.
    Article stated it was the same gpu w/ a comparison to a 660...thats 1v1 I'd like to see 2-3 of each board type bench. I'd also say its obviously hard to read, where I said Vsync handles screen tearing and similar problems that this article is talking about, but hey insult me all you want meanwhile not even remotely reading yourself.


    @Lemmiwink when in the world was a 7970 $100 more than a 680?????
    Last edited by Milkshake86; 2012-12-13 at 02:38 PM.

  7. #7
    Quote Originally Posted by Lemmiwink View Post
    This is actually quite funny. I have a 7970GHZ edition card as well. But I ended up using my GTX680 instead cause I feel it runs games smoother despite having marginally lower frames.

    For HTPC videocard though, the 7970GHZ is a a beast and the GTX680 doesn't even come close. Too bad this has nothing to say for gaming cards....

    And before I get attacked by someone who is obviously biased towards amd: I have bought an AMD card for every generation, this is the first nvidia card I have personally owned since the GTX280. My 7970 is an ASUS Matrix platinum card, which is supposed to be handpicked and it is clearly the best gpu on the market when it comes to frame and in multi screen the GTX680 looks like a 10 year old card compared to it.

    My point here being: So many people say the GTX680 is a waste of money bla bla bla. Iirc the 7970 was a good 120$ more than the GTX680, delivering on average 4-5 frames more per second. It looks better but runs hotter, despite it being a handpicked ROG card.

    The GTX680 is, how I see it, after comparing the two in many games the superior video card if given these criterias: 1080p-1440p 1 screen(Anything more than this and the two cards can no longer be compared). Well optimized engine that doesn't require extreme amounts of video memory (metro2033 is not such a game, don't know why it's tested), not the highest AA(When a game utilizes heavy AA, the GTX680 falls behind by alot.)

    Now, you can say what you will about pricing, but I feel the GTX680 is the first top card from nvidia that has been sold at a reasonable price. GTX580 was way more expensive when released. At least in Norway. Now it is very possible to pick up a GTX680 for around 400 euros, in comparison the Matrix will set you back around 550 euros. At this time, I feel like the Matrix was a dumb investment despite being "best gpu on the market"
    Just about pricing, last time i checked the normal matrix card was 470. Whereas the cheapeast proper cooled 680 is 460. So i geuss it all depends in what country you live, if the 680 is the cheaper card to get. But than i would still get the 670 over the 680 i think, which is still 100 euro cheaper (again, i geuss it depends where you live)

    edit

    Also read the arcticle really fast. The named that reviews look at average fps, and just about every one does that there luckily exceptions. But the main factor which determines the playability of a game is not average fps, but minimum. And i geuss it is somehow related with what the arcticle states.
    Last edited by Zeara; 2012-12-13 at 03:27 PM.

  8. #8
    Quote Originally Posted by Milkshake86 View Post
    Article stated it was the same gpu w/ a comparison to a 660...thats 1v1 I'd like to see 2-3 of each board type bench.
    Quote Originally Posted by Milkshake86 View Post
    but hey insult me all you want meanwhile not even remotely reading yourself.
    Are you picking a fight? Again?

    The first article tested two Radeon 7950's along with bunch of other cards, this one I linked has third R7950 tested because people thought that the R7950 scores were too low in the first place. The conclusion seems to be there's a systematic fault in all three R7950 cards tested.

    Quote Originally Posted by Milkshake86 View Post
    I'd also say its obviously hard to read, where I said Vsync handles screen tearing and similar problems that this article is talking about
    This is screen tearing, it has got nothing to do with very late frames.



    The bug with R7950 cards is that the rendering completely stops for a long time leaving same frame hanging visible causing visual jerkiness.
    Last edited by vesseblah; 2012-12-13 at 05:10 PM.
    Never going to log into this garbage forum again as long as calling obvious troll obvious troll is the easiest way to get banned.
    Trolling should be.

  9. #9
    Quote Originally Posted by vesseblah View Post
    This is screen tearing, it has got nothing to do with very late frames.
    Quote Originally Posted by vesseblah View Post
    The bug with R7950 cards is that the rendering completely stops for a long time leaving same frame hanging visible causing visual jerkiness.
    I actually don't get it. Once you're saying it has nothing to do with very late frames, and in the 2nd quote you're saying it's having a rendering break?

    A frame is meant for refreshing?

  10. #10
    Quote Originally Posted by Faithh View Post
    I actually don't get it.
    The purpose of vsync is to make sure new frame starts drawing at the top left corner of the screen and nowhere else. With vsync turned off the new frame render can start anywhere, for example right in the middle causing screen tear like in the picture posted above.

    With vsync turned on the GPU stops processing the frames for short periods of time to make sure they all line up with top of the screen, vsync off the frames are pushed through as fast as possible. All frames takes roughly the same time to draw from start to finish regardless of if it's on or off.


    The bug/problem with R7950 they spotted in the frame latency test shows some frames arriving extremely late, taking ten times longer to render than most. Of course there is slight difference from frame to frame in the time it takes to complete drawing it as can be seen from the charts, but the spikes are exceptional.
    Never going to log into this garbage forum again as long as calling obvious troll obvious troll is the easiest way to get banned.
    Trolling should be.

  11. #11
    Brewmaster Majesticii's Avatar
    Join Date
    Feb 2010
    Location
    Netherlands
    Posts
    1,352
    Adaptive Vsync <3

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •