About Microstutter and progress to alleviate it

[cmdrdredd]

HurleyBird made a thread that got closed about an old (1year) TR article regarding something Nvidia calls "Frame Metering" that will hold a frame and not push it right away so it creates a slightly smoother experience and help with microstutter. Does AMD use something like this as well? I had never heard of this before but it sounds interesting and if it could be built upon, it would really help make multi GPU setups better. Sometimes a single GPU no matter how fast just isn't enough for everyone, as many of you no doubt know personally. SLI and X-Fire are becoming more popular options but microstutter always comes into the discussion at some point. The latest is that SLI offers slightly less than X-Fire, but it's never really gone. Has anyone heard of any progress to eliminate it? Can it even be eliminated?

I think Microstutter is as relevant to a multi GPU setup as scaling is. You can have 100% scaling across 4 GPUs and have a ton of stutter and it makes all your max and average FPS seem worthless because it feels like a mess.

Thoughts?

 

[imaheadcase]

Never experienced it with SLI 580s so couldn't tell you. Its prob a physical problem and not a hardware one. Some people are just more sensitive to it I suppose. I would suggest a single card setup for those.

 

[SirPauly]

Having three or four GPU's actually helps alleviate or helps curb micro-stuttering. This was an outstanding article from Alex, when he was at Rage3d -- now is the editor at Beyond3d.

http://www.rage3d.com/reviews/video/ati4870x2cf/index.php?p=2

I had some debates about micro-stuttering with Alex in the past and was so glad to see him investigate it and offer that article.

Now, Kepler, suppose to have hardware frame metering and tried to get some official information in the GTX 690 Live Review and received something like this from nVidia, " Today we're not ready to officially discuss hardware frame metering!

 

[cmdrdredd]

Ill check out that article thanks. Maybe nvidia and AMD are both working on stuff internally and not disclosing too much yet because it may be something of a game changer if done right?

 

[KingFatty]

Something about "holding" a frame makes me think that it would cause the benchmark scores to be lower when it comes to FPS?

I guess I wonder if by making microstutter less apparent, you'd be hurting the FPS scores, which would be ripe for criticism in all the testing websites who'd scream bloody murder that the FPS scores are lower.

It's like improving a subjective experience to help microstutter vs. causing a tangible/objective decrease in apparent FPS scores?

I guess I'm possibly creating a false dichotomy thinking you can't improve one without hurting the other, but it seems to me like this is a case of avoiding "taking your medicine" because, even though eliminating microstutter would improve the experience, it would be a short-term loss in apparent FPS?

 

[cmdrdredd]

I dunno if the fps changes because a single frame is held. It likely happens too fast to make a difference.

 

[GotNoRice]

I used 2x 4850, followed by 2x 4870x2 before I went to a GTX680, and eventually 2x GTX680. The only time I ever noticed micro-stutter with the 2x 4850 or 2x 4870x2 setup was when the FPS dropped below 60 for whatever reason. Switching from that Quad Crossfire setup to a single GTX680 wasn't some sort of epiphany of smoothness either; it pretty much felt like a side-grade.

 

[Silverforce11]

I think Microstutter is as relevant to a multi GPU setup as scaling is. You can have 100% scaling across 4 GPUs and have a ton of stutter and it makes all your max and average FPS seem worthless because it feels like a mess.

Thoughts?

Thoughts are: if you notice microstutter and it bothers you, don't use CF. Some of you may also notice MS on SLI setup as well.. so do your research/testing before committing to multi-GPU.

I don't notice it, so i've used CF in the past and it was a great perf/$ setup.

 

[cmdrdredd]

Thoughts are: if you notice microstutter and it bothers you, don't use CF. Some of you may also notice MS on SLI setup as well.. so do your research/testing before committing to multi-GPU.

I don't notice it, so i've used CF in the past and it was a great perf/$ setup.

Thats a fair take. However as more and more people want performance a single gpu just cannot give at higher resolutions we begin looking at sli or xfire so microstutter is a real concern for many. Like you I dont get bothered by it but I am interested in any ways to make it less noticeable for the future.

 

[Silverforce11]

Thats a fair take. However as more and more people want performance a single gpu just cannot give at higher resolutions we begin looking at sli or xfire so microstutter is a real concern for many. Like you I dont get bothered by it but I am interested in any ways to make it less noticeable for the future.

The problem is uneven temporal frame rendering.

Thus, timing the output/display of the processed frame to even-up the gap between frames should work, and it does: NV's frame metering seems to solve most of the problem for a lot of users.

But when some ppl claim to notice mouse input lag, 2ms monitor refresh limitations etc etc.. i dunno. Maybe these ppl just need to wear "herp derp normal vision" glasses??

 

[BrightCandle]

Having owned a pair of 7970's and now 2x 680's I can tell you comparatively NVidia's SLI stutters less. The fraps frame time captures show quite clearly that in the grand majority of games AMDs cards stutter and do so at nearly 100% of frame time (that is alternate frames take twice as long to render) which is basically as bad as you can get. NVidia's SLI does stutter, but its not as severe and it doesn't happen in anywhere near as many games.

In practice for me it meant that the AMD cards were unplayable at anything other than exactly 60 fps. If it dropped one 1 frame the game was nearly unplayable.

With NVidia however it feels smooth down to about 45 fps in many games. With a single 680 its smooth to 30 fps in those same games, but with SLI that feels like its sub 30 and not performing well. So there is clearly a small impact of stutter present and the fraps times show that dual cards does produce a less smooth frame rate but it doesn't look all that bad.

Neither solution eliminates it yet, but NVidia's solution from what I have seen is better than AMD.