Multi GPU Thoughts - Micro Stuttering Redux

[Final8y]

First I would like to mention that it is not the first time that someone brings problems with Multi GPU implementations into the spotlight, as you may already have heard about "micro stuttering". Micro stuttering was described as an uneven frame rate when using two GPUs, something that may happen for other reasons like lack of RAM, software accessing the HDD at the same time, software using the CPU(s), etc. Many things can disturb frame rate but the issue with micro stuttering was that the uneven frame rate seemed to only occur when SLI or CrossFire are in use. PCGH had a look at the problems and published some data a while back and it was an interesting look at the problem. The issue is that they didn't notice the problem's origin. They noticed that frames were arriving at uneven intervals, which mitigates the purpose of a higher frame rate providing better responsiveness, which is what is expected from these setups. Keep this in mind: better frame rate must be equal to better responsiveness, or it's a moot point.
Still, an uneven interval should still provide better responsiveness nonetheless, if the frame rate is going to almost 2x. This is where it gets tricky: looking at just the frame times doesn't show you the real problem that's going on with contemporary multi GPU solutions, the big problem is that the responsiveness is in fact worse!

Before we dwell any further into this issue, we need to look at multi GPU rendering modes and have a more detailed look at how the issue presents itself to the user.

http://www.siliconmadness.com/2011/09/multi-gpu-thoughts-micro-stuttering.html

 

[Rroff]

While I can kinda see his point in regarding to testing SLI for potential problems... 9400GT SLI is not exactly the best of starting places.

Also nVidia didn't try to push SFR at the expense of AFR as he claims, they added it as another alternative method as it produces better results in some titles (tho generally a fairly small number of titles) with 2 cards and makes it easier to render with 3+ cards without hideous latency issues amongst other things - also as mentioned crossfire has its own version of SFR (scissoring mode).

Also RE his conclusion - nVidia has long had the option for the user to change rendering mode if they so wish - tho its usually better to go with the one nVidia has tested...

 

[Final8y]

I have not read it closely at all but some of what he is using is way past its sell by date.

Also i keep seeing the 30fps bench and really most people would be upgrading if they were getting an average of 30fps in multi GPU in most of their games.

 

[Rroff]

Yeah gotta agree there - unfortunatly multi GPU does not work well if your having to resort to it just to get 30fps... where multi GPU is at its best is when you use it to ensure you get as high as possible minimum fps and as close to 60fps (or higher) as possible - multi GPU with low end cards rarely makes sense... mid-range solutions sometimes where its cost effective and one card produces "ok" performance.

 

[Final8y]

Yeah gotta agree there - unfortunatly multi GPU does not work well if your having to resort to it just to get 30fps... where multi GPU is at its best is when you use it to ensure you get as high as possible minimum fps and as close to 60fps (or higher) as possible - multi GPU with low end cards rarely makes sense... mid-range solutions sometimes where its cost effective and one card produces "ok" performance.

Indeed.

 

[cavemanoc]

I have not read it closely at all but some of what he is using is way past its sell by date.

Also i keep seeing the 30fps bench and really most people would be upgrading if they were getting an average of 30fps in multi GPU in most of their games.

From the article:

. If you can bring your solution to the point it is rendering 30FPS per GPU, you will notice artifacts if present or that it simply isn't more fluid with SLI/CrossFire, despite what benchmarks tell you.

I cannot stress enough that this issue may be completely unnoticeable in setups that put out 200+ FPS, so be careful with that during testing.

The whole point of the setup was to allow us to see the effect in action - this is only possible around the 30fps mark - beyond that all you get is the 'feels unresponsive' - which is difficult to quantify.

It is a fascinating read (worth it guys - not just a skim...) - effectively he has concluded that AFR is a compromise that is not working and the manufacturers need to look again to find a better solution, rather than just relying on faster frame rates to mask the issue.

The only bit I thought was weak was the OS issue he identified - Server 2008 had none of the uneven frame-rate issues which everyone assumes is responsible for micro-stuttering, yet even there the ghosting is apparent. Either the uneven rate isn't responsible for the stutter, or the ghosting is caused by another, unidentified, issue - this side definitely needs more investigation.

 

[Rroff]

^^ Yeah problem is that not the only weakness of that type in the article, and while I appreciate its a very complicated issue that no one really fully understands, for someone like myself who has a reasonable grasp of it the inconsistancies in the article just muddies the water more.

 

[Final8y]

From the article:



The whole point of the setup was to allow us to see the effect in action - this is only possible around the 30fps mark - beyond that all you get is the 'feels unresponsive' - which is difficult to quantify.

It is a fascinating read (worth it guys - not just a skim...) - effectively he has concluded that AFR is a compromise that is not working and the manufacturers need to look again to find a better solution, rather than just relying on faster frame rates to mask the issue.

The only bit I thought was weak was the OS issue he identified - Server 2008 had none of the uneven frame-rate issues which everyone assumes is responsible for micro-stuttering, yet even there the ghosting is apparent. Either the uneven rate isn't responsible for the stutter, or the ghosting is caused by another, unidentified, issue - this side definitely needs more investigation.

Everything is a compromise at some level.
Maybe we should have all set our CRT monitors to 30Hz just so we can say how CRT was a compromise of showing motion.

But i don't get any off the issues he has as he is using very old SLi 9400 GT.

 

[Strife212]

I never thought of that, surely AFR is a fraud, it doubles your framerate but if every other frame is the same then what benefit is there?

 

[drunkenmaster]

As he says, life is a comprimise, micro-stuttering is in, and everyones weighing in with their uneducated opinion and ridiculous suggested fixes, yet not one of these sites has had a chat with Nvidia or AMD to ask them about the problem, the fixes, or if they even know about it.

When someone has the balls to phone AMD, ask them to do an interview, go speak to AMD, Nvidia, a hardware, a driver guy and ask them to explain microstutter and how to get rid of it, I won't bother reading, because its just the same crap being repeated over and over again.

This "inducing" of stutter to make a point is even worse, if you buy a high end setup with great framerates and the issue is hidden........ thats good, why would I want to kill performance to bring out a problem.

This is on top of far too many daft reviewers completely ignoring things like vsync, randomly decided to try one or two arbitrary games, or randomly looking at only one or two performance markers, was it Tom's recently who decided to see if "normal" fps gives us a real representation of actual performance, and proceded to do a review with only microstutter comparisons, without ever actually comparing the outcome to a "normal" fps chart, and seeing if the recommendation changed.

THey are all trying to look as inteligent as each other by pointing out they know what microstutter is, then putting out another equally useless article that misses half the main issues.

 

[Crezzlin]

When you say micro stutter is it small stutters in game play or lines that stutter across the screen when moving fast with the mouse etc? I have this problem when running eyefinity but can’t find a fix apart from disable the crossfire and run one card.

 

[Adster3000]

When you say micro stutter is it small stutters in game play or lines that stutter across the screen when moving fast with the mouse etc? I have this problem when running eyefinity but can’t find a fix apart from disable the crossfire and run one card.

Microstutter is uneven frame distribution. So the "gaps" of time between rendered frames aren't similar, causing it to look like a lower FPS than is actually displaying.
If you run without vsync and usually get 100+ fps, it's not very noticeable. If you run with v-sync however (assuming the refresh rate is 60/75Hz,) it is much more noticeable. 60FPS would seem like 30-40FPS, for example.

The "lines across the screen" is called screen tearing. This can be fixed by enabling v-sync.

 

[Crezzlin]

Microstutter is uneven frame distribution. So the "gaps" of time between rendered frames aren't similar, causing it to look like a lower FPS than is actually displaying.
If you run without vsync and usually get 100+ fps, it's not very noticeable. If you run with v-sync however (assuming the refresh rate is 60/75Hz,) it is much more noticeable. 60FPS would seem like 30-40FPS, for example.

The "lines across the screen" is called screen tearing. This can be fixed by enabling v-sync.

Thanks for your reply, screen tearing! , i also get this when i drag a application across the screens fast, im sure i didnt have this problem before so could be a motherboard fault. :s

 

[Rroff]

What happens with vsync is a whole lot more complicated than that, in theory anything enforcing regular frameupdates like vsync makes it smoother, infact the story is a bit more complicated especially with nvidia cards and you can end up trading off smooth performance for increased input lag

 

[Adster3000]

Thanks for your reply, screen tearing! , i also get this when i drag a application across the screens fast, im sure i didnt have this problem before so could be a motherboard fault. :s

If anything, I'd blame drivers. Maybe try older ones?