7970 x2 Crossfire Microstutter unbrearable

[BrightCandle]

That's the problem with "microstutter", though. Lot's of people complain, and there's never any video to help determine if the issue is MS or not.

He hated his 5970. I have a 5970 and have no issues at all. If I play games at too high of settings I Get lag, not microstutter. I've posted videos of my experiences.

Certainly if the current round of reinstalls doesn't fix this I'll record using Fraps/DXTory at 60 fps. I likely wont be able to share on youtube the results because it will reduce the frame rate but I can always dropbox it.

 

[imaheadcase]

People have to realize that everybodys definition of smooth is different. I've had friends marvel over my sli/crossfire setups and comment how smooth the gameplay was while I was seeing every little frame stutter driving me insane. I have also seen people playing game @ 25fps telling me how great their system plays.

I have/had

6600 gt sli
7800 gt sli
7800 gtx sli
8800 gt sli
gtx 460 sli
gtx 480 sli
x1800 crossfire
4870 tri fire
5870 crossfire

They all have had microstutter. It is fact.



http://techreport.com/articles.x/21516/11

That article is pretty good and is a must read.


If *you* don't see microstutter I envy you because I wish I didn't. I would love to have more GPU power for my u3011


*edit*

The recommendation to limit frame rates to just below the minimum you get is the most insane thing I have ever heard. Why spend double the money on cards to do such a thing?

I got SLI GTX 580 and never seen mico stutter at all. It's a fact.

No reason to limit fps.

 

[boxleitnerb]

Because you have no direct reference point for comparison.

If I had the money, I would put up two machines, one with SLI, one without it and have them run the same game(s) at (roughly) the same fps. The difference in smoothness would be quite clear I can imagine.

 

[acefightr]

Single card for the win.

 

[BallaTheFeared]

It's pretty easy to disable your second card. Not everyone went straight to SLI/CF right off the bat either, upgrading from a single card is more common.

I'm well aware of MS (as are many other people), and when it occurs. So far in my testing/playing it has only occurred for me in Heaven, nothing else.

I went from 1920x1080 to 5800x1080, took a huge fps loss, and still even with a much larger view picture and large fps loss I do not notice MS in any games I currently play.

Toms did a review of Nvidia and AMD cards and showed even at 1200p it required 6950CF to get rid of most MS, while 3x 6950CF at 1200p produced smoother frames than CF did, thus showing even CF 6950 for Tom's was producing MS.

Drivers are key, the better your drivers the better you can eliminate or mask it's occurrence.

 

[Grooveriding]

http://www.youtube.com/watch?v=XWwGkz_Zx5A

This can give some idea of it. If you had two identical systems and monitors in front of you, but for one single gpu and the other multi, you wouldn't be able to miss it if they were both looping the same run at the same fps. You just get used to it when you use it all the time (some people never do).

 

[BallaTheFeared]

I don't see any MS there, I see it here though...

590 SLI

http://www.youtube.com/watch?v=qGXGtuUVQgE

 

[WMD]

Having occassional MS is an acceptable trade off for higher absolute performance not possible with a single card. 2 mid range cards in CF gives you much better performance/ dollar over a single card solution.

 

[Grooveriding]

Having occassional MS is an acceptable trade off for higher absolute performance not possible with a single card. 2 mid range cards in CF gives you much better performance/ dollar over a single card solution.

Value wise for sure, but gaming quality is much better in a single GPU. You can slap a couple mid range cards together and get 10 or 20% more perf than a single gpu beast, but I'd rather have that single gpu than that gain. Generally the single gpu card will play your games more smoothly and deliver a better experience.

Ignoring price multi gpu is only worth it if you need a good chunk more power than any singke gpu card has. Of course it is hard to ignore price.

 

[mariyoshikong]

As someone who has done a SIGNIFICANT amount of time researching microstutter with crossifre (2, 3 and 4 GPUs) there are a few key points I would like to add to this discussion that may clear up a few things.

1) What most people call microstutter is really just a difference in render times from one frame to the next. i.e. 10ms 25ms 10ms 25ms

2) The result of this is that games sometimes appears to be running at a much lower frame rate than what is being reported. For me, this starts to become noticeable once the frame time maxima are above 30ms. Everybody has different eyes, and this WILL vary from person to person.

3) In every case I have seen, microstutter shows up when the GPUs are nearing 100% load. As the load on the GPUs decreases, the difference in frame times also decreases until it is effectively gone (looks just like running a single GPU). This is why using a frame rate limiter fixes microstutter for most people. It is artificially reducing the GPU load, and in turn the microstutter.

4) The notion that adding a third GPU will eliminate microstutter is wrong. By adding a third (or fourth) GPU to the equation, you start moving in the direction of becoming CPU bound. Once this starts to happen, GPU usage begins to decrease, and so does microstutter. If you look at the data from the Tom's Hardware microstutter article (where they state that adding a third GPU eliminates microstutter) you will notice that their testing is all done at 1920x1080 using Call of Juarez. At this resolution, 3 GPUs are almost certainly CPU limited and thus no microstutter.

If i get a chance this weekend I will post some of my data and graphs to help illustrate these point better.

 

[boxleitnerb]

1) true. I would like to add that at higher fps there still will be differences between SLI/CF and a singleGPU setup, but they will be less noticeable. Only at lower fps where the SLI/CF setup actually microstutters it will be easily apparent
2) agreed
3) yes and no. I agree with everything you said but the frame rate limiter also lessens/fixes microstutter if the overall framerate doesn't change. For instance if you get 31fps in a scene and have microstutter and you limit the fps to 30 it will be more smooth, yet the GPU load is almost the same.
4) yes I never get why people always point to this article when the conclusion is clearly wrong. Only one set of data was presented, this is in no way conclusive. Further testing has to be done on this subject, but the 2/3/4 GPUs should render at the same framerate. You could reduce microstutter with 2 GPUs as well by reducing resolution or detail, thus lessening GPU load. No magic here.

 

[Dadofamunky]

the recommendation to limit frame rates to just below the minimum you get is the most insane thing i have ever heard. Why spend double the money on cards to do such a thing?

 

[boxleitnerb]

To increase image quality. If you're playing on relative tame settings for such a setup or a shooter, then it is not really necessary.

But if you want to play with all the eyecandy and your fps are quite low anyway, why not? What is better:

a) 17fps@single GPU = stuttering
b) 33fps@SLI/CF = stuttering
c) 30fps cap@SLI/CF = smooth

I vote c)

 

[antihelten]

If you look at the data from the Tom's Hardware microstutter article (where they state that adding a third GPU eliminates microstutter) you will notice that their testing is all done at 1920x1080 using Call of Juarez. At this resolution, 3 GPUs are almost certainly CPU limited and thus no microstutter.

Actually at those settings 3 GPUs are almost certainly not CPU limited, seeing as 4 GPUs achieve 15-20% better performance than 3 GPUs.

Furthermore in the article, they actually mention testing their setups across multiple different games with the exact same results.

With that said, we definitely need more data before we can declare 3 or more GPUs as the panacea for microstutter.

 

[imaheadcase]

To increase image quality. If you're playing on relative tame settings for such a setup or a shooter, then it is not really necessary.

But if you want to play with all the eyecandy and your fps are quite low anyway, why not? What is better:

a) 17fps@single GPU = stuttering
b) 33fps@SLI/CF = stuttering
c) 30fps cap@SLI/CF = smooth

I vote c)

Then you should not be running SLI/CF to begin with.

 

[omeds]

TRI SLI/CFX has less microstutter than dual or Quad.

Vsync helps eliminate microstutter.

Capping fps helps eliminate microstutter. Use a 125fps cap on both 60hz and 120hz monitors for best results.

Being slightly CPU bound helps eliminate microstutter.

Enjoy your MGPU system.

 

[Keysplayr]

Then you should not be running SLI/CF to begin with.

This comment makes no sense. In this example, if he wasn't using SLI/CF he would have an unplayable 17fps stutter show. Why would he not want, and really need to run SLI/CF?

 

[Lonbjerg]

Maybe I'm too old but I've been staring at CF and SLI rigs since the early days and never notice microstutter. When things run at constant 60+ fps its so smooth i can't tell...

Its like mouse lag with CF/SLI, some ppl i know claim they notice it..

Since you notice it in every game, i guess you are just more sensitive to it and thus multi cards are not for you.

Microstutter can occour evenat +100 FPS.
It's not about the frames per second...it's about the interval between frames and that time fluxuation:

http://techreport.com/articles.x/21516

 

[mariyoshikong]

Actually at those settings 3 GPUs are almost certainly not CPU limited, seeing as 4 GPUs achieve 15-20% better performance than 3 GPUs.

Furthermore in the article, they actually mention testing their setups across multiple different games with the exact same results.

With that said, we definitely need more data before we can declare 3 or more GPUs as the panacea for microstutter.

That is a good point. I don't have a copy of the game to test, but my guess is that losses from crossfire scaling going from 2 -> 3 and 3 -> 4 GPUs is resulting in the load per GPU being reduced. i.e.

2 GPUs 100% + 100% = 200% of single GPU performance

3 GPUs 90% + 90% + 90% = 270% of single GPU performance

4 GPUs 80% + 80% + 80% + 80% = 340% of single GPU performance

Even though we see performance increase, the overall load on the GPUs is decreasing. I have seen this happen in almost every game where performance scaling is less than ideal.

When I went from 2 GPUs to 3 GPUs I still saw microstutter in every single game that I did before, most notably Crysis, Battlefield 3 and Skyrim.

 

[omeds]

Thats strange, 3 GPU's has less ms than dual. Was it SLI or CFX?

 

[mariyoshikong]

All of my testing is with crossfire.

 

[The2ff]

I've used SLI and Crossfire exclusively since the 8800 days. On launch day, I bought 2 7970's to continue that trend. Though my framerate went through the roof in BF3 with everything enabled, it still felt 'choppy', probably more noticeable because of the cash outlay and expectations I built in my head.

I decided to run a single 7970 in BF3 for comparison and I'm pretty pissed off because it is SOOOOOOOOOOOO smooth. My frames are ~55 with everything on including 4x msaa and it feels/looks incredible compared to getting 90fps with crossfire enabled.

For so many years I figured I simply needed more frames and wondered how any review site could say 45fps is playable, but now I see... and now I have a 7970 sitting in 3watt low power state doing absolutely nothing.

 

[Vesku]

... and now I have a 7970 sitting in 3watt low power state doing absolutely nothing.

Can always get more monitors. ;p

 

[The2ff]

Can always get more monitors. ;p

Already running 3 on the first card, but you make a good point.

 

[Tempered81]

I've used SLI and Crossfire exclusively since the 8800 days. On launch day, I bought 2 7970's to continue that trend. Though my framerate went through the roof in BF3 with everything enabled, it still felt 'choppy', probably more noticeable because of the cash outlay and expectations I built in my head.

I decided to run a single 7970 in BF3 for comparison and I'm pretty pissed off because it is SOOOOOOOOOOOO smooth. My frames are ~55 with everything on including 4x msaa and it feels/looks incredible compared to getting 90fps with crossfire enabled.

For so many years I figured I simply needed more frames and wondered how any review site could say 45fps is playable, but now I see... and now I have a 7970 sitting in 3watt low power state doing absolutely nothing.

Sounds like you found out what microstutter is