Tempted by some cheap GTX460 SLI action? Might want to check this out first before laying down your

[Rroff]

Well, good for you I guess

Personally I am uninterested in people's opinions on matters which can be quantified numerically. Opinion has no place in mathematical analysis, and microstutter comes down to the analysis of framerate irregularity.

My aren't your grumpy.

I've never seen a properly functioning multi GPU setup that gets a 35% framerate variation with your program in a scenario where a single GPU isn't getting a similiarly high variation in framerate - and I see 10% variations on both kind of setups in some games and very low variations on both in others.

You need a consistant 15-20% variation in frametimes thats happening all the time before anything more than <1% of the population would notice it - 10-15% variation over a 60+ second varied scene raw numerically mean nothing, your numbers don't necessarily quantify anything without other supporting considerations.

 

[An Exception]

I never noticed any stutter running 2 460 1gb,s in Sli... Tried a good few games and was ok no stuttering. ??

What res?

 

[RavenXXX2]

Thread title is misleading to attract attention, there is nothing the 460 SLI is doing wrong that can't happen on another ATI or Nvidia mulit-gpu setup, look at the vid below, maybe I should start a thread on how buyers should be aware of the 5970 as it will microstutter like hell at low FPS and a 480 is a smooth as silk, see where I'm going with this.

 

[MOOGLEYS]

What res?

1920x1080......

 

[RavenXXX2]

NVM

 

[Duff-Man]

My aren't your grumpy.

I've never seen a properly functioning multi GPU setup that gets a 35% framerate variation with your program in a scenario where a single GPU isn't getting a similiarly high variation in framerate - and I see 10% variations on both kind of setups in some games and very low variations on both in others.

You need a consistant 15-20% variation in frametimes thats happening all the time before anything more than <1% of the population would notice it - 10-15% variation over a 60+ second varied scene means nothing, your numbers don't quantify anything - pauses like in that video aren't symptoms of "microstutter" - you have a whole 100ms or so pause every 2-3 seconds.

Wow.

Not only do you not understand what microstutter is, but you are taking it upon yourself to trash my analysis without having any understanding of it. Let me point out a few things for you:

1. Microstutter is the variation away from the INSTANTANEOUS average; that is, the local smoothed average over the surrounding 9 frames. This is constantly varying over the game scene, and the microstutter index reflects the AVERAGE of these values variations from the local mean. If I output a MS index of 20%, then it means a 20% average variation over the entire benchmark. Not some kind of maximum. There will be locations where the variation is far, far more (see some of the extreme plots in my XS thread). So, your "pulled-out-my-ass" criteria for what "more than 1% of the population" would notice are already being met.

For example, THIS is a "worst case" scenario during a roughly 20% MS index run of 3dmark vantage (plot shows frame nnumber vs framerate):




2. My program culls the largest 1% variations from the local frametime average specifically to take account of paging and other macroscopic "hitching" phenomena. The output result is a quantity influenced entirely by microscopic (frame-by-frame) phenomena.

3. You say you have not seen any multi-GPU variation which is not reflected in single GPU results? Did you even bother to read the link I gave you? A couple of random examples from that thread alone:

Crysis 2560, 4xAA:
dual GTX 480 MS index: 15.5%.
Single GTX480 index: 2.0%

Heaven benchmark:
Dual GTX480: 25.9%
Single GTX480: 5.9%
Dual 5870: 30.2%
Triple 5870s: 54%

... There are plenty more in there, and elsewhere on the web. I could repost all the plots of frametime here, but what would be the point?


If I seem "grumpy" then it is because I grow tired of people talking rubbish about things that they either don't understand, or refuse to take the time to read into. I have provided plenty of links for you to look at which will educate you on the matter if you choose to read them. If not, don't bother talking rubbish about things you clearly don't wish to take the time to comprehend.

 

[Rroff]

If your program is building an average based on the local average over the entire length then I apologise. I do understand what microstutter is I mis-understood how you were analysing it.

You still can not compare say single GTX480 against dual GTX480 and say it microstutters (any worse than a single card) - you need to be comparing against very similiar performance i.e. GTX460 SLI against a single GTX480. When I compared GTX260 SLI against a GTX470 I saw very similiar levels from both in my normal scenarios i.e. RE5 benchmark, heaven, etc.

 

[arknor]

someone made a tool that measures micro suttering i saw it posted on another forum, anyone who has an sli setup should google for it.

people ran it on single cards and still had slight micro stutter readings

 

[Rroff]

someone made a tool that measures micro suttering i saw it posted on another forum, anyone who has an sli setup should google for it.

people ran it on single cards and still had slight micro stutter readings

your thread reading is worse than mine

I didn't realise duff-man made the tool but I used it awhile ago with: 7950GX2, 8800GT SLI, GTX280, GTX260, GTX260 SLI and didn't see multi GPU giving any significantly different levels of microstutter between similiar performance setups regardless of the number of GPUs. Also tested is when I swapped from GTX260 SLI to a GTX470 as the 470 feels far more "consistant" but very little difference to the microstutter levels. Wish I'd posted my findings at the time now - may have to put the GTX260 back in later to run some numbers.

 

[Scougar]

I see scared of bogey micro-stutter man, so it must be a pointless thread, and not one that educates others about the effect that micro-stutter can bring to potential gamers, just because you know what it is, doesn't mean every does.

So why didnt you detail the thread as general and not 460 specific? I see thread maker attention seeker.

(read the following VERY slowly)

The point of the video was to capture the difference between a single and muti GPU setup running at the same frame-rate, not to compare who has the most FPS.
They both were running at 30 FPS while 5870 was just about playable the 460SLI was hideously stuttery!

But to answer your question, at whatever res your play at (except extreme high res), it's very probable a fast single GPU is the best choice for smoother gameplay than two slower cards that give you modestly more FPS.

And it's the last point that makes this thread useful, as it will help someone out there avoid a costly mistake by making an informed decision even if you have already heard of the issue.

I agree with you about single card over dual etc, and what i recommend, but you are still scare mongering. Costly mistake? Very unlikely.

I agree with some of your points but you have approached this all wrong.

 

[tdream]

Actually, so long as you're not right on the threshold, then double-buffer vsync will completely "fix" microstutter.

Where's the evidence of that?

On what threshold? Stick with numbers, it's less vague. Do you mean almost 60 fps but not quite.

 

[Duff-Man]

Where's the evidence of that?

On what threshold? Stick with numbers, it's less vague. Do you mean almost 60 fps but not quite.

As far as evidence goes, I've posted a link to the XS thread I made a few times now. I have a similar thread on hardforums. There are a few examples with vsync, but I can run a couple more later on tonight if you like.


As for the "threshold": With double-buffer vysnc this refers to any of the points of discrete change in output framerate. So, for a 60hz screen, that would be 60fps, 30fps, 20fps, 15fps etc (basically integer divisions of the maximum refresh rate).

- With no microstutter, you will be at 30fps until your GPUs can output 60fps, at which point you will jump to 60fps.
- With microstutter, there will be a small window whereby you are able to output *just above* 60fps, but the irregularity of output means that you don't get a solid 60fps; instead you jump between the two (i.e. a 1/60th of a second gap between frame1 and frame2, then a 1/30th of a second gap between frame2 and frame3, then 1/60th, then 1/30th etc).

This is a very difficult situation to get hard numbers on, because it is hard to recreate consistently for testing. But it is there. Luckily, once you get above something like ~61-62fps or so, the gap caused by the card(s) 'resting' while the framebuffer is full drowns out any irregularity in framerate output. From this point on you have a near-perfect 60fps framerate displayed, in sync with the monitor. Any evidence of microstutter disappears.

The same thing will happen on the threshold of 20fps to 30fps, 20 to 15, 15 to 12 etc etc.

 

[An Exception]

Thread title is misleading to attract attention, there is nothing the 460 SLI is doing wrong that can't happen on another ATI or Nvidia mulit-gpu setup, look at the vid below, maybe I should start a thread on how buyers should be aware of the 5970 as it will microstutter like hell at low FPS and a 480 is a smooth as silk, see where I'm going with this.

Raven for the last time, stop trolling this thread and generally behaving like a fanboy dishing out crackpot accusations, you just haven't got the IQ to pull it off.
Yeh the 5970 does micro stutter, but it's not about Nv Vs Amd it's about Multi Vs Single GPU, but you just posted up a benchmark showing the 480 at much higher FPS than the 5970 in heaven.
Hardly a fair test to compare Multi GPU Vs single GPU and the effects of micro stutter...

 

[An Exception]

So why didnt you detail the thread as general and not 460 specific? I see thread maker attention seeker.

Titled with 460's because everyone is talking about them being sooo good value with great scaling and FPS compared to GPU's like 5870/50's & 480/70's when in reality the later actually offers better REAL performance & therefore more value.
And of course they are the cards in the video...


Edit:

Costly mistake? Very unlikely.

Well it is if your having SLI issues with the 460's, but also because for less money a single GPU that cost's less would probably offer better performance due to the lack of micro-stutter.

 

[RavenXXX2]

So why didnt you detail the thread as general and not 460 specific? I see thread maker attention seeker.

This, I see no complaints of stuttering with 460 SLI on here or anywhere else, so the thread title has no relevance other than to bait people. Picks a vid of youtube using a setup and res that 99% of 460 users will never use, who's to say that the 460's were not over their 1GB buffer thus causing hitching, here's a vid showing 460 SLI in action with crysis, butter smooth, fail thread.

 

[MOOGLEYS]

I have no idea what Micro stutter is but at 1920x1080 with all settings max i noticed no stuttering at all in nearly all of the latest games....

 

[Rroff]

Microstutter is when your ostensibly outputting say 30fps - but the actual interval between those 30 updates in 1 second isn't even - if theres a big enough irregularity it won't look smooth and even more noticeably your input might not work quite as you expected.

Thing is the actual % of stutter alone doesn't matter... 35% "microstutter" when your longest update interval delay is 5ms is probably unnoticeable to 99% of the population.

 

[MOOGLEYS]

Microstutter is when your ostensibly outputting say 30fps - but the actual interval between those 30 updates in 1 second isn't even - if theres a big enough irregularity it won't look smooth and even more noticeably your input might not work quite as you expected.

Ah i see, Tbh my eyesight is crap so i doubt i would notice anyway...

 

[An Exception]

This, I see no complaints of stuttering with 460 SLI on here or anywhere else, so the thread title has no relevance other than to bait people. Picks a vid of youtube using a setup and res that 99% of 460 users will never use, who's to say that the 460's were not over their 1GB buffer thus casing hitching, here's a vid showing 460 SLI in action with crysis, butter smooth, fail thread.

Allot smoother, but I can still see it stutter.

Single card looks ok to me.
Still, a clocked 5850 should be smoother.

5850

[/QUOTE]

 

[MOOGLEYS]

Allot smoother, but I can still see it stutter.

Tbh i can see a tiny amount of stutter in that video.. If thats what all the fuss is about then i,m kind of shocked...Its hardly a deal breaker...... lol...