Microstuttering masterclass

[LifeOnMars]

It's an interesting subject indeed. I do feel though that some people complain of this stutter when they haven't researched the particular game in question. For example, I have to fix the following games with little console tweaks or configs to get them running stutter free and uber smooth. (This may not be the case with other people's setups.)

All Valve games -

I set an fps max of 61 and it gets rid of any stutter/hitching that I get with default settings.

Far Cry 2 in DX10 -

Frame skipping unless a max fps is selected in the games config file....DX9 is fine regardless.

Doom 3 - I forget the command but it's to do with altering the games tic rate.

That's just 3 games off the top of my head and there are many more which take little tweaks to get them running as you would expect. I do feel that alot of people are quick to judge the cards rather than their other hardware or the game engine in relation to their hardware.

 

[cadaveca]

I have a single GPU setup, and I have micro-stuttering. Where is my option?

Single card solutions are NOT prone to microstutter. It is another problem all on it's own that affects single card solutions.


I guess many people don't understand that the term Microstutter was coined for an issue first noticed, really, thanks to 3870x2 GPUs, which didn't really have enough bandwidth to communicate with each other, and drive high resolutions.

This lack of bandwidth lead to a delay in the time it took already rendered frame to end up in the main GPU frame buffer, and led to "hitchy" frame display, although the actual number of rendered frames was more than enough to push a nice healthy 60FPS or more.


I've had a video on my YouTube channel of this stutter happening with the 3870x2's for years now, and over the time that has passed, I've seen mroe users report Microstutter, but very few are actually experiencing it...most are "victims" of other problems, like system bottlenecks caused by mismatched GPU and CPU power.

Microstutter is a real phenomena, and can affect many systems, but there are "protections" built into the OS to deal with it, and AMD's own "Application Profiles" software is yet another way that hardware makers have the opportunity to deal with it.

The Radeon HD 3870 X2’s bridge chip was limited to PCI-Express 1.1, meaning data transfer rates between each GPU—and the PCI-Express x16 interconnect—was limited to just 2.5GB/sec via the bridge chip and a further 0.9GB/sec via the CrossFire interface. This means that the maximum interconnect bandwidth on the Radeon HD 3870 X2 is 6.8GB/sec.

On occasion, AMD believes this lack of bandwidth contributed to some of the performance problems and so the company hopes to resolve this with some additional logic inside the GPU.

- http://www.bit-tech.net/hardware/graphics/2008/08/13/amd-ati-radeon-hd-4870-x2/2

SidePort was introduced to the 4870x2 to fix the problem, if needed, of insufficient bandwidth. The 4870x2's bridge chips also offered dual broadcast functionality, to lessen CPU-to-GPU communication. These were two specific hardware design steps, taken to avoid Microstutter.

 

[qubit]

It's an interesting subject indeed. I do feel though that some people complain of this stutter when they haven't researched the particular game in question. For example, I have to fix the following games with little console tweaks or configs to get them running stutter free and uber smooth. (This may not be the case with other people's setups.)

All Valve games -

I set an fps max of 61 and it gets rid of any stutter/hitching that I get with default settings.

Far Cry 2 in DX10 -

Frame skipping unless a max fps is selected in the games config file....DX9 is fine regardless.

Doom 3 - I forget the command but it's to do with altering the games tic rate.

That's just 3 games off the top of my head and there are many more which take little tweaks to get them running as you would expect. I do feel that alot of people are quick to judge the cards rather than their other hardware or the game engine in relation to their hardware.

Oh yes, Doom 3, what a crock of ****. I remember how the animation would jump slightly at roughly 0.5 second intervals even with vsync on and no dropped frames. I never was able to get this problem out of it, even with all the googling and forum posts I could throw at it. Drove me nuts. Note that this was all before I found the awesome TPU.

In the end, I didn't really like the game anyway (not really into jumpy horrors) and stopped playing it, sidestepping the issue.

 

[qubit]

@cadaveca: I'd be interested to see this microstuttering video of yours. Linky please!

So tell me, have you seen MS when the dual chip card is locked to vsync and the system is rendering without dropping any frames? Because from what I understand, you only see MS when the system is dropping frames, which will be just the same with a single chip card.

 

[cadaveca]

I posted this video a bit over 3 years ago now:

[yt]-5SXeR0torc[/yt]

You'll see the stutter near the middle of the video, when I am facing the wall, and rendering load is actually fairly low, and framerate has increased.

And to me, that's the big sign that you've got microstutter...when performance increases, but perceptual performance is lessened.

 

[Mussels]

please give the exact time you're talking about, i cant see shit just by watching the video.

 

[human_error]

I've had crossfire since x1950xtx crossfire (including a 3870x2, 4870x2 and a 5970) and I have never noticed microstutter at all.

I find the poll is very loaded as the options are either "I have microstutter" or "I have a single GPU" - where's the "I have multiple GPUs and don't get microstutter" option?

 

[qubit]

I find the poll is very loaded as the options are either "I have microstutter" or "I have a single GPU" - where's the "I have multiple GPUs and don't get microstutter" option?

No, my poll is not loaded. All my question asks is how much the phenomenon of microstuttering bothers you, nothing more. It's got nothing to with anyone's particular setup.

Note that the 5th option was added by W1zzard, not me and skews the poll. I did make a noise about it, but it then seemed fairer to leave it in after so many people had voted it.

 

[erocker]

No, my poll is not loaded. All my question asks is how much the phenomenon of microstuttering bothers you, nothing more. It's got nothing to with anyone's particular setup.

Note that the 5th option was added by W1zzard, not me and skews the poll. I did make a noise about it, but it then seemed fairer to leave it in after so many people had voted it.

So does "It doesn't bother me" mean that I have microstuttering and it doesn't bother me or does it mean I don't have microstuttering thus it doesn't bother me? I agree that the poll isn't loaded, it is vague.

 

[human_error]

No, my poll is not loaded. All my question asks is how much the phenomenon of microstuttering bothers you, nothing more. It's got nothing to with anyone's particular setup.

Note that the 5th option was added by W1zzard, not me and skews the poll. I did make a noise about it, but it then seemed fairer to leave it in after so many people had voted it.

Yes it is loaded. There are 3 options for "it bothers me", one for "I ignore it" and one for "I like it" (which is a joke option surely). The final option added by w1zz does give people with 1 GPU something to vote (which is good - people like putting their experiences in polls like this), but there's no option there for "it does not exist in my multi-GPU setup".

The poll may have been made to query how much it bothers people, but due to the massive weighting towards "it is a problem" it gives a disproportionate representation of the amount of setups with microstutter (anyone skimming the thread and looking at the poll options would get the impression that there is always microstutter unless using only 1 GPU).

 

[qubit]

So does "It doesn't bother me" mean that I have microstuttering and it doesn't bother me or does it mean I don't have microstuttering thus it doesn't bother me? I agree that the poll isn't loaded, it is vague.

That's it, I'm gonna strangle someone! erocker, give me your home address and telephone number and we'll arrange that showdown! j/k

It's sort of the first option, but it's not making a statement about any particular system. I simply ask if you've seen the effect (doesn't matter where) then how much does it bother you.

I noticed that no-one "kinda likes it" lol.

I think the question is clear, but if you think it should be rephrased, I'm open to offers. In fact, reading through the thread, quite a few people have suggested different poll options, so feel free to rejig it!

If it is rephrased, then I suggest that all the votes be cleared.

 

[qubit]

Yes it is loaded. There are 3 options for "it bothers me", one for "I ignore it" and one for "I like it" (which is a joke option surely). The final option added by w1zz does give people with 1 GPU something to vote (which is good - people like putting their experiences in polls like this), but there's no option there for "it does not exist in my multi-GPU setup".

The poll may have been made to query how much it bothers people, but due to the massive weighting towards "it is a problem" it gives a disproportionate representation of the amount of setups with microstutter (anyone skimming the thread and looking at the poll options would get the impression that there is always microstutter unless using only 1 GPU).

Nope, it's not loaded - with the original five options. You had 2 pro choices, 2 anti choices and one middle choice. Perfectly balanced.

However, check out my reply to erocker above yours. If you're keen to have it changed, make your suggestions and we'll see what our mods come up with. Can't say fairer than that!

 

[erocker]

That's it, I'm gonna strangle someone! erocker, give me your home address and telephone number and we'll arrange that showdown! j/k

It's sort of the first option, but it's not making a statement about any particular system. I simply ask if you've seen the effect (doesn't matter where) then how much does it bother you.

I noticed that no-one "kinda likes it" lol.

I think the question is clear, but if you think it should be rephrased, I'm open to offers. In fact, reading through the thread, quite a few people have suggested different poll options, so feel free to rejig it!

If it is rephrased, then I suggest that all the votes be cleared.

Nah, I get what you're saying. The way you describe it though, I can't see how anyone would actually like stutter. So, the way the poll was posted initially, there's only one logical answer. I also have no interest in making/changing polls.

 

[qubit]

Nah, I get what you're saying. The way you describe it though, I can't see how anyone would actually like stutter. So, the way the poll was posted initially, there's only one logical answer. I also have no interest in making/changing polls.

You're terribly right about the bold bit. It was a bit of a joke option, in place of the "doesn't bother me very much" option, lol. You know me, I couldn't resist a subtle gag.

 

[LAN_deRf_HA]

 

[Mussels]

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


^ this is very relevant to the topic

 

[cadaveca]

Good link, Mussels.



I'll stil lsay, I think it's impossible to properly measure "Microstutter" without using a device that replaces the monitor. To truly know if differing render times are an issue, we need to know if ther is actually any difference in teh way the data is sent out to the monitor, not just when the frame is actually complete.

 

[qubit]

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


^ this is very relevant to the topic

I've now read some of that article and I think this is issue is important enough for these kinds of measurements to be included in TPU graphics card reviews. It's all those hitched, stuck frames that can make a real difference to the fluidity and playability of a game. Basically, hitches are annoying as hell and must be minimised as much as possible.

This is what I found a couple of years ago when I compared my HD4870 with a much older (and supposedly lower performing) 8800 GTX I got for cheap. The 4870 was indeed faster in some situations (and surprisingly slower in others) but had noticeably more hitches to it than the 8800. It wasn't long until I eBay'd the 4870 and got myself a GTX 285 and haven't looked back. That card gave me a really smooth experience, as does my GTX 580 now.

Whether AMD have improved hitching performance since then I don't know, but I do know that my nvidia cards work very well, so I'm sticking with nvidia.

Incorporating these tests in TPU reviews will help the knowledgeable enthusiast make a much more informed choice for their next video card purchase.

Here we go, from the article:

We're counting through all five of our 60-second Fraps sessions for each card here. As you may have inferred by reading the plots at the top of the page, the Radeons aren't plagued with a terrible problem, but they do run into a minor hiccup about once in each 60-second session—with the notable exception of the Radeon HD 6970. By contrast, the Nvidia GPUs deliver more consistent results. Not even the older GeForce GTX 260 produces a single hitch.

See what I mean?

 

[erocker]

I've now read some of that article and I think this is issue is important enough for these kinds of measurements to be included in TPU graphics card reviews. It's all those hitched, stuck frames that can make a real difference to the fluidity and playability of a game. Basically, hitches are annoying as hell and must be minimised as much as possible.

This is what I found a couple of years ago when I compared my HD4870 with a much older (and supposedly lower performing) 8800 GTX I got for cheap. The 4870 was indeed faster in some situations (and surprisingly slower in others) but had noticeably more hitches to it than the 8800. It wasn't long until I eBay'd the 4870 and got myself a GTX 285 and haven't looked back. That card gave me a really smooth experience, as does my GTX 580 now.

Whether AMD have improved hitching performance since then I don't know, but I do know that my nvidia cards work very well, so I'm sticking with nvidia.

Incorporating these tests in TPU reviews will help the knowledgeable enthusiast make a much more informed choice for their next video card purchase.

Here we go, from the article:



See what I mean?

Only problem is, microstuttering doesn't exist on all multi-gpu setups. That said I've experienced it before with I believe my 4870 setup and my GTX 460 SLi setup. 3870's no problem along with 5850's and 6970's. Since it is something that is noticable, it would be nice if it is at least mentioned in a review if microstuttering is present for a certain test.

 

[qubit]

Only problem is, microstuttering doesn't exist on all multi-gpu setups. That said I've experienced it before with I believe my 4870 setup and my GTX 460 SLi setup. 3870's no problem along with 5850's and 6970's. Since it is something that is noticable, it would be nice if it is at least mentioned in a review if microstuttering is present for a certain test.

Yeah, I'll +1 that.

I'm still not convinced that the so-called "microstuttering" seen with a dual GPU setup is any different from ordinary stuttering seen with one GPU.

Tell me, did you see what you believe to be microstuttering with your 4870 xfire setup with vsync locked and no dropped frames? That would indeed be something different. From what I understand, this microstuttering is only happening on vsync unlocked rendering with the system rendering well under the monitor refresh. You're gonna get noticeable judders there, no matter what, which is why I'm not so convinced about the alleged microstuttering effect.

I have to say, that about 5 years ago or more now, I saw stuttery animation on nvidia cards with vsync locked and fraps showing no dropped frames. It was a weird kind of wobbly effect, hard to describe. Very noticeable and very annoying. ATI back then were buttery smooth, so I switched brands.

This funny on nvidia was a driver glitch, because the older Detonator (remember them!) drivers worked smoothly, while the newer ForceWares didn't. It took nvidia well over 2 years to fix this one.

 

[cadaveca]

I'm still not convinced that the so-called "microstuttering" seen with a dual GPU setup is any different from ordinary stuttering seen with one GPU

It's very simple. MICROSTUTTER is caused by the frames being transferred from one card to another, and is something NOT caused by performance deficiencies. This problem, as i stated earlier in the thread, is what began the entire "Microstutter" conversation.

One a single card, if there is stutter because of a longer time to render a frame, that's a PERFORMANCE DEFICIENCY, and is clearly indicated by a faster version of the same GPU not having the same problem(in the link from Mussels, 6950 had the problem, but 6970 did not = NOT Microstutter).

I think it's important to seperate the problems into CAUSES, and not just an EFFECT, and hence my stance on this. Many things can cause issues like this, and sometimes the problem can be solved, but it's going to be impossible to solve a problem without properly identifying the issue in the first place.

 

[Mussels]

cad: read the article i linked to. microstutter is NOT limited to dual GPU only. Its more common there, but it certainly happens on single GPU's as well. dropped frames results in microstutter, and dual GPU's get lots more dropped frames - they are definitely related, but its not dual GPU (and up) only.


and FYI, that article missed one thing: while they say that ATI cards seem to suffer it more, they only tested current and last gen cards. 8800GTS 320MB for example, was really infamous for its microstutter.

 

[cadaveca]

Did the 640MB cards have the same problem? Can't blame "Microstutter" for poor performance when the poor performance CAUSED the "microstutter". There is a distinct difference.

 

[qubit]

It's very simple. MICROSTUTTER is caused by the frames being transferred from one card to another, and is something NOT caused by performance deficiencies. This problem, as i stated earlier in the thread, is what began the entire "Microstutter" conversation.

One a single card, if there is stutter because of a longer time to render a frame, that's a PERFORMANCE DEFICIENCY, and is clearly indicated by a faster version of the same GPU not having the same problem(in the link from Mussels, 6950 had the problem, but 6970 did not = NOT Microstutter).

I think it's important to seperate the problems into CAUSES, and not just an EFFECT, and hence my stance on this. Many things can cause issues like this, and sometimes the problem can be solved, but it's going to be impossible to solve a problem without properly identifying the issue in the first place.

Maybe you're right, it certainly sounds plausible. This is why I'd like to see an authoritative explanation of this phenomenon. Have a look at the question I asked erocker a few posts ago. If he replies Yes to MS with vsync locked and no dropped frames, then it definitely indicates a dual GPU specific problem and therefore not something you'd see with one GPU only.

But then Mussels has just said that single GPU's show this too, so I really don't know what to think. The thing I'll say though, is that Mussels has a way of being right on anything technical, which adds weight to his argument.

 

[eidairaman1]

major slowdowns in FPS is what ticks me off.