Microstutter in latest-gen cards - examples included

[Zaxis01]

I'm running 2x 5870's in cf and ran AVP for about 30-40 secs and these are my results.

[janolle]

A nifty little program. Sure beats doing tens of graphs on Excel just to convince a bunch of people that avg FPS means nothing unless you are sure that no microstuttering occurs.

I just deleted a really big bunch of framtime data I had on my harddrive. I had a few really wild ones using Unigine, Stone Giant and Stalker:CoP.

Here's a graph of one run I had (Single 5870, converted to fps from frametimes):

The stutter percentage would probably be pretty low on this one but it sure is very visible to the eye as it periodically looks like a 20-40% frop in framerates. Alot of people don't believe their eyes because the FPS number stays the same.

[Pantsu]

FC2 used to be one of the most microstuttering games out there, it was obvious even on a single GPU. Eventually Nvidia did fix it though in some driver release, so it's at least in part fixable if they'd bother to, but most likely they won't cause it might lower the "real" fps result. I hear Stalker is one good (bad) example of ms too.

Lot's of people are probably considering to go GTX 460 SLI, but if microstutter takes that 20% off your fps it's not really as good as GTX 480 now is it? Of course it's dependant on the game and drivers and whatnot. All in all if you go SLI or CrossFire, you should go big time, as in GTX 480 SLI, not some dual 5770 setup where you'll end up with low fps and high stutter.

[janolle]

Lot's of people are probably considering to go GTX 460 SLI, but if microstutter takes that 20% off your fps it's not really as good as GTX 480 now is it? Of course it's dependant on the game and drivers and whatnot. All in all if you go SLI or CrossFire, you should go big time, as in GTX 480 SLI, not some dual 5770 setup where you'll end up with low fps and high stutter.

I had a 5870 crossfire configuration for a while and it was actually worse than a single card. When looking at frametimegraphs it was clear that the experienced fps couldn't possibly be more than 20% higher compared to a single card. What made it worse was the randomness of the stuttering. To me crossfire was a huge downgrade... higher numbers but even worse experienced performance

[Pantsu]

Here's a shot of one my old FC2 .csv files. This was done on a single GPU, and was the worst case of a dozen tests.



You can imagine what it would look like on dual gpu setups. Somewhere between April and July though Nvidia addressed the problem, and now it's smooth with max 5% difference.

[Arseface]

Thanks for the results guys

Interesting to see some real microstutter from single GPU setups, even if it was due only to a game-specific driver issue. I think there have been similar problems with Fallout 3 in the past (I don't know if they have been fixed by now or not). And yes, I completely agree that the whole "SLI 260s vs a single GTX480" issue is the place where microstutter comes into play the most. It really does make a difference to the price : performance payoff of the two setups. When you're adding a second GTX480 it's just a case that you're not getting quite the performance increase you expect, but the bump is still significant. But I would never double up slower cards - the real-world performance will always be better with a single faster card.

I'm running 2x 5870's in cf and ran AVP for about 30-40 secs and these are my results.
http://i51.tinypic.com/2lt6l95.png

Great to see that there are at least some circumstances where the microstutter from 5870 x-fire setups is minimal I found a similar thing with the GTX480s on crysis 1920*1200.

Any chance you could run the heaven benchmark?

Although.. are you sure your formula isn't buggy? The "Global Average Timeframe" is pretty high in that single card run. It's probably due to the low fps, but never hurts to ask

You can see that the global average frametime and framerate are equivalent:

framerate(fps) = 1000 / frametime(ms)

[Arseface]

Here's a graph of one run I had (Single 5870, converted to fps from frametimes):

The stutter percentage would probably be pretty low on this one but it sure is very visible to the eye as it periodically looks like a 20-40% frop in framerates. Alot of people don't believe their eyes because the FPS number stays the same.

Wow, that looks terrible...

And yes, that would have a relatively low microstutter index, particularly as I cull the worst 1% of frametime variations from my formula (in order to account for paging etc).

Not quite sure what's going on there if it's a single GPU run. Obviously some major issue with whatever game that is. Either that or constant DRAM to VRAM paging?

[janolle]

Not quite sure what's going on there if it's a single GPU run. Obviously some major issue with whatever game that is. Either that or constant DRAM to VRAM paging?

I have no idea what causes that but a closer look shows that it's almost excatly the same thing as microstuttering in cf (a slower frame followed by a quick one). Microstuttering isn't just a multi-gpu problem... although it's ALOT worse with 2 or more GPUs.

Heres a comparison of single vs. cf in Stalker:CoP.

[Carfax]

OK I did another run, and this one seems to be much more accurate. I think I left FRAPS running after the first benchmark had actually finished (exit screen) which was why my results were whacked!

Anyway, same settings as the first one, Heaven benchmark (2.0) @ 2560x1600 4xAA 16xAF, extreme tessellation etc, with 480s overclocked to 850/1700/4200.



I can't say I noticed any microstuttering with the naked eye. The benchmark appeared to run quite smooth.

I also had my max pre rendered frames set to 6. As I mentioned earlier, if you're running SLi, it's best to have that setting at either 5 or 6 to get the best gaming experience and performance.

Makes everything a lot smoother...or at least for me it does.

[janolle]

OK I did another run, and this one seems to be much more accurate. I think I left FRAPS running after the first benchmark had actually finished which was why my results were whacked!

Anyway, same settings as the first one, Heaven benchmark (2.0) @ 2560x1600 4xAA 16xAF, extreme tessellation etc, with 480s overclocked to 850/1700/4200.

http://a.imageshack.us/img714/7959/heavena.png

I can't say I noticed any microstuttering with the naked eye. The benchmark appeared to run quite smooth.

I also had my max pre rendered frames set to 6. As I mentioned earlier, if you're running SLi, it's best to have that setting at either 5 or 6 to get the best gaming experience and performance.

Makes everything a lot smoother...or at least for me it does.

I'm not sure if fraps measures what you see on your screen when using pre-rendered frames. It's possible that it measures the output before the buffer. Pre-rendering will of course smoothen the experience but it also causes lag that most gamers don't like. Renderin 5 frames ahead will cause a 100ms lag at 50fps and that's huge!

[Carfax]

I'm not sure if fraps measures what you see on your screen when using pre-rendered frames. It's possible that it measures the output before the buffer. Pre-rendering will of course smoothen the experience but it also causes lag that most gamers don't like. Renderin 5 frames ahead will cause a 100ms lag at 50fps and that's huge!

I only notice more lagging or jerkiness when I go above 6. Setting it to 5 or 6 though, makes my gaming experience smoother with no noticeable lag compared to the default value of 3.

I'm sure the type of CPU you have makes a difference as well. I doubt using such high values with dual cores (or S775 quadcores) for instance would improve gameplay. Most likely, it would degrade it.

[janolle]

I'm sure the type of CPU you have makes a difference as well. I doubt using such high values with dual cores (or S775 quadcores) for instance would improve gameplay. Most likely, it would degrade it.

I don't mean to be rude but could you explain this to me?

I don't see how a framebuffer could be affected by cpu performance in any way. As I see it it's a simple matter of framerates.

[FinBenton]

More cpu-cores -> more microstutter. When use only 1 core, you get almost none of microstutter.

[Arseface]

More cpu-cores -> more microstutter. When use only 1 core, you get almost none of microstutter.

Well this is true, but only because using a single CPU thread you'll almost always be CPU limited. And as I already demonstrated, when the CPU is the limiting factor then the degree of microstutter is always low, because the GPUs are waiting for the CPU to finish its workload every frame, and therefore sync to the regular output from the CPU.


Case in point; Crysis 2560*1600, 4xAA with only a single CPU thread active:




As you can see, even at this resolution the GPUs were not at full load. And the result (37fps) is slower than the eight-thread result I posted earlier, even taking into account microstutter (apparent framerate 44.9, down from 51.9).

I'm sure you could achieve the same effect by under-clocking the CPU, but neither is very attractive as a solution.

[Carfax]

I don't mean to be rude but could you explain this to me?

I don't see how a framebuffer could be affected by cpu performance in any way. As I see it it's a simple matter of framerates.

Hehe, I'm not sure I'm qualified to explain it to you.. I was merely postulating in truth, since SLI rigs (Crossfire too in fact) perform much better on Core i7 than on previous CPU architectures.

Click here for an example.

And the above link was done with 280s. Can you imagine what it's like with 480s, which are roughly twice as powerful?

I suppose it has to do with the CPU's abilities to "feed" the GPUs. The multithreaded nature and high data throughput of the Core i7 allows it to send data faster to multiple GPUs so they don't have to wait.

Increasing the amount of framebuffers increases the CPU workload, but if the CPU isn't fast enough to keep up, then I would assume performance would suffer....much as it does when I go above 6 framebuffers.

Using high end cards like the 480 or HD 5870 in multi GPU configs with anything less than a Core i7, hampers their performance though to be quite honest..

[janolle]

Hehe, I'm not sure I'm qualified to explain it to you.. I was merely postulating in truth, since SLI rigs (Crossfire too in fact) perform much better on Core i7 than on previous CPU architectures.

Click here for an example.

And the above link was done with 280s. Can you imagine what it's like with 480s, which are roughly twice as powerful?

I suppose it has to do with the CPU's abilities to "feed" the GPUs. The multithreaded nature and high data throughput of the Core i7 allows it to send data faster to multiple GPUs so they don't have to wait.

Increasing the amount of framebuffers increases the CPU workload, but if the CPU isn't fast enough to keep up, then I would assume performance would suffer....much as it does when I go above 6 framebuffers.

Using high end cards like the 480 or HD 5870 in multi GPU configs with anything less than a Core i7, hampers their performance though to be quite honest..

Pre-rendering of frames simply dealys the output a certain amount of frames. The cpu has very little to do with that.

A faster cpu (or lower detail settings) may get higher framerates and that will reduce the lag that buffering causes. Even at 100fps the lag with 6 pre-rendered frames is unacceptable for games that require instant response.


At this time it's a choice between instant response or microstutter. Let us hope that it will work better in the future.

[Arseface]

At this time it's a choice between instant response or microstutter. Let us hope that it will work better in the future.

I found only a minor improvement by increasing the number of frames rendered ahead (see above).

As far as I see, vsync is the only way to reduce microstutter. Which, I guess, also increases response time (though not by as much). So it's a tradeoff between microstutter, or reduced performance and somewhat increased response time.

[demonkevy666]

I had heaven in windowed mode and I noticed one sit was a bit smoother then other in crossfire. :/
anyone know where I can change the driver to do super tiling in stead of AF rendering ?

seem like it dose super tiling in windowed and AF in full screen.

[i5L4ND]

I had heaven in windowed mode and I noticed one sit was a bit smoother then other in crossfire. :/
anyone know where I can change the driver to do super tiling in stead of AF rendering ?

seem like it dose super tiling in windowed and AF in full screen.

http://www.xtremesystems.org/forums/...d.php?t=252301
That will do it, I think.

[Zaxis01]

Heaven Benchmark Using 2x 5870's

[Arseface]

Thanks zaxis

Looks like the 5870 x-fire setup is just as bad as SLI GTX480s when it comes to microstutter though

[Asmola]

I'm running 2x 5870's in cf and ran AVP for about 30-40 secs and these are my results.

Dont know what is my problem but im getting much worse microstuttering on AVP benchmark, might be different while playing?

CPU stock:


CPU overclocked:


Benchmark it self doesnt look so bad, but those numbers are really bad..

[Arseface]

If you're running at settings which are largely GPU-limited, then you will see the microstutter (as you have found above).

In CPU-limited scenes you will see very low microstutter, as the GPU output syncs to the CPU, which it is waiting for after each frame.

Given that you have gained (virtually) nothing in this benchmark from overclocking the CPU, I would guess you're largely GPU-limited in this case. I suspect the scene that Zaxis benchmarked was largely CPU limited.


... Looks like I don't need the entire game in order to run the benchmark, so I'll grab it and do some comparisons using the GTX480s. Good to see more ATI results coming in now though - just what I'm after in order to compare the two technologies It would help if you could let me know the exact settings you used (resolution, AA/AF, detail levels etc), and when you started the FRAPS benchmark / how long you ran it for, so I can replicate as close as possible



Edit:

1920*1200, 4xAA, 16xAF, textures=very_high, SSAO=on, tessellation=on, shadow quality = high.

Ran the FRAPS bench for the entire run (stopped as the benchmark fades out)







Ouch... Pretty bad Although, as above, the sufficiently high framerate made it seem pretty smooth.

[Carfax]

Thanks zaxis

Looks like the 5870 x-fire setup is just as bad as SLI GTX480s when it comes to microstutter though

Did you see my earlier scores?



Thats for the heaven benchmark, @ 2560x1600 4xAA 16xAF, extreme tessellation etc..

A value of 18% isn't too bad is it?

[janolle]

A value of 18% isn't too bad is it?

I would say that 18 is easily noticed if you pay attention to what fps readings you're getting and how it looks. What you see is an average performance drop of 18% compared to the fps reading you're looking at. And that's an average reading wich it is probably much worse in some situations.

I once saw a nifty little program that you can test microstuttering with different variables. You should try it out if you can find. I think it's made by some german guy.