HD5970 Microstuttering tests

[eRazorzEDGE]

for everyone who wants a direct comparison of single gpu and multi gpu graphs of frame times pcgh has a few articles that show a huge difference between these 2 setups.
e.g.:
- http://www.pcgameshardware.de/aid,63...fikkarte/Test/
- http://www.pcgameshardware.de/aid,65...fikkarte/Test/

And starting back at post 94, I put together a little single 5870 vs dual 5870's frame time graph comparison. It shows that the single vs. dual GPU frame rendering times are damn near identical at the settings I ran. Also, others have posted their own personal comparisons.

Now, granted the links you posted I believe were with somewhat older cards, excluding the 4870x2. But as I've said before, we need people who can quantify this graphical behavior instead of people posting these graphs that may or may not be proof of anything.

[rge]

the german website pcgameshardware.de put a lot of effort in proving microstuttering and trying to show it with videos. \

Thanks for links, especially the video of ms. And that is what I thought ms was, basically a variance where the frame rate dips down from say 40 fps alternating with 20fps, ie below 30fps where the brain can see a smooth image. And the graphs show it nicely.

And I agree, something has improved since that 2008 shots, since now the single gpu graphs look just like the multi gpu graphs on the 10+benches I tried on varying games and posted several in this thread after eRazorzEDGE illustrated the controls (single gpu) looked same now. In fact sli had less variation frame to frame then single gpu in 3dmark6, maybe that has just been well optimized for sli. Crysis is only game I have seen where single gpu variance is slightly less than sli, others were exactly same amount.

And yes most could easily tell difference between 30 and 60 fps if viewed side by side, it is just when you are varying framerates from well above 30 to well below 30 microstutter will be evident to everyone. Air force pilot study showed they could identify a plane with viewing 1/220 second frame, correspond to 220 fps, though the identification depended on afterimage.

But if varying framerates from 70 to 45 fps also is suttle microstutter then my single gpu microstutters the same as sli, since both graphs look same. Though I cant see any microstutter, I will admit if saw a smooth fps of 70 vs alternating 70 to 50, I might be able to see a difference. And no doubt there will be some games where sli is worse, hard to believe that would not be case, but there is not any visible ms in any of 7 games I recently played that I can see, nor any difference graphically.

[Dresdenboy]

Is vsync on in these tests? I ask, because it looks like a timing problem: slight variations in actual calculation time for a frame could jump around the monitor's frame time border, alternatingly being in time and a bit too late.

[prava]

First and most importantly:



A huge clap for eRazorzEDGE for his factual job.


First of all, people needs to stop talking about FPS when talking about MS. Guys, when dealing with MS, miliseconds and individual frames is the only things that matters as I couldn't care less about how many frames do I have at the end of the second but how they are distributed among that second: an even distribution will never have MS no matter how low the framerate is. Of course, the higher the framerate the lower the chance of having MS but they are not totally related, you can have 80fps and have MS yet have 30fps and not have MS at all. It's totally and absolutely logical If you understand the concept on itself.

I'm talking about comments like this:

The people who don't notice microstuttering are playing games in which the minimum momentary FPS doesn't go below 50 or 60 something. When your momentary FPS is 50-100-50-100-50-100 you don't see a 75 FPS level fluidity, you see 50 FPS level fluidity. But when your FPS's are so high that you don't see microstuttering, then your second GPU doesn't add anything, does it? All you have done is to pay huge amounts of extra money just to see that 75 FPS indicator whereas you are getting a mere 50 FPS.

To me, microstutter is when you get large swings in framerates that constantly dip enough below the 30fps threshold such that the normal viewer would see start, pause, start, pause when playing certain segments of the game. And if caused by SLI, the problem gets better, visually and graphically, with turning off sli. And microstutter is apparently fixed with updated drivers, settings or code, just like any other bug.

as long as ur fps stay above ur monitors refresh rate u cannot experience the microstutter phenomenon.
games that run with fps rates under the monitor refresh rate are stuttering, no matter if ur on a multi-gpu or a single-gpu system.
microstuttering only gets extreme if ur reaching very low fps rates around 30fps, then u can actually notice it. but who the hell plays ego shooters on 30fps??
to me the whole microstutter stuff is just theoretical nonsense. my monitor displays 60 frames per second. so i don´t care if one frame is rendered 0.002s or 0.004s after another, as long as there is always a different frame to display 60 times in a second.
microstutter has nothing to do with a drop in fps. ur fps stays the same, the only thing that happens is, that frames aren´t put out "almost" in sync like they would on a single gpu solution. and this is only noticeable in really low fps environments, when u actually wouldn´t play the game, even on a single gpu system.

Why can we have MS with 80fps? Its darn simple: if we have an even distribution there will be always a new frame new the screen is ready to update. But, if we have spikes, something like (measured in miliseconds): 8-9-8-9-40-8-9-8... even if, at the end of the second, we get the total of 80frames, you have stuttering. Heck, we could even have 1-1-1-1-1-1-100-1-1-1-1-100-1-1-1-1-100 and at the end of the second we would be deffinitely above 100frames yet it would be unbearable. That's why we can't talk about FPS in here as it's not accurate at all.

Has anyone noticed screen tearing below 60fps, while playing Crysis and then noticed it go away when turning Vsync on?

BenchZowner mentioned the strafing technique, so I tried it while looking out at the harbor in the level Contact, with the sun behind me and a couple of palm trees directly in front of me at a distance of roughly 5 to 10 feet away. While getting a constant FPS of around 57 FPS and Vsync off, I noticed screen tearing of the edges of the palm trees, but with Vsync on and the same 57 FPS, it went away. It was the same when I used only one GPU.

Anyone have an explanation or can verify what I'm seeing on my end?

Again, it's pretty simple. Uneven distribution of frames means that the screen will be given two frames while drawing, which means tearing. Vsync means that the gpu's only output frames exactly when the screen is up to refresh, nothing else.

and now don´t come and tell me that u can see more than 60fps on a 60Hz lcd display.

You are actually wrong. Displays can show more than 60fps otherwise we would not have the tearing problems, in the sense that if the screen is drawing and you get it a new frame it will update its buffer and keep drawing with the new information so yes, we can display more than 60fps on a 60hz display.

[orangekiwii]

You can also see that it's smoother.

Its easy to tell the difference between 200 fps and 60 fps in games. Even on my 60hz lcd it's easy. The 200 fps is just more fluid.

[KoHaN69]

This is an old debate, and I'll say same thing I said last time regarding:

Getting a video of it is much more difficult than understanding it technically.
Why ?
Any screen capturing utility slows down the system's performance significantly.
A handheld camera can't record high FPS as they're outputted by the VGA.

YES! The best way to do this is record this with a high-speed camera.

[rge]

First of all, people needs to stop talking about FPS when talking about MS. Guys, when dealing with MS, miliseconds and individual frames is the only things that matters as I couldn't care less about how many frames do I have at the end of the second but how they are distributed among that second: an even distribution will never have MS no matter how low the framerate is.

Why can we have MS with 80fps? Its darn simple: if we have an even distribution there will be always a new frame new the screen is ready to update. But, if we have spikes, something like (measured in miliseconds): 8-9-8-9-40-8-9-8... even if, at the end of the second, we get the total of 80frames, you have stuttering. Heck, we could even have 1-1-1-1-1-1-100-1-1-1-1-100-1-1-1-1-100 and at the end of the second we would be deffinitely above 100frames yet it would be unbearable. That's why we can't talk about FPS in here as it's not accurate at all.

So basically what you are saying is if something is traveling at 500 miles per hour, but has yet to go one mile it is not really traveling at 500 miles per hour, and hence we should not talk about it in mph terms...think about it for a minute, you might get the analogy.

1 second = 1000 microseconds. 30 fps = 1 frame every 33 microseconds.

THERE is no difference in stating fps versus microseconds of frames, unless you actually think until you have reached one second you cant know the speed of that frame in fps.

For example let me translate your numbers, 8 8 8 40 8 would be 125 fps, 125 fps, 125 fps, 25fps, 125 fps.

You are confusing fps with some bizarre notion that unless you have waited full second you cant measure transient frames per second. For example you can measure over .25 seconds then multiply x 4.

So if I get pulled over for a ticket for driving 45 miles per hour in a 35, can I tell the police officer I havent driven a full mile yet, hence I was going to drive 10 miles per hour under 35 for next 1/2 mile, therefore I was doing exactly 35mph, since you dont know my speed until I have driven a full mile, that is in terms of mph.

Or was this a joke?

[RaZz!]

So basically what you are saying is if something is traveling at 500 miles per hour, but has yet to go one mile it is not really traveling at 500 miles per hour, and hence we should not talk about it in mph terms...think about it for a minute, you might get the analogy.

1 second = 1000 microseconds. 30 fps = 1 frame every 33 microseconds.

THERE is no difference in stating fps versus microseconds of frames, unless you actually think until you have reached one second you cant know the speed of that frame in fps.

For example let me translate your numbers, 8 8 8 40 8 would be 125 fps, 125 fps, 125 fps, 25fps, 125 fps.

You are confusing fps with some bizarre notion that unless you have waited full second you cant measure transient frames per second. For example you can measure over .25 seconds then multiply x 4.

So if I get pulled over for a ticket for driveing 45 miles per hour in a 35, can I tell the police officer I havent driven a full mile yet, hence I was going to drive 10 miles under 35 for next 1/2 mile, therefore I was doing exactly 35?

Or was this a joke?

ehm, i don't get what you're trying to say with this post. in my opinion you basically said exactly the same thing as prava. the only thing i don't understand are the numbers for the said example: 125fps, 125fps, 125fps, 25fps, 125fps (the bolded part in the quote). this is wrong.

to sum it up:
fps stands for frames per second, but we're in in a dimension between fps.
the example 8-8-8-40-8 stands for the time in ms between each rendered frame. in other words: 1st frame is displayed after 8ms, 2nd after 8ms, 3rd after 8ms, 4th after 40ms, 5th after 8ms. in total this means 5 frames per 72ms. we're still far away from frames per second at this point. but what's important about microstuttering is, that the total amount of frames isn't the crucial measurement, instead the time between each frame is important.

or did i miss smth about this little discussion?

btw: sry for my ninja editing. i always tend to hit the "post reply" button too quick

[STEvil]

So basically what you are saying is if something is traveling at 500 miles per hour, but has yet to go one mile it is not really traveling at 500 miles per hour, and hence we should not talk about it in mph terms...think about it for a minute, you might get the analogy.

1 second = 1000 microseconds. 30 fps = 1 frame every 33 microseconds.

THERE is no difference in stating fps versus microseconds of frames, unless you actually think until you have reached one second you cant know the speed of that frame in fps.

For example let me translate your numbers, 8 8 8 40 8 would be 125 fps, 125 fps, 125 fps, 25fps, 125 fps.

You are confusing fps with some bizarre notion that unless you have waited full second you cant measure transient frames per second. For example you can measure over .25 seconds then multiply x 4.

So if I get pulled over for a ticket for driving 45 miles per hour in a 35, can I tell the police officer I havent driven a full mile yet, hence I was going to drive 10 miles per hour under 35 for next 1/2 mile, therefore I was doing exactly 35mph, since you dont know my speed until I have driven a full mile, that is in terms of mph.

Or was this a joke?

Frames vs Microseconds gives you a more accurate measurement allowing you to see if you are experiencing poor frametimes. If you only measure frames over a period of a second you will only see the number related to a bundle of frames vs a number than can be related to a single frame.


I want you to attempt to drive your car down a road at speeds between 10 and 60mph using maximum braking and acceleration maintaining an average speed of 40mph then come back and tell me it was smooth.


edit - while you're at it, lets see some graphs using frames per 10 seconds or 60 or 100, rge...

[annihilat0r]

blabla

Are you joking? You are basically saying the same thing as I was saying, yet in a completely wrong manner (1-1-1-1-1-100 won't be above 100 frames) and you're quoting my words and opposing what I have said.

[prava]

So basically what you are saying is if something is traveling at 500 miles per hour, but has yet to go one mile it is not really traveling at 500 miles per hour, and hence we should not talk about it in mph terms...think about it for a minute, you might get the analogy.

1 second = 1000 microseconds. 30 fps = 1 frame every 33 microseconds.

THERE is no difference in stating fps versus microseconds of frames, unless you actually think until you have reached one second you cant know the speed of that frame in fps.

For example let me translate your numbers, 8 8 8 40 8 would be 125 fps, 125 fps, 125 fps, 25fps, 125 fps.

You are confusing fps with some bizarre notion that unless you have waited full second you cant measure transient frames per second. For example you can measure over .25 seconds then multiply x 4.

So if I get pulled over for a ticket for driving 45 miles per hour in a 35, can I tell the police officer I havent driven a full mile yet, hence I was going to drive 10 miles per hour under 35 for next 1/2 mile, therefore I was doing exactly 35mph, since you dont know my speed until I have driven a full mile, that is in terms of mph.

Or was this a joke?

You are wrong. In this case we don't care about FPS, we care about the distribution of the frames among time. A second is too big to be used as a unit to quantify anything as the whole title says everything: microsuttering, or micro-pauses. Besides, the FPS thing is not a speed quantifier (*in most of the cases, as having more fps in a game will not make it render faster) nor anything related so no, if we have 0.30sec passed we don't know how the FPS will be as this unit measures all the frames drawed within a second and a computer is not a car in the sense that you can always predict the speed at the end of it. Think about it, how can you talk about 125-125-125-125-50-125-125 when all of those numbers relate to onlye ONE frame? That is nonsense. 200fps is not smoother than 100fps because there are more frames, but because they will be more even distributed among time. That's why the total number is useless.

Are you joking? You are basically saying the same thing as I was saying, yet in a completely wrong manner (1-1-1-1-1-100 won't be above 100 frames) and you're quoting my words and opposing what I have said.

That was an example. 1-1-1-1-1-1-100-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-100 will be above 100 FPS at the end of the second if we would follow the pattern; and I quoted your words because you also use the FPS stuff in the wrong way, such as rge does.

[rge]

snip

snip

snip

You are all assuming by fps it can only mean the number of frames measured in 1 second. While that is one way of making the calculation, that requires an assumption which in this case is incorrect.

You can also measure fps by recording 1 frame/time it takes. for example 1 frame rendered in 33ms is speed of 30 fps. THAT DOES NOT MEAN I AM TALKING ABOUT ALL THE FRAMES IN 1 SECOND. and my god this is such a simple concept.

30 fps does not necessarily = average frames per second nor does it HAVE to mean the number of frames in one second, because speeds can vary frame to frame. It does not mean I measured all the frames in 1 second, anymore than an instant speed of 50 mph means I measured all the miles in an hour, it can refer to a fraction of an hour, ie instant speed.

fps can denote an instantaneous speed, just like mph, or any other speed

Again, the car is traveling 50 mph. Does that mean I measured 50 miles after 1 hour, well it could or it could mean an instantaneous speed reading.

Given the term can clearly be used either way, and given we have graphs or ms per frame, it is clear to me we are talking instantaneous speeds. Just like if a police officer tells you 50 mph you can assume instantaneous speeds. Or if someone says you need to drive 50 mph in order to reach your destination, one can assume averages.

And Razz you are right, I am using same terms as Prava, 1 frame rendered in 8 microseconds is EXACT same thing as an instantaneous speed of 125fps, 1000/8=125. It is Prava that assumes fps can only mean number of frames averaged over 1 second, which is wrong. If it were true, then you can not get a speeding ticket by going 50 mph, until you have driven an hour.

And if you dont like me using fps as instantaneous speeds, or simply dont understand simple math conversion, ignore me, because I am going to continue to do so, as it is clearly correct.

[annihilat0r]

balbla

What you fail to understand is that you don't need a whole SECOND to measure FPS. In my post I was talking about "momentary" FPS which would be equal to 1000 divided by the time it took to draw the next frame, which is what you are talking about.

[rge]

What you fail to understand is that you don't need a whole SECOND to measure FPS. In my post I was talking about "momentary" FPS which would be equal to 1000 divided by the time it took to draw the next frame, which is what you are talking about.

yeah it was very obvious that we were all talking about momentary FPS, especially given the graphs, but the johnny come latelys that dont even bother to read then post...not that I have ever done that Arguing semantics +/- math and reading comprehension issues.

But I will try this the next time I get pulled over.

Officer: "you were going 55mph in a 45 mph zone, you were speeding."

Me: "no sir, 55 miles per hour, means you have to measure me for an hour, otherwise you have to call my speed something else. For example, take frames per second. If one frame is rendered in 33 microseconds, while technically that is a speed of 30 frames per second, you cant call it that until you measure for a whole second, and if you dont believe me, I will show you the post on the net."

Officer: "are you on drugs?"

me: "no sir, you are not getting it. You see miles per HOUR. You did not watch me the whole HOUR, therefore technically it is not correct to state my speed in miles per HOUR, you have to use some other terminology. Because when you say miles per HOUR, it ONLY means you watched me go for 1 hour, then calculated my speed, it can not refer to my momentary or instantaneous speed."

And yes, everyone who has posted in this thread understands the simple concept that AVERAGE frames over 1 second is not indicative of the milliseconds it takes to render one frame, ie the instantaneous speed of frames per second.

And anyone who bothered to actually read this thread before posting would realize we are talking about fps as momentary fps.

BTW, can one of those semantics posters please answer this question.

A frame is rendered in 33.3 microseconds. At what speed was that frame rendered?

[Behemot]

A frame is rendered in 33.3 microseconds. At what speed was that frame rendered?

1 frame/33.3us or 0.03 frame/us regulated to one

[annihilat0r]

Officer: "you were going 55mph in a 45 mph zone, you were speeding."

Me: "no sir, 55 miles per hour, means you have to measure me for an hour, otherwise you have to call my speed something else.

[ottoyu34]

Nicely done. This confirms what I see with the game Dirt 2.

[eRazorzEDGE]

Again, it's pretty simple. Uneven distribution of frames means that the screen will be given two frames while drawing, which means tearing. Vsync means that the gpu's only output frames exactly when the screen is up to refresh, nothing else.

But it was under 60 frames per second. On the single GPU it was only 30 FPS. It happens consistently and is very reproducible.

I'm just not really getting exactly how it does it. It's my understanding that the display refreshes 60 times a second w/o any "output" or any sort of timing signal. I could be wrong on that however. If I'm right, how does just switching on Vsync, and getting below the monitors refresh rate in FPS, make the screen tearing go away?

That would mean that the system or GPU needs to know the exact time to render a frame so that monitor doesn't refresh while the single contains half of one frame and half of the next.

I guess I'll just have to buy a high speed camera and do some more tests.

[One_Hertz]

I really wish review sites started posting frame rendering time graphs instead of the semi useless fps figures. I bet there will be differences in stability between different cards..

[rge]

I really wish review sites started posting frame rendering time graphs instead of the semi useless fps figures. I bet there will be differences in stability between different cards..

Yep, graphs are the only way to go. 60 fps average, or 16.6 millisecond avg render time per frame doesn't tell the same story as a graph of render time vs frame number or render speed vs frame number, whichever way you prefer to express it.

[STEvil]

You are all assuming by fps it can only mean the number of frames measured in 1 second.

That is because that is the defintion of per second.

You are correct that 120fps could be 1 frame per 33ms but the measurements we are interested in are milliseconds per frame, not frames per millisecond, second, minute, hour, etc.

[rge]

That is because that is the defintion of per second.

Frames per second do not imply one second has passed that is an error. Just like miles per hour do not imply one hour has passed, that is also an error. They are simple units of measurement in speed, and absolutely imply nothing else. If one had to wait a second or an hour to make measurements, math would be bizarre.

If you were given a simple math problem that stated, one frame is rendered in 33 microseconds, please state the speed of that frame using units of frames per second, what would be your answer? Sorry cant, have to wait and count for a second?

Or if you were given another math problem, object traveled 1 mile in 15 minutes, what is the speed in miles per hour?
Is your answer going to be, sorry but miles per hour, means an hour, so no answer until it travels an hour.

fps does not imply one second has passed, anymore than miles per hour imply an hour has passed. I honestly cant believe someone is actually even saying that.

You are correct that 120fps could be 1 frame per 33ms but the measurements we are interested in are milliseconds per frame, not frames per millisecond, second, minute, hour, etc.

So you are saying one frame in 33ms is different from 33ms to render 1 frame? Or 10 minutes to drive 10 miles, is different than 10 miles driven in 10 minutes?

I think the issue is when review sites use fps they are taking an average of several minutes of data, (or even if took 1 second average), they gave the term fps a bad name, since the average fps is not representative of what is happening frame per frame. But so what if use milliseconds per frame (same thing to me just different units, and I will use your preference of stating it since it matters to you) they would simply give you an average milliseconds per frame, unless you expect them to tell you every frame they measure in words? Well they could simply start telling you the speed of every frame rendered as well, but either would be difficult to present in text.

Again, like One Hertz pointed out, changing the units isnt the solution, you will still get averages, the solution is to change the format of presentation. Showing graphs or various segments in game that show speed of each frame (or time) is the solution, units used I could care less, my graph above is a mirror image of each other.

[eRazorzEDGE]

Showing how long it takes for each frame to render is ridiculous, unless we can prove that it actually means something.

There's a reason why the "review sites" list FPS (max, min, average), it's be cause it's a whole hell of a lot easier to understand than these graphs we're all producing. We don't even have any factual information on why, even on single GPU setups, each frame takes a different amount of time to render... even if the alternating high's and low's stay at their same respective values over a period of time.

Sure we can add this type of testing method to the traditional FPS tests, but these will never outweigh the traditional FPS tests. For right now, people should remain objective and try not to draw any conclusions from these graphs.

On the subject of FPS understanding... it's simple. You take the number of units measured in a given time frame and VOILA!

Here's a lil example for the hard of reading:

20 frames rendered in 0.50 seconds = 40 FPS

or

1000 frames rendered in 25 seconds = 40 FPS

Take 5th grade math, the whole adding/subtracting fractions thing really works well here.

[STEvil]

Showing graphs or various segments in game that show speed of each frame (or time) is the solution,

Milliseconds per frame, as the graphs have been showing is the correct measurement.

Measuring frames per second is an average measurement over a period, not an instantaneous one. This is pretty basic math.

[Dimitriman]

Seconds:
0.1 - 0.2 - 0.3 - 0.4 - 0.5 - 0.6 - 0.7 - 0.8 - 0.9 - 1.0
Frame-Frame-Frame-Frame-Frame-Frame-Frame-Frame-Frame-Frame

= 10 Frames per second - No Micro Stutter

Seconds:
0.05 - 0.4 - 0.45 - 0.5 - 0.8 - 0.85 - 0.88 - 0.92 - 0.96 - 1.0
Frame-Frame-Frame-Frame-Frame-Frame-Frame-Frame-Frame-Frame

= 10 Frames per second - Micro Stutter


Isn't that how it works? I don't see whats so hard to understand