HD5970 Microstuttering tests

[Manicdan]

if everyone who said they saw MS on their PC gave me 5$, then i might have enough money to quit my job, buy a bunch of new and old hardware, and spend a few months and figure it all out. or i could just retire to a small island with that money.

[annihilat0r]

Here's my result from a single 5870@stock at 1920x1200 with everything on Enthusiast.

[Manicdan]

now that is really bad, can you do the same test but drop the textures down one notch (what game btw?)

[annihilat0r]

Oh, forgot that It's Crysis Warhead on the Ambush timedemo. Yeah, it's horrible. I recorded frametimes for approx. 1000 frames, and 500 frames were like this. The rest was quite nice.

[jmke]

what was your framerate? if it's low, I wouldn't be surprised that you'll notice stuttering

[annihilat0r]

You can calculate from there, I guess it was around 40 FPS. I can't say I have noticed stuttering, but it was only a timedemo and I wasn't controlling the character.

[Teemax]

if everyone who said they saw MS on their PC gave me 5$, then i might have enough money to quit my job, buy a bunch of new and old hardware, and spend a few months and figure it all out. or i could just retire to a small island with that money.

I guess with the $100 or so that you'd probably collect, you could buy a small piece of land in Haiti these days.

[Solus Corvus]

Here's my result from a single 5870@stock at 1920x1200 with everything on Enthusiast.

http://i50.tinypic.com/52b1gj.png

Could you re-run with render ahead set to 0 frames?

I guess with the $100 or so that you'd probably collect, you could buy a small piece of land in Haiti these days.

Not funny.

[Eastcoasthandle]

Could you re-run with render ahead set to 0 frames?


Not funny.

This is something I would like to see from all CF solutions provided in this thread if possible.

[Solus Corvus]

With CF/SLI you have to render at least 1 frame ahead, IIRC.

[eRazorzEDGE]

How do you do that?

[cegras]

Doesn't this just prove that 'microstuttering' (oh god, I hate that word) simply occurs when the instantaneous load on the GPU is too much for it to handle?

[eRazorzEDGE]

These graphs don't really prove or disprove anything, except that the same thing happens on single GPU's the same as multiple GPU's.

What we need is someone who can...
A.) actually see the MS
B.) reproduce said MS
C.) replay same scenario without MS
D.) record both frame interval data and shoot hi def, minimum 60 frames per second, video and then see if we can't find a link somewhere.

[Xoulz]

These graphs don't really prove or disprove anything, except that the same thing happens on single GPU's the same as multiple GPU's.

What we need is someone who can...
A.) actually see the MS
B.) reproduce said MS
C.) replay same scenario without MS
D.) record both frame interval data and shoot hi def, minimum 60 frames per second, video and then see if we can't find a link somewhere.

Why do single GPU's use AFR..?

[Solus Corvus]

How do you do that?

ATI Tray Tools used to have an option for it. I'm not running an ATI card at the moment so I can't verify it still does.

Why do single GPU's use AFR..?

They don't. But they do render ahead frames unless told not to.

[rge]

None of the graphs we have seen in this thread are microstuttering, it is just swings in framerates seen with 1 or 2 gpus.

People that were complaining about microstuttering in Crysis were seeing movement, pause, movement, pause, it was obvious to all that viewed it, similar to watching 3dmark6 and watching the cpu bench run at 1-4 fps. The corresponding graphs had a rhythmic dropping in framerates from 30-40fps down to 10fps in certain segments of the game that corresponded to start, pause, start, pause (again that was obvious and all could see it). Then turning SLI off, the rhythmic decrease in framerates improved and the graphs showed less swings, though still have some microstutter but was much better. See the link, it has graphs showing real microstutter.
http://www.overclockers.com/micro-st...and-crossfire/

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.

I played large segments of codmw in last day with fraps recording entire time with 1gpu and then sli, and no microstutter with 1 or 2 cards visually or graphically. In fact the variation was identical even with graphing 32,000 frames at a time (max in excel btw) only sli showed this same variation at a higher framerate.

I dont think ms exists on any of the 7 recent games I played, and my guess now that people are looking for it, so are the developers, and like any other known bug, it is now very rare to see.

But there will always be those that are now hypertuned to looking for any glitch and say, "see that jitter, right there, didnt you see that ms". um, no. "Well your not looking hard enough...see right there, that jitter."

And by the way I ran with 0 prerendered frames, 1,2 and default 3 with sli and 1 gpu. The default 3 had the best graphically, the smallest swings in codMW, though that may vary game to game. And 0 prerendered is bad, kills lot performance with sli. 1,2 simply had larger swings than default 3.

[BenchZowner]

If you want to see microstuttering all you have to do is open Crysis, and walk sideways left & right ( strafe ).

[eRazorzEDGE]

Ok, been playing Crysis for a little while today. Any stutter is from the hard drive being accessed. I know this, because when I see it start to stutter, I can see the HDD light solidly lit. Maybe I'll get FRAPS to take a video for you.

[BenchZowner]

Ok, been playing Crysis for a little while today. Any stutter is from the hard drive being accessed. I know this, because when I see it start to stutter, I can see the HDD light solidly lit. Maybe I'll get FRAPS to take a video for you.

In a scene that has been pre-rendered ?
Not happening, unless you think Crytek's engine is so dumb and cr*p to dump data that is in use

It happens in any SLI/CF system here no matter what.
But it doesn't happen on a single card system.
If you still think it's the HDD, then set texture streaming to off in order to use texture precaching and check again.
For what it's worth, there are a dozen system executables that can be accessing the disk ( check the file i/o per executable @ task manager if you doubt it )

[HWaddict]

All this talk about Microstuttering sounds like jitter in the Audiophile world (witch I left Years ago).
...

[eRazorzEDGE]

In a scene that has been pre-rendered ?
Not happening, unless you think Crytek's engine is so dumb and cr*p to dump data that is in use

It happens in any SLI/CF system here no matter what.
But it doesn't happen on a single card system.
If you still think it's the HDD, then set texture streaming to off in order to use texture precaching and check again.
For what it's worth, there are a dozen system executables that can be accessing the disk ( check the file i/o per executable @ task manager if you doubt it )

No, it doesn't happen in any SLI/CF system here no matter what. You can say that it's happening until your ass bleeds, it doesn't make it true.

Yes, I realize there a dozen things that are running in the background and can access the disk. And I also realize that when the HDD is being accessed, it slows frame rendering. I'm not sure why it does it and I really don't want to postulate a theory right now, but go ahead and turn on a disk defragmenter while your playing a game and tell me your FPS stays at it's maximum.

Also, here is a video I uploaded to YouTube... http://www.youtube.com/watch?v=ETn7PNvclxY

It's about 1 minute long. The original was around 650MB and was smooth as butter, but every conversion process that I've tried on the file makes it come out with way worse that original quality and gives the appearance of slight stuttering, when in fact, there was none. So I uploaded the raw file to YouTube, but they converted it so w/o sending everyone the direct file, you can't see what I see.

[rge]

I tried straffing side to side in multiple places in Crysis, and looking at scenery in distance there is a little jerking noted not evident when walking straight, but cant tell if that is because the character bounces up and down when straffing or not. But it visually looked exactly same with sli vs 1 gpu, and there was no pausing or microstutter visually noted. I graphed straffing with sli, and walking straight with sli. I then graphed straffing and walking straight with 1gpu. All looks same to me graphically and visually, except FPS were less with 1 gpu.

[eRazorzEDGE]

I just tried strafing while looking away from the sun, towards the harbor, with palm trees directly in front of me (3 at roughly 5 to 10 foot distances away), and I could see some lag on my display... more like screen tearing. I could see the edges of the palm trees clearly being drawn in a new spot on the screen, from top to bottom. I was getting around 55-58 FPS at this time with two 5870's in Crossfire. I tried enabling Vsync and the "tearing" went away even thought I was below 60 FPS.

I then went on to try it with my second 5870 disabled and while only getting 30-32 FPS, I could still see the "tearing". So then I tried Vsync again, and the problem went away. I recorded the frame times with FRAPS for both Single and Dual GPU's, with and without Vsync and the graphs look very similar to RGE's above... constant ~5ms fluctuation between frames.

I also tried to capture video with FRAPS, but the same image tearing doesn't show up on the video and alas, I don't have a digital camcorder.

I also tried the r_texturesstreaming=0 and e_precachelevel=1 to no avail. It still had the screen tearing, but there's no hint of "microstutter". However, it will randomly stutter on occasion, but I can't say this is "microstutter" as most games do this when something is reading/writing from/to the hard drive.

[Vinas]

Well, I could never notice a difference back when I had a 3870x2. Today I still can't notice any difference with the so called microstutter. I only believe it exists because people claim they can see it here.

To me the quality between my 5870 is about the same but with more framerates than my old 3870x2... ?

[BenchZowner]

Crysis strafing or not runs as smooth as butter on my rigs with GTX 285's and 5870s as long as it's one gpu, Vsync disabled or enabled, doesn't matter, no tearing, no ms, nothing.And trust me, I'm very picky and my eyesight is well... very very good.
As soon as I add another VGA on the rig/s I get ms in several parts of the game.

One way or another, that's my experience and what I can see using all the cards I have here, it's my own experience and realization of what the monitors are displaying, I'm not going to try to convince or force any of you to stand by me and my opinion