HD5970 Microstuttering tests

[kadozer]

Very well done. So the question is, is the person who did the article stupid or do they have an agenda?

Just a bad article. No baseline should be the first clue. Not everyone has the same exact setup that they ran.

[jaredpace]

Microstutter in action:

- Time: 1:12-1:16 on 5970 screen watch carefully as he rounds the corner around the pole, then rewind and watch how much smoother it looks on the 5870 (even though 5870 has less fps).

- Time: 1:21-1:26 on 5970 screen watch carefully at the Camouflage netting on the top of the FOV, and again watch the screen underneath it and see how smoother it is on the 5870.

http://www.youtube.com/watch?v=emG7ZNIsxw8&fmt=22

Single GPU rendering is less likely to show this type of stuttering compared to multi-GPU.

[eRazorzEDGE]

I edited my previous post to include the Supertiling technique of rendering on Crossfire. It appears that Supertiling, in the graph, varies less from frame to frame, but when viewing the actual run, it looks almost as if there is more screen tearing. So I don't know.

Perhaps some other would care to test this out?

Microstutter in action:

- Time: 1:12-1:16 on 5970 screen watch carefully as he rounds the corner around the pole, then rewind and watch how much smoother it looks on the 5870 (even though 5870 has less fps).

- Time: 1:21-1:26 on 5970 screen watch carefully at the Camouflage netting on the top of the FOV, and again watch the screen underneath it and see how smoother it is on the 5870.

http://www.youtube.com/watch?v=emG7ZNIsxw8&fmt=22

Single GPU rendering is less likely to show this type of stuttering compared to multi-GPU.

If you look closely, it almost appears as though it's a checker board type effect on the 5970... which is what Supertiling is. If I had a video camera, I would show the difference between the different crossfire techniques.

[Manicdan]

im about to start a "Give eRazorxEDGE a Poster-of-the-Year award" poll, and i expect to see everyone in here vote yes

thanks for the work, simply proving that its not always there, and gives confidence to people who have it and want to get rid of it. i will take a alot of tests with alot of different hardware and games and hundreds of graphs to see if there is any kind of obvious cause to microstuttering, but no one should have to go through all that pain.

[Vørtex]

Just to backup what eRazorzEDGE has been saying about Crossfire being based on Supertiling since X8xxx.

"Supertiling is the default rendering mode for Crossfire in all D3D applications that CATALYST A.I., the ATI mechanism for game profiling in their driver and one without any user modifiable settings or game attributes to affect Crossfire rendering mode, has no idea about. If ATI have performance profiled an application and deemed it unsuitable for Supertiling or more suited in a performance sense to Scissor or AFR, CATALYST A.I. is the mechanism to enable that. Therefore if a Crossfire user is to disable CATALYST A.I. completely, Supertiling will be used for everything across the board, regardless of whether its the best performance mode as decided by ATI or not."

Source: http://www.hexus.net/content/item.php?item=1651&page=9

[Levish]

Very well done. So the question is, is the person who did the article stupid or do they have an agenda?

I'm going to say that the problem likely isn't entirely attributable to the card, lots of things contribute to heavy variability of FPS but the primary culprit (imo) is RAM and how the App handles RAM and loading of textures.

The poster that mentioned LoTRO, I'm going to say to try fully disabling superfetch/prefetch might help his stuttering problem on Vista.

[eRazorzEDGE]

Thank you guys, it's much appreciated. And welcome Vortex, excellent first post

Also, about the stuttering itself, I think it's more apparent do to the fact that 2 (or more) cards are throwing out more frames per second so in a sense, you're getting really bad screen tearing, except that on an ATI card it looks more like a checkerboard-type of tearing. And since people can really only focus on one part of a screen at a time, they don't get the whole picture, so-to-speak. It just appears like a stutter, but it's really screen tearing only at certain parts of the screen.

On a related note... I just recorded a video of the Crysis benchmark, but it's only running around 24 FPS when I start recording, which is really low considering my hardware and previous benches. Also, when it's running at that low FPS, the images don't move from one part of the screen to the other as fast as they do when I'm getting say 60 FPS, so it's an illusion of no stuttering when I can clearly see the "stuttering/tearing" at higher FPS with Supertiling on. Does anyone know if this can be increased or do you just have to get a camera to properly record?

[naokaji]

i will take a alot of tests with alot of different hardware and games and hundreds of graphs to see if there is any kind of obvious cause to microstuttering, but no one should have to go through all that pain.

While I agree that no enduser should have to go through all that testing, I believe that Nvidia and ATI should bother with it because after all they are the ones who want our hard earned money for their products.

@eRazorzEDGE
Impressive what you did there, good job

[Manicdan]

While I agree that no enduser should have to go through all that testing, I believe that Nvidia and ATI should bother with it because after all they are the ones who want our hard earned money for their products.

i fully agree with that statement, but i wonder, has either company even acknowledged that microstuttering exists? i think thats the first and easiest step, but somehow thats the hardest step for the company to make.

[Levish]

i fully agree with that statement, but i wonder, has either company even acknowledged that microstuttering exists? i think thats the first and easiest step, but somehow thats the hardest step for the company to make.

There are just so many variables I would question if everyone is experiencing microstuttering (or to the same extent).

Valid testing to come to the conclusion that its the vga would be:

Tested across OS's
OS versions (32/64bit)
driver versions
superfetch off and on (i would personally leave it fully disabled for testing)
indexing off and on
Different hard disks tested
video settings and so on.
appropriate peformance logs ran (disk queue length, available physical memory, % processor etc)

[Manicdan]

yup, so many things to consider, basically any part of the system that can add to latency is able to be a culprit. i almost wonder if the motherboard has a greater impact than video cards do.

[Levish]

I'm sure, and making sure slot is running x16 pci-e 2.0. My p45 mobo for example makes both slots x8 even if its only a pci-e x1 in the other x16 mechanical x8 electrical slot.

[eRazorzEDGE]

I was just playing Dirt 2 online, so I decided to pick up a capture and this is what I got. Same system, 1920x1200 8xAA Ultra settings, With AFR, not the default Supertiling.




From the OP, for comparison:



Basically the same settings, except I'm using 8xAA instead of 4xAA, and I'm no where close to what they're showing. Except for that part about 2/3 into the run... dunno wtf that is

[Manicdan]

that was probably when you hit the water puddles in the map?

[eRazorzEDGE]

Not sure, I played a few races after that, so I don't even remember what map this was

[Manicdan]

also keep in mind how many frames your chart shows, if you stretched out that one section, it would be almost as wide as their entire chart. so your seeing a few spikes per second, their chart shows 20ish

[eRazorzEDGE]

Yeah, I stretched it out in OpenOffice and its' about 120 frames, and given the FPS in that field, roughly a second worth of larger than normal fluctuation. What's weird though, is their chart is fluctuating in the order of 15ms, while mine is only shifting at most 5ms on average.

I haven't seen anyone else do any comparisons, anyone else want to try this?

*EDIT*

@clip... I could only put 100 frames like the guys at that forum did, but that doesn't show anything but what someone is wanting you to see.

[Manicdan]

you can try to declock your chips so your framerate gets worse and see if that brings you in the range of 20ms per frame average, or turn on vsync and see what that does

[eRazorzEDGE]

I'll try to underclock the cards, see what happens.

Lol, I'm no whiz, but I did pick it up in about an hour this morning

It's simple really, I use OpenOffice Calc, just open the CSV file. It'll ask you how you want to format it, just select COMMA as the only thing separating the text. Then use this code... =$B2-$B1 in cell C2. Now copy that cell and highlight every cell in column C from C3 down to whatever row your text ends on and then paste. Highlight column C and click on the Chart button and VOILA!

[Pyr0]

If you have an ATI card, or cards, look in the registry. Search for MVPUmode... default is 33... 33 IS EQUAL TO SUPERTILING.

maybe it's different for the X2 cards, but in win7 and using the beta 8.69 package from msi, the default MVPU in my registry is 31
i've never used agent god's crossfire util.

[Eastcoasthandle]

Microstutter in action:

<snip>

Unfortunately this doesn't offer anything constructive because the frame rates were omitted.

[eRazorzEDGE]

maybe it's different for the X2 cards, but in win7 and using the beta 8.69 package from msi, the default MVPU in my registry is 31
i've never used agent god's crossfire util.

MVPU is 31 in mine also, I think that's crossfire enabled... but I'm not sure.

The next one down, MVPUmode is the one I'm talking about.

[Pyr0]

i don't have an MVPUmode entry anywhere in the registry

i'd hazard a guess that yours are from agent god's app

[eRazorzEDGE]

[JoeBar]

Great work eRazorzEDGE!