Eyefinity Crossfire Stuttering Resolved !

[Wijkert]

That is some seriously nice work Haldi! I chose Dirt3, because in a racing game microstuttering is most annoying (you need the flow the most). When playing it indeed feels like some frames come to late as if the driver tries to make up for lagging behind. This of course is annoying as hell. Will edit this post soon with a new frametimes.csv without crossfire enabled, so we can compare the two.

Edit: Here are the frametimes for exactly the same scenario(game, track, settings etc.) accept for crossfire being disabled. And here are my system specs:

2500K @ 4.5 (1.365v)
2x 6950 Dirt 3 edition
XFX Pro 650W
Z68X-UD3H-B3
Ripjaws 1600mhz
C300 64GB, 1TB F3
CM 690 II Adv.
Win7 Ultimate
12.1 preview (as for microstuttering all the drivers I tried, feel the same )

[Haldi]

Mah..... I'm Out all day. Eating Fondue <3

Will be back in 6till 9 hours.

P.s asked a friend of mine, he programmed a Java tool which will open a .csv file and create a .png from the Graph. So it's just gonna be one click work to get a nice looking graph!

Hope we can get as much Crossfire users to do it as possible! With a lot of Games!

[Wijkert]

P.s asked a friend of mine, he programmed a Java tool which will open a .csv file and create a .png from the Graph. So it's just gonna be one click work to get a nice looking graph!

Hope we can get as much Crossfire users to do it as possible! With a lot of Games!

Looking forward to that tool.

I must say I realy like this community. People tend to actualy help others, instead of flaming the hell out of people that ask for help.

[Haldi]

How much people tend to help others alway's shows how much interest they have themself!

So, here the new Graph. Looked especialy to make it from 0 to 35 ms like the other one! so you can Compare. (that might be a problem with the Programm, guess i'll make it from 0 to 60ms So from 999fps till 16 FPS its everything shown! Lower than 16 FPS won't work! but still..... from 0 to 35 looks quite good.... but just lower than 28 FPS won't show -.- Maybe making graph from Average ms +20-20ms.... might work, if programm supports it!)


And then there are the New Graphs that show the statistical Spread of Frames. Made by the Programm. You can change a lot of settings, but i hope i'll manage to modify it so it will work as one click! (And still able to change afterwards if you want to!)
(changed the NO CF one to maximum 1800 frames on Y axis, so you can compare!)
Edit: Added old Graph for comparission!

Attached files

[Cat Toaster]

I'm running a Sapphire 5970 Toxic (which are 2x5870s OCed to 900MHz on one PCB with 2GB VRAM each) on an Intel i5-2500k on an Intel DP67BG Motherboard for one year now. I only have issues with Eyefinity-Crap-Titles like the most recent ones: Skyrim, LA Noire, Witcher 2, Fable3 while others simply run great like BF3, Batman Arkham City (DX9), Deus Ex:HR.... (at least after several Game-Patches, AMD Drivers and CAPs)

Other games that run simply epic on that kind of setup are "Mafia 2" and "Just Cause 2". The rather old "Race" Series by SimBin, Civ5, Black Prophecy and the latter Total War Series-Titles all run great with almost maxed out settings on EF+CF.

I can't see how my Intel Board, CPU, Chipset are possibly affecting my gaming experience anywhere in a negative way. If it sucks, than it's all about the developers that fail to support Crossfire or Eyefinity or both properly.

Hyperthreading of course is a complete other issue in games (no matter which VGAs you use btw).

In my humble opinion if one starts building a solid system for EF/CF, this starts with the backbone of all: The Motherboard. And Asrock is not on my list of solid motherboard-makers, not even there "Premium-Mother" Asus is on that list. However, I'd blame Asrock for this particular experience and that nvidia-chip especially. Not Intel, not HT, no "few lanes" to blame in this case. At least I can´t confirm that.

[Wijkert]

In my humble opinion if one starts building a solid system for EF/CF, this starts with the backbone of all: The Motherboard. And Asrock is not on my list of solid motherboard-makers, not even there "Premium-Mother" Asus is on that list. However, I'd blame Asrock for this particular experience and that nvidia-chip especially. Not Intel, not HT, no "few lanes" to blame in this case. At least I can´t confirm that.

Well Asrock is a sister company of Asus. A couple of years ago they made cheap (crappy) hardware, but that isn’t true anymore. Besides, it is not the motherboard, since I am using a Gygabyte board.

If you want to help, maybe you can post a frametimes recording when both CF and EF are enabled.

[Haldi]

Delphium was so kind to provide me with some Frametimes.csv for Dirt 3 :)

On a i7-930 X58 HD5870 Crossfire.

Kinda hard to read something out of these.... HD5870 has more difference.... swapping between 8ms and 14ms ! HD6970 was more constantly! but then again, we don't have these fast Frames (2ms!) but also some slower frames but they are maximum 25ms. Not as high.

Attached files

[Wijkert]

It seems to me that there is more stuttering with the 5xxx series, but I believe that was already an excepted fact. Good to see the evidence though!

If people (especially people who don't have stuttering, or notice it at least) would post there frametimes it would help our comparison?

@Bushmaster, you said that buying an AMD system has helped you. Can you post your frametimes as well?

If in the near future someone on these forums has two 7xxx (the reviews of the 7970 got me excited) cards in crossfire, I would be really interested in frametimes, to compare with frametimes with the 6xxx and the 5xxx series.

[Wijkert]

After reinstalling win7 and updating my bios to 11f (UD3H-B3), it feel like those spikes are gone. The general stutter is still there though. Haldi can you please make a new graph using this file, so we can compare the two benchmarks?

[Haldi]

Nope i can't i'm in Hollydays xD i'll be back samewhen next year.

[Wijkert]

Nope i can't i'm in Hollydays xD i'll be back samewhen next year.

Watched your video a couple of times to figure out how make those graphs. You made a new column (sum=B3-B2), but after calculating that for one cell, I can't figure out to do that for the whole column. Maybe you can tell me how?

Anyway, have fun on your holyday!

[Gilly]

[quote]Nope i can't i'm in Hollydays xD i'll be back samewhen next year.

Watched your video a couple of times to figure out how make those graphs. You made a new column (sum=B3-B2), but after calculating that for one cell, I can't figure out to do that for the whole column. Maybe you can tell me how?

Anyway, have fun on your holyday!

You drag and hold the lower right corner (black box thing) all the way down. Or can use Fill Down command I believe!

[Haldi]

and Done.


Doesn't seem to be much of a difference! Even though in this part of the Graph (Frames 1500-2000) you have less, in the overall it's the same as you can see!

I7-2500K
HD6950 CF
Bios 11F

Attached files

[Haldi]

i know, double post's aren't welcome... but i didn't want to have a confusion with the attachements!

There you go! First Beta "sort my .csv file" script.

just open the script. Select the file. wait till "done" message pops up (automaticly disapers after 2sec, you also have a "work in progress" Tooltip at the left upper corner cuz it take some time.... 30-50sec) and Finished.
Now you have a new .csv file with exact same name, just a -edited added.

Thats a single row CSV (i know............) that shows Time inbetween each Frames in ms, which you can open with Excel or whatever you like to Create a Graph.

And you can also open it with "histmaker.jar" (which is pre-Beta and i truly shouldn't post it here.....)
For "Bezeichnung fuer Werte" you insert: Frames
"Ganze Zahl fuer Breite eingeben!" is for size of balk i always use 50
"DiagrammTitel" is the Title. Normaly i enter Game, GPU + CPU, +additional info
"Label für x-Achse" is "Time between Frames"
"Lable für y-Achse" is "Amount of Frames"
"Diagram Speichern": Do you wan't to save a .png Screenshoft of the file? Yes yo do! Select Location where you want to safe it and a Name.

Thats it! Now you can freely zoom and watch the diagramm as you like!
You can still click "Speichern unter" to make a new screenshot.



Still working on a fully automated way so you just have to open the file once.... but will take some time.

Attached files

BetaCSVTools.zip (1.9 MB)

[Wijkert]

That is some very nice work. The script that you posted works. I hope that people that care about this problem start using it, so we can get to the bottom of this.

[Haldi]

Updated Histmaker to v0.3 So now you just have to enter Title!

Way less work to do ;)

Attached files

histmaker 0.3.jar_.zip (1.6 MB)

[Wijkert]

Read this article on microstuttering again. The author says in the conclusion that he would always prefer a single fast card over two midrange cards in SLI/Crossfire. I would agree although running three screens makes it more difficult to get enough gpu power with only one high end card. In the graph on this link page you can see similar results as are posted in this thread, although SLI seems to have less microstutter. I would really love to test that out myself, but those cards aren’t available. Can someone test it for me.? Just post the frametimes.cvs and we can compare results. Thanks in advance.

[Haldi]

I absolutly don't get this graph....

First they say that you calculate the time between each Frame pops up, but then they make a graph which does NOT show time in milli seconds on the left side Y axis! why ? Because the less time in ms you have between each frame, the more FPS you get! So what's shown on the Y axis has to be FPS. but because there are normaly more than one Frame per second this is absolutly ridiculous. I assume they took the time between each frame and calculated the FPS from that value! unrealistic!

But it totaly agree with him ! Only go Crossfire/SLI if you absolutly have to and no other exception.

[Wijkert]

@Haldi

In your profile I see that you have 2x7970's, nice! How is the microstuttering?

[Wijkert]

I know, posting twice in a row sucks, but since this has nothing to do with my previous post, I will do it anyway.

I have been reading about the problem of microstuttering and have encountered this article. Haldi obviously already read it(since he introduced the whole frametimes instead of fps way of testing for smoothness), but it is a good read for people that want to know more. It seems that SLI has indeed less microstuttering compared to crossfire, but both technologies have this problem. In the conclusion the author said that using fraps for frametime recording might not be precise enough. Although it is probably still more precise then using fps. The above article is quite technical (even scientific). I like it a lot, but some may find that a bit too much. Both Nvidia and AMD say that they are working on a solution, but not for current technologies, but for future ones. That doesn't help us right now and probably not in the near future as well. In the following article the same author writes about a possible solution, which isn't open to the public yet. It is a form of better v-sync, which might also help out with lessening of microstuttering. Let’s hope it will.