Microstuttering tests on ATI Radeon HD 4870 X2

[Sampsa]

Hi,

I'm working with full article of microstuttering issue including tests in Crysis and Race Driver: GRID with ATI Radeon HD 4870 X2, ATI Radeon HD 4870 CrossFireX, ATI Radeon HD 3870 X2 and GeForce 9800 GX2.

Article will be published on Finnish hardware and overclocking site Muropaketti.com and translated to English.

Here is a sneak peak of what to expect:





I'm also waiting answers from AMD regarding this issue which I'll update to the article later.

-----------

Update:

Here are charts showing the difference between frametimes with ATI Radeon HD 4870 X2, 4870 CrossFireX and single 4870 in Race Driver: GRID:



And GeForce 9800 GX2 and GeForce 9800 GX2 in single GPU mode:




Still waiting input from ATI for my questions about microstuttering.

---

Update 2

Ok here is a short explanation about testing methods and graphs:

First of all, I'm using Race Driver: GRID demo and CrossFire and 9800 GX2 seem to work just FINE. With 1920x1200 resolution, 4xAA and 16xAF I get 66,9 FPS with single 4870, 73,4 FPS with 4870 CF and 75,6 FPS with 4870 X2:



(Red bar keskimääräinen is average FPS)

Second, the graphs I've showed here related to microstutterin ARE NOT intended to display performance but difference between frame rendering times. I have only recorded 2 seconds of gameplay and used first 30 frames from data.

If performance is 60 FPS.. that would mean graphics card renders 120 frames in 2 seconds. Since I have only used 30 frames to analyze the results and draw the graphs, it means gameplay of half a second (0,5s, 500 ms)!!! That's not enough to draw conclusion about perfromance...

... it might not be enough to draw conclusion about microstuttering either but in my own tests 30 frames already show clear pattern with ATI Radeon HD 3870 X2 and rendering problem where every other frame is being rendered after different time than every other.

[kromosto]

nice test

and nice to see average for new ati cards is nearly 2.5ms maybe less as seen on the graph .

[fellix_bg]

Wow!
3870X2 is bumping the frames like mad! That would be awful feeling, especially in a fast racing game.

I wonder, can this behavior be demonstrated properly on a video record w/ FRAPS?

[Shintai]

So 4870X2 card doesnt hold any "microshutter" advantage.

[MpG]

These tests are being done with a fast processor? One of the theories/thoughts put forwards has been that a processor that's "too fast" will contribute to the microstuttering, since the GPU's get information too quickly.

Looking at that graph, the worst offender is also the slowest card by a noticeable margin. A faster graphics card(s) would process the given information faster, reducing the tendency for the frames to appear vastly out of sync.

[Sampsa]

These tests are being done with a fast processor? One of the theories/thoughts put forwards has been that a processor that's "too fast" will contribute to the microstuttering, since the GPU's get information too quickly.

Looking at that graph, the worst offender is also the slowest card by a noticeable margin. A faster graphics card(s) would process the given information faster, reducing the tendency for the frames to appear vastly out of sync.

Done with Intel Core 2 Duo E8500.

[bowman]

These tests are being done with a fast processor? One of the theories/thoughts put forwards has been that a processor that's "too fast" will contribute to the microstuttering, since the GPU's get information too quickly.

I guess that's where the 'smoothness' of the Phenom comes from.

(sorry, couldn't resist, disclaimer: I have three AMD processors)

In any case, didn't expect anything better, and won't be going multi-GPU any time soon..

[Hu1kamania]

Seeing as how vsync apparently fixes a lot of (if not all?) of microstuttering, I would like to see more multi GPU graphics compairisons with vsync enabled in order to see which graphics cards can keep a higher minimum frame rate. Vsync makes for an overall more enjoyable play experience...except in twitch shooters or other games when reaction time is more important, as it causes input delay.

[003]

Test needs to include dual 4870X2s and also a single GPU configuration for comparison. Otherwise, great results, it appears that CF micro stuttering has been reduced to nil with the HD4000 series.

[annihilat0r]

Vsync is not good in multi GPU due to the lack of triple buffering.

[GrimReaperGuy]

The graph should really begin at what is labeled 'Frame 2'; there is no data before that.
Did you test a single card also? The addition of a control would make the lower end of the data set more meaningful.

[Rammsteiner]

I guess that's where the 'smoothness' of the Phenom comes from.

(sorry, couldn't resist, disclaimer: I have three AMD processors)

, nice joke though.

Anyhow, I would have expected they fixed microstuttering really. If they wouldnt have fixed it I think they would have let down quite some customers despite the performance.

[003]

Seeing as how vsync apparently fixes a lot of (if not all?) of microstuttering

It also inherently introduces mouse lag simply by the way it works, and this is easily noticeable in fast paced FPS games such as UT2004.

[Seraphiel]

Strange results... Only the 3870 X2 suffers from any significant microstutter.

[JohnZS]

Interesting, so reading the graphs it suggests that the HD4000 series cards have improved greatly on Micrstuttering over the previous generation of cards, and that SLi (on the 9800 GX2) is a shade worse than the 4000 series but nowhere near as bad as the 3000 series Microstutter!

[annihilat0r]

Oh.

DAMN SLI.

I just moved to a 4870 from a 8800GT SLI setup. In Crysis, the framerates are lower. THE GAME IS MUCH SMOOTHER. THE CONTROLS ARE DEAD ACCURATE. Even 25FPS could be considered playable whereas in SLI controls sucked below 40FPS

Damn ME if I ever buy a SLI or CF system again.

[Eson]

So let me get this correct; singel GPUs has perfect frame sync? Like a new frame every X.ms?

[Seraphiel]

So let me get this correct; singel GPUs has perfect frame sync? Like a new frame every X.ms?

No they do not, but they are pretty close to it. It all depends upon the scene and what is to be rendered.

[Akabeth]

The 4870x2 have sealed its place in my checkbook

Thanks very much

[BulldogPO]

Sampsa, did you test HD3870 X2 with same driver version also?

[Hu1kamania]

It also inherently introduces mouse lag simply by the way it works, and this is easily noticeable in fast paced FPS games such as UT2004.

Yeah, thats was what I meant when I said it causes input delay.

[Macadamia]

Interesting, so reading the graphs it suggests that the HD4000 series cards have improved greatly on Micrstuttering over the previous generation of cards, and that SLi (on the 9800 GX2) is a shade worse than the 4000 series but nowhere near as bad as the 3000 series Microstutter!

Being second last place in this case doesn't give it any awards. If anything it means that 8800GT/GTS/98GTX SLI is probably as bad.

If it's noticable in high res that the cards are supposed to play at (3870X2, 9800GX2) it sucks. No passing through that.

[adamsleath]

Damn ME if I ever buy a SLI or CF system again.

...so im not the only one who was turned off? thought i was alone there...

singel GPUs has perfect frame sync?

what's to sync? it only has one gpu

to sync or not to sync. that is the question.

can cf run well with vsync off?

[Hu1kamania]

Vsync is not good in multi GPU due to the lack of triple buffering.

It may not be optimal, but what is the harm?

[fireice2]

Can't wait to see the full article!