• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Mutli GPU solutions Worthless?

Dr_Fartypants said:
...
How about this - tonight I will look into the best and most reliable way of recording frame output times. I'll write up a mini-guide on how to do it, and make a thread. Users on this forum with both single- and muli-GPU setups can then report back their results in a variety of games, and we will be able to quantify the magnitude of this effect. Sound good?
Click to expand...

That sounds fair. Though I'm willing to bet numerous people will be surprised at their findings. At least Fraps is objective, people looking at the screen for something they don't want to see it is subjective and a bit of psychology is also in play. It very much reminds me of the rainbow effect with single dlp projectors, some people see it, some people don't.
 
I can remember reading an article about this in relation to the brand new Rage Fury Maxx card that was available at the time, and the article went into depth into why alternate frame rendering was a really bad idea, and produced the kind of problems that you are discussing here.

I really do agree that we do need a more meaningful measure of performance than frames per second. Even minimum frames per second does not give an accurate reflection of the fluidity of gameplay, as at the end of the day this is what we notice when playing games.

With regard to all the guys that report no problems with their multi gpus setups, then that is good news, but I expect the problems only to be visible when the cards are really stressed, and there are not a lot of games that do that with the current multi cards.
 
tdream said:
That sounds fair. Though I'm willing to bet numerous people will be surprised at their findings. At least Fraps is objective, people looking at the screen for something they don't want to see it is subjective and a bit of psychology is also in play. It very much reminds me of the rainbow effect with single dlp projectors, some people see it, some people don't.
Click to expand...

This is starting to feel a lot like the LCD vs CRT debate. All I can say is that if I had experienced this problem as profoundly as others have, that 2nd card would have come straight out and gone in another PC or been sold off.

A 2nd card simply allows me a much smoother game play experience at 1920x1200 with all the aa/af I want, I really can't put it any other way.
 
Well the great Finnish overclocker Sampsa has his 4870x2, here is his post (copied from XS)

-----------------

Good news! I can confirm that based on my own tests microstuttering is gone on R700!

I've tested with R700 (ATI Radeon HD 4870 X2) and R680 (ATI Radeon HD 3870 X2) in Crysis (1600x1200 and High settings). I used Fraps and enabled Frametimes logging. I recorded 2 seconds from exactly the same point in game (loaded from save game). Based on my recorded data, with ATI Radeon HD 3870 X2 frames are rendered after ~21,5 and every other frame after ~49,5 ms. With ATI Radeon HD 4870 X2 all framres are rendered after ~ 21,9 ms.

-------------------

He notices the discrepancy too, but fortunately it seems to be gone from the 4870X2 :)
 
tdream said:
Well the great Finnish overclocker Sampsa has his 4870x2, here is his post (copied from XS)

-----------------

Good news! I can confirm that based on my own tests microstuttering is gone on R700!

I've tested with R700 (ATI Radeon HD 4870 X2) and R680 (ATI Radeon HD 3870 X2) in Crysis (1600x1200 and High settings). I used Fraps and enabled Frametimes logging. I recorded 2 seconds from exactly the same point in game (loaded from save game). Based on my recorded data, with ATI Radeon HD 3870 X2 frames are rendered after ~21,5 and every other frame after ~49,5 ms. With ATI Radeon HD 4870 X2 all framres are rendered after ~ 21,9 ms.

-------------------

He notices the discrepancy too, but fortunately it seems to be gone from the 4870X2 :)
Click to expand...
That is fantastic, hope to see some nice pricing on this card, then i'll have one on the way :D
 
I fail to see how SLI could be a/effected by 'microstuttering', since one GPU works out odd number of lines and the other GPU even lines. I just assummed multi gpu cards worked the same way.. :)
 
keyser said:
I fail to see how SLI could be a/effected by 'microstuttering', since one GPU works out odd number of lines and the other GPU even lines. I just assummed multi gpu cards worked the same way.. :)
Click to expand...

Eek!

Yeah, that's how 3dfx's implementation worked, back in 1998! These days the increased complexity of scene-rendering would require far too much inter-GPU communication to make this a viable solution. Both ATI and nvidia use alternate frame rendering techniques for the most part these days. AFR requires the least inter-GPU communication, and so is the way to push out the most number of frames when averaged over a set period of time.



Also - fantastic news about the 4870x2, if it turns out to be true in general :D I'll be eagerly awaiting the details of the inter-GPU interconnect implementation to see how AMD solved the problem :)
 
Last edited:
tdream said:
That sounds fair. Though I'm willing to bet numerous people will be surprised at their findings. At least Fraps is objective, people looking at the screen for something they don't want to see it is subjective and a bit of psychology is also in play. It very much reminds me of the rainbow effect with single dlp projectors, some people see it, some people don't.
Click to expand...

I think the easiest thing, to get consistent and directly comparable results, would be for me to write a small program to read in FRAPS data and perform local level statistics. This way, people can choose whatever game they want and run for however long they want. I can also post the source if people are worried I'm somehow biased.

Might not get around to it tonight (got other stuff I need to do) but I'll do it some time this weekend. If the 4870x2 really has cracked the 'microstutter' issue, it would also be a nice tool to demonstrate this.
 
I think the main problem with this entire concept is: Does it really matter? What delta between frametimes do you need for this effect to become noticeable? Is it even noticeable if you're not looking for it? Does it even happen with every dual card rendering mode (AFR, SFR, SLI, there're quite a few)? A lot of people I see with dual card setups claim that they don't notice it or haven't had it.
 
Lightnix said:
I think the main problem with this entire concept is: Does it really matter? What delta between frametimes do you need for this effect to become noticeable? Is it even noticeable if you're not looking for it? Does it even happen with every dual card rendering mode (AFR, SFR, SLI, there're quite a few)? A lot of people I see with dual card setups claim that they don't notice it or haven't had it.
Click to expand...

Whether it's noticable will be subjective, and could vary significantly from person to person. The effect will also become less noticable at higher framerates (but then the benefit of adding a second card reduces significantly at higher framerates anyway).

It should only occur with alternate frame rendering modes (AFR), but this is the mode used in 90% of games anyway, since it pumps out the most frames when averaged over sufficient time, and so looks the best in benchmarks.

A quantification of the extent of the effect will prove (or disprove) that the phenomenon actually exists. What it won't change is whether or not people notice it. However, it will be more information for people to consider before choosing their next graphics setup. In my eyes more information is always good, whether you choose to look at it or not.
 
Okay, I've written the program :)

It's very simple, but it seems to work well enough. I've got to do other stuff now, but I'll test it later tonight and make a thread with it tomorrow so we can put this 'microstuttering' concept to rest once and for all. For the sake of our children, and our children's children :p


If someone with a multi-GPU setup (or even better one person with SLI and one with crossfire) could send me a sample FRAPS output that would be nice. If you could run a FRAPS benchmark in crysis (say) for something like 60 seconds or so in-game, that would be great. For now, just stare idly into space in a quiet location. In theory it should work walking around or in action just as well (as long as there is no loading), but I'll test what effect that has on the stability of the results later. You will need to select the 'frametimes' tickbox in the 'FPS' tab of FRAPS, then run the FRAPS benchmark in-game (default key is F11 I think).

If you're willing to do this, please send the resulting *.csv file saved in the FRAPS/benchmarks folder to:

arseface1 (at) hotmail (dot) com :)


Also, can anyone recommend a reasonble free file-storage system that I can use to store the .exe? It's not big, around 60kb when zipped.
 
Last edited:
You must log in or register to reply here.
Back
Top Bottom