3DMark06 flawed for multi-processors?

M

Mysterae

Gawd
Joined
Sep 11, 2005
Messages
648
I haven't seen anything like this posted before, but I could have missed it. Although this is really cpu related, I posted this in here as it's a more relevant audience.

While trying to narrow down the reason for certain games stuttering, I have found what I would call a flaw in 3DMark06.

3DMark06 and 3DMark05 displayed the same stuttering effect that I and many others have seen in games. I used those benchmark programs as they were easy and quick to test with.

Relevant System Specs:
AMD X2 4400 @ 2.64Ghz
X1900XTX & X1900CF (XTX @ 690/800)
2x 1G Ballistix @ 240Mhz 2-2-2-7 1T
DFI RDX200

Fresh reinstall of XP-SP2 all updated.

The only cure for the stuttering effect in 3DMark from this excellent thread was setting the affinity to one cpu core. It was after looking at the scores more closely I spotted something different.

Both cores being used in 3DMark06:

SM2.0 Score - 4100
HDR/SM3.0 Score - 4328
CPU Score - 1575
Total Score - 8419

Only one core being used in 3DMark06:

SM2.0 Score - 4118
HDR/SM3.0 Score - 4474
CPU Score - 1026
Total Score - 7266

Nothing wrong with that you may say, as the first (and higher) cpu score is using both cores, so the score should be higher. But that extra cpu (flaw) is hurting 3D performance as well as killing the overall score. However, the main flaw is that it's not utilising the cores correctly because of the stuttering in visual quality when both cores are being used - just like in many games.

So much for being optimised for multi-cpu's!

Of course, it could be my system that's causing this, but I doubt it. Go on, try it yourself!
 
Mysterae said:
Nothing wrong with that you may say, as the first (and higher) cpu score is using both cores, so the score should be higher.
Click to expand...

Why? You do realize that even in a correctly configured SMP setup, two CPU's doesn't mean perfectly scaled "double" performance? In other words, with a second CPU installed, you do not get a %100 increase in CPU power.
 
You didn't realize that once everything is optimized for dual cores both applications and drivers fight for CPU cycles?

If you take video drivers that are not optimized for SMP, applications that are will perform better and vice versa...such as why Quake 4 works better with SMP disabled when using newer video drivers that are optimized for smp.

Basically, the application/game is trying to do processing on the second core at the same time the video drivers are trying to offload work onto that same second core.

Now tell me a PPU isn't going to make a difference. :D
 
Sir-Fragalot said:
Why? You do realize that even in a correctly configured SMP setup, two CPU's doesn't mean perfectly scaled "double" performance? In other words, with a second CPU installed, you do not get a %100 increase in CPU power.
Click to expand...
You're putting words into my post! I never said it should be "double", only higher - as one would expect from an application that is supposed to be optimised for multi-processors. The way I see it the cpu test is optimised for multi-core but the rest of the program isn't, leading to an erroneous final score.

BBA said:
You didn't realize that once everything is optimized for dual cores both applications and drivers fight for CPU cycles?
Click to expand...
I didn't actually. I thought it would all work in harmony and give a pleasant gaming experience :p j/k! Isn't how you describe it how single cpu systems work? Is it this "load switching" that's giving me this stuttering in 3Dmark and games? And yes, a PPU will make a big difference depending on how much % cpu load is taken up by physics calculations, but will this be negated by the cpu required to make it work and the craptastic dual core drivers they will write for it :D !

I think the main flaw has been missed by your replies - the tests stutter when both cores are used. It's only really visible when the fps are above 30 as it's masked by the slideshow. Perhaps the reason it's not been reported yet is because not many get to see 06 above 30fps and many that do are only concerned with a high score than image quality. There, I said it!
 
Might want to check your system config or drivers more.

Mysterae said:
The only cure for the stuttering effect in 3DMark from this excellent thread was setting the affinity to one cpu core. It was after looking at the scores more closely I spotted something different.
Click to expand...

I had stutters in 3dMark06 and a couple games until I installed the appropriate patches in that thread:
- AMD X2 CPU driver
- MS Hotfix
- Registry tweak to enabled hotfix

That resolved all the problems I had in all titles, *and* 3dMark06. I have never had to set affinity on any title to get it working stutter-free with an X2 CPU.

And I *think* there are more than a few of us with X2 CPUs running 3dMark06. If this problem was endemic of the application, it would have been noticed DOZENS of times, if not HUNDREDS....but it has not been.
 
cyks, read your thread but see that it didn't come to much, did you just settle with the affinity?

dderidex - I tried all those things and tested right after the application of each one - nadda - didn't fix anything. Set the affinity and wham - complete night and day change!

dderidex said:
And I *think* there are more than a few of us with X2 CPUs running 3dMark06. If this problem was endemic of the application, it would have been noticed DOZENS of times, if not HUNDREDS....but it has not been.
Click to expand...

I suppose I was being an arse when I made the comment about not everyone seeing it, but I'm going to stick to my conviction. My comment was not about people with X2's, but to the fact that not many people see 3DMark06 above 30fps so they wouldn't see the problem. Glad your system is stutter free. Mine is now, with a certain affinity :D
 
dderidex said:
Could just be a problem with the chipset or Crossfire in general, though.
Click to expand...

Could be, and very simple to find out.

I would ask anyone with a dual core to run 3DMark06 as they do normally and note all four scores down (total, SM2, HDR/SM3, CPU) then reboot. After rebooting run 3DMark06 again but this time set the affinity to a separate core (in my case it was the second one) and post the scores.

Be interesting if there is a difference between ATI and NV on this.
 
dderidex said:
Could just be a problem with the chipset or Crossfire in general, though.
Click to expand...


Last time I read that post.. jpinard complains about slow downs with those drivers on single core cpu's, not dual core cpu's. I could be missing something but I'm not going to read teh whole thing again.

BUT: In all honesty, the nforce 4 chipset does have a few problems with smp. The ATi (R200 at least) works fine with dual core and does not have the errors the nforce chips do. I verified this against both NF4 SLI (3800 X2 and 175) and NF4 Pro SLI (with dual 254) against the R200 with several programs, such as Creative Labs DVD-Audio player, game performance, etc... The cure for nvidia is to force affinity to single cpu for the offending application, while the same applications have no problems on the ATi chip systems.
 
Yeesh, just add /usepmtimer to your boot.ini and don't bother with all that affinity garbage
 
I have the problem in this ATI chipset. Setting the /usepmtimer has no effect.

It could be the ATI drivers that's causing the stutters, don't know. Since I have X1900's, there's little selection of drivers to try - ones from the cd or 6.2.
 
I now do not think it's a problem with 3DMark - pass the ketchup and I'll eat my words! However, it's still happening and haven't found a fix. CPU, motherboard, drivers? argh!

Here's some test figures to illustrate what's happening (it's just easier to show with 3DMark):

OC_prob_01.gif


I posted this on DFI-Street incase it was a problem with the RDX200 mobo, but still nothing. Am I the only person having this issue?
 
You get similar results with games that support SMP, turning off smp gives better performance with some graphics driver.
 
You must log in or register to reply here.
Back
Top