GenL

Unless it is confirmed for other PhysX apps, it's a FluidMark issue i'd say, or something with a driver... Windows has nothing to do with it, i'm sure.
Try Warmonger, it's pretty physics-intensive. Maybe you'll notice the same thing with it... or maybe not.

ghost_zero5

OK. I have now tried some things. Interesting points first:
When using 190.38 it seems that I have a good PhysX performance if the CPU usage is high enough. PhysX FluidMark itself doesn't have too much CPU usage and therefore I ran into this slowdown issue once I everything was loaded.

So if I had to guess this is related to AMD's Cool & Quiet feature.
HOWEVER, I don't get why this has an influence at the PhysX performance if the GPU is doing the calculations? Seems like a driver bug.


@GenL:
I don't think that Warmonger is a good PhysX test application as it isn't so PhysX intensive as it is pretty old. Mainly because it doesn't uses newer PhysX features.
I can even run it fluidly with software PhysX of my processor only too.
OK, I have to admit here that it seems that the older PhysX features - those done be AEGIA - seem to be better Multi-Core optimized as Warmonger uses all CPUs equally. PhysX FluidMark or Sacred 2 only use ONE Core mainly which, of course, can't result in a good PhysX performance.

If I would use a game to test PhysX performance it would be Sacred 2 with the newest Patch as it is really PhysX intense (especially because of the particle count).

darthcyclonis

I don't notice a big difference in PhysX from Beta 1 to RC 1 to RTM. Of course Nvidia has had newer drivers out since Beta 1 and they were not that good.

ghost_zero5

Well I had a speed drop. Maybe not that high and I used the same driver at Windows 7 RC1 and Beta 1 but maybe RC1 changed some of the default Cool & Quiet settings.
I have now tested this by changing those settings in the control panel (Power settings) - there you can change the "Minimum Performance" of the CPU from default 5% to 100% and once I do this I have 44 FPS even when PhysX FluidMark is the only thing that runs.
But this really sounds like a driver bug as the CPU usage was only 10-20 % with the reduced speed. This shouldn't change anything when using Hardware PhysX which it definitely does because with the CPU alone I would never get above the 10 FPS in FluidMark.

Does anyone know if Cool & Quiet also slows down the PCIe speed - then this might make sense - or something like this? From what I know not because the reference speed should still be the same, just the multiplier becomes slower.

GenL

Sorry what? I can't run it with more than 25-30 fps on my q9450 3.6 ghz...

About Cool & Quiet... I'd say it's overall useless when it comes to stability + performance. For example, i remember that one pc with amd core and enabled Cool & Quiet feature was always giving a bsod at some particular game when loading some particulat level. Only disabling this feature helped.

Anyway, there are still more possible cases (other than pcie speed) i think - for example, PhysX still can consume some cpu performance, it's not about % of usage. We know almost nothing about importance of cpu in PhysX-on-cuda-on-gpu process. Nvidia wants to sell hardware, Ageia wanted the same thing. It's quite possible that the whole PhysX thing is really just a joke, and vendor is falsely decreasing performance for pure cpu process. Maybe it's exaggeration, but you can't say that it's completely wrong for ageia and nvidia devices... well that's offtopic

...or it may be some state when Cool & Quiet can't properly decide the needed state, and begins to switch them over and over again, resulting in additional cpu usage, which is not guaranteed to be displayed in Windows tools. That's just an opinion...

ghost_zero5

Actually about Cool & Quiet it depends on the CPU used. The Phenom (1) had a pretty buggy Cool & Quiet - from what I remember - , however the Phenom II fixed all of those issues (especially it now also works well with OverClocking - didn't do so with Phenom (1) - , I believe you can also define its speed values now) and what is the best included the possibility (I believe it was with Phenom II) to have Cool & Quiet per Core - meaning that only one core gets a higher speed when one core is enough for it.
And it is senseless to use it because you aren't always just playing. Of course, you could disable it for playing purposes but if it works fine otherwise, why should you? Take your BSOD example, it makes sense to disable it when you play that game but having it disabled the whole time, not really.

Regarding CPU importance: Actually, when your whole CPU isn't used, then PhysX doesn't consume much CPU which means that it doesn't require more CPU power or that there is a bug in the nVidia driver which means that it does need more CPU power but somehow isn't showing it. There are no other options. It isn't unlikely for it to be option two meaning it needs more CPU power but the driver isn't telling it or something like this. Especially, when I start PhysX FluidMark it has FULL CPU speed up to the point when it reaches the maximum particle count, once that is reached it uses the lower speed AND you probably already guessed so but once that count is reached also the CPU USAGE is about 10 % lower. However, that might be related to the point that up to that particle count there are always MORE particles generated and that is likely to require more CPU usage than just having the same particle count there.

Btw. the Q9450 has by default 2.67 GHz, did you overclock it?
Regarding Warmonger: I admit that I might have used the "fluid" a bit wrong BUT it is definitely very well playable, it could be better but it is playable.
That is why I said it isn't a good test-application because if you take Sacred 2 with newest Patch I get 2 FPS WITHOUT GPU PhysX. That is really bad

CQd44

I've been looking into why the Resident Evil 5 demo performed so horribly on ATI cards (specifically the 4xxx series) and came across this thread.

I managed to get PhysX working on Windows 7 RTM 64 bit with an 8800gtx with the 182 drivers (185.68 would not install as my card isn't listed in the setup file). My ATI card is a 4890.

I ran fluidmark and my framerate was a constant 60 FPS. I see some of you guys had framerate spikes of up to 500. Was it supposed to run at 60 FPS or did some of you just have high framerates at the beginning or something?

Also, as a side note, both GPUs can do Folding@home separately! The instance utilizing the 8800gtx folds at 1200 iterations/second while the ati 4890 folds at about 400. Good to know I get CUDA as well as PhysX!

Decebal

I think it's at least immoral to automatically disable PhysX when other non nvidia cards are detected. It should be our choice to enable or disable software components. We paid good money for their products and they just limiting our rights.
So, any modded drivers?
And where can I download the 182.xx drivers for win 7 x64 ?
(sorry for my bad english)

darthcyclonis

Not as of yet...They are being worked on.

CQd44

I need to make a correction. I installed the 181.71 drivers for windows 7. These can be found if you search for "beta or archived drivers" from the nvidia page.