Widescreen Gaming Forum

After I failed to use a pci-e x16 8400GS with a x16 to pci-e x1 adapter as my third card, I've gotten a native x1 HP 8400GS which works.

The 180.48 drivers are installed for both 8800 Ultra and 8400 GS and I've verified that I can enable phyx processing on it with fluidmark:

physx disabled: 50+% cpu use, ~10fps
physx on 8400GS: 1-5% cpu use, ~10fps
physx on 8800Ultra/sli: 1-5% cpu use, 60fps (capped by vsynch)

Now I'm wondering, does the 8400GS simply suck at physx or is there something else wrong? I did have to manually add it to the driver .inf for it to install since only 8800 and up are in the current 180.48 drivers.

Chances are that it sucks at PhysX calculations... IIRC, the 8400 isn't just a hobbled card when compared to the 8600/8800 cards - it's totally castrated.

True, but it is doing the exact same performance as a 3ghz core 2 duo though.

So, I guess while it won't improve physx performance, it'll free up about 45% of the cpu for something else, and that'll likely help since that'd be 90% of the first core for single-threaded titles (aka, most games).

Of course, I'm still wondering about that whole, need to edit .inf, part, maybe it doesn't work that well if I have to add it in manually.

Hmm, guess I should've verified it before posting it from memory,

I just redid em and the results aren't what I remembered:

physx disabled: ~50% cpu use, ~10fps
physx on 8400GS: 1-5% cpu use, ~10fps
physx on 8800Ultra/sli: ~60% cpu use, 60fps (capped by vsync)

So that leaves me to think even if fluidmark says it's hardware physx, it's not working the same. And it's quite odd that it would use so much less cpu for the same thing as 2 8800 ultras, unless that's the sli cpu overhead.

Unfortunately, I can't dedicate one 8800 to phyx and the other to gfx since they list as one in the listbox, regardless if sli is enabled or not.

I've lost some of my interest in PhysX thanks to the lack of new games using it, but from what I read (and iirc) Fluidmark does not quite mimic the actual use of PhysX hardware in games. It was just designed to give th better results with beefy cards. So don't draw too many "logical" conclusions from your results...
Other than that, 8400 GS on PCI-E 1x sounds like it's under minimal requirements for CUDA to even work and for the rest of the system to be aware of it :nudgenudge

And last time I checked, besides limitations bound to SLI, the secondary card had to be connected to a monitor too before you can select it as PhysX hardware under Vista.

Well my results just seem totally off. I mean, 60% for the same physx on 2 8800 vs 1-5% for one 8400GS just doesn't fit.

I don't know what exactly the choice of the 8800ultra means either, is it that it only lists the 2nd 8800 as in dedicate one of the 2 to sli or if it considers both as one 8800 in the list.

I do know that the hard lockups I've associated with physx being enabled on sli-8800 also happen when the 8400 is supposed to be working so either they have a common cause/physx driver issue or I pick 8400GS and it's still doing the work on the 8800, just doing a bad job of it, hence, little cpu use. Like it's running on the 2x8800 but at 8400 speed or something.

I've mainly used fluidmark because everything else in the 'geforce phyx pack' or whatever, is long loading, crashing, crap.

Right now the 8400 has no monitors attached, and it is listed in the physx choice too.

"8400 GS on PCI-E 1x sounds like it's under minimal requirements for CUDA to even work" <== compared to what exactly?

Well, I don't know, compared to a GeForce officially supporting hardware PhysX...

I read your results like this :
- the 8400GS lifts the PhysX burden from the CPU. No "visible" change in Fluidmark except the numbers, but in a real PhysX game with AI and stream loading of levels and whatnot, you can expect better framerates from using the 8400GS. In terms of PhysX, the 8400GS can't do more than the CPU, but it does it somewhere else. Performance in a real game would probably still suck, but not directly because the CPU is crawling under PhysX calculations.
- the 8800s don't just use more CPU time for "the same PhysX" : it's because they allow a 600% framerate boost that they need (and can) "dialog" with the CPU more intensively, too. Don't forget that hardware PhysX still needs CPU time, not to calculate the PhysX events anymore, but to integrate the PhysX calculations from the GeForce to the rest of the data flow. (Erm, I don't mean I know how all this really works, but that's how I see it.) So, the more PhysX calculations there are, the more the CPU load increases. Only this time, the framerate increases too, and pretty dramatically.

Yeah, that's pretty much how I read em too, and if it wasn't for the constant hard lockups that happen whenever physx is enabled (on 8400 or 8800s) I'd leave it set on the 8400.

So far though, performance isn't any better with it or with sli so for the time being, I've disabled the 8400 and 1 8800 and running even without sli.

No loss of performance that I can tell.

Sli is marketing bs in my case it seems, gonna wait a long time before I need to buy a new card though, the current gen aren't much faster to begin with.

PhysX is BS, because it has no practical use and only raises theoretical questions that can't be answered.
SLI is not ;) but that's another story.