PhysX For CUDA, Linux Support A Given?

[phoronix]

Phoronix: PhysX For CUDA, Linux Support A Given?

Earlier this month it was revealed that NVIDIA Corporation would be buying up AGEIA Technologies, which is the maker of the PhysX SDK and the PhysX PPU (Physics Processing Unit) hardware. That same day we had then asked the question whether is NVIDIA buying AGEIA good for Linux? (The responses.) AGEIA had produced a PhysX software SDK binary for Linux but have never released a Linux driver to enable the offloading of these physics calculations to their PPU hardware. In NVIDIA's Q4'07 financial results conference call, it was revealed by their CEO, Jen-Hsun Huang, that they are already busy porting PhysX to using the CUDA interface and will be available as a free software update for GeForce 8 owners...

http://www.phoronix.com/vr.php?view=NjMzNA

[Svartalf]

This is interesting, but since CUDA is limited when you're running under X, unless they are supporting CUDA on a second GPU in the SLI slot not running in SLI mode, this is of rather limited usefulness.

[Michael]

This is interesting, but since CUDA is limited when you're running under X, unless they are supporting CUDA on a second GPU in the SLI slot not running in SLI mode, this is of rather limited usefulness.

You can run a second NVIDIA GPU with CUDA in a secondary PCI-E slot not connected to a monitor (and not necessarily an SLI board) while the first monitor is connected to a screen.

[yoshi314]

great. ati users left out, as usual :/

[Svartalf]

Then, that's AWESOME. I'd figured that they'd thought along those lines, but not having had a chance to play with any G80 hardware, I didn't know if there was "issues" with that configuration or not.

I guess if they're going to CUDA with PhysX, in addition to their dedicated hardware for it, I might be inclined to go looking for an SLI motherboard and Laptop, then...

[Svartalf]

great. ati users left out, as usual :/

Heh... I'm SURE that they'll get their act together...eventually. I'm a bit concerned about them claiming that the Workstation market's their main market for Linux and that they need to have closed drivers in that space- if that were honestly the case, you'd expect to see "better" results.

[ethana2]

We desperately need better GPU scheduling. OpenGL is used by everything from firefox3 to amarok, and compiz doesn't play nice. A GPU scheduler needs to find its way into the linux kernel, like the GPU memory manager did. It's just another processor, and it needs to be managed right.

[Svartalf]

We desperately need better GPU scheduling. OpenGL is used by everything from firefox3 to amarok, and compiz doesn't play nice. A GPU scheduler needs to find its way into the linux kernel, like the GPU memory manager did. It's just another processor, and it needs to be managed right.

/me nods his head...

Now, if only we could get that, get all the players to give us basic functionality on the cards in an open manner, we'd have quite a bit fixed in that space.

[deanjo]

Does anybody know if on a 2 GPU card, is Cuda able to access the second GPU on the same card? So in other words, if your using a dual GPU card 9800GX2, will you be able to use one GPU for X and the other for Cuda?

[Michael]

Does anybody know if on a 2 GPU card, is Cuda able to access the second GPU on the same card? So in other words, if your using a dual GPU card 9800GX2, will you be able to use one GPU for X and the other for Cuda?

Yes, I am fairly certain that's supported.