Are there "discrete" CUDA drivers?

SunnyD

Belgian Waffler
Jan 2, 2001
32,674
145
106
www.neftastic.com
I was thinking, as there has been some discussion that PhysX in a mixed video card environment (NVIDIA + ATI) is virtually impossible because the OS doesn't take kindly to multiple video drivers installed...

PhysX is based on CUDA no? I vaguely recall a product from NVIDIA... maybe it was Tesla? That wasn't a video product at all, but rather more of a pure CUDA product.

This leads me to the question, if NVIDIA could put out a ... erm ... "non-video" product like Tesla, would they not also be able to fairly simply create discrete CUDA/PhysX drivers as well, thereby allowing ATI cards to drive video and the NVIDIA cards to drive PhysX?

Of course this hypothetically situation would likely be answered with "Well why the hell would they want that? I mean they want to sell you a video card for video!" Well the reason is simple, aside from their greed in trying to go vendor lock-in until the ATI PhysX driver is hacked out, they have an absolutely great potential market for all those 8400's, 8500's and 8600's sitting on store shelves collecting dust and eating into their margins. It would be the PPU for the masses, doing what AEGIA only dreamed they could do.

Mayhaps keys and Rollo can look into this and see if NVIDIA's magic crystal 8-ball says anything other than "Outlook not so good."
 

WelshBloke

Lifer
Jan 12, 2005
30,449
8,111
136
Well why the hell would they want that? I mean they want to sell you a video card for video!


:p





(If you want physX with ATI cant you just get on of the original Ageia PPU's?)
 

QuixoticOne

Golden Member
Nov 4, 2005
1,855
0
0
It would be nice. One doesn't need the bloat of a full video driver just to load/run CUDA applications. It would help for GPGPU because it would save system memory and be more efficient on the host CPU usage.

It would give them an easier way to let people run CUDA GPGPU code on different operating systems where their video drivers don't exist or are not well supported up to the revision of the OS in question -- e.g. FreeBSD 64 bit, OpenBSD, some of the Solaris derivatives, et. al.

It would remove conflicts with X-windows and Microsoft Windows since you wouldn't have the device show up as a console video device, so you wouldn't have to load X or wouldn't be restricted from running GPGPU applications on a headless MS Windows box where you RDP into it.
 

aka1nas

Diamond Member
Aug 30, 2001
4,335
1
0
AFAIK, you still need to install forceware drivers for the Tesla cards so they can access CUDA.