ATi for graphics + NVIDIA for CUDA possible?

Z

zz2

Limp Gawd
Joined
Jun 28, 2004
Messages
181
Would it be possible to have ATi and NVIDIA cards in same PC. ATi card for graphics and NVIDIA card only for CUDA processing??
 
Damn, I was wondering if I could do that too, haha :( Keep my 8800GTS and get a 4870 for the graphics.
 
Yes, it is possible.

You can mix and match graphics cards (to a certain extent).

You can use them both to power different monitors, or use one for a monitor, and the other for a different application.

People in special design and CAD do it all the time.
 
Its very true and one of the firsts things I learned studying for MS certifications for vista. Starting with Vista, Microsoft is planing on upgrading the driver models for several of the product lines (Audio, Network, Video). One of the restrictions on Vista for stability reasons it will only have one active Video driver running at one time. That is why certain video capture cards where caught out.
 
but it would be possible if NVIDIA made a seperate CUDA only driver?
 
Is cuda all that important to you? Then I'd just recommend you getting one of those overpriced cards anyway.
 
No, not right now, but NVIDIA works with developers so there is good chance of CUDA taking off in the near future (PhysX, video encoding...). It would be nice to know if I could buy ATi card + mb with 2xPCIe slots now, and later on if CUDA proves worthy buy additional card just for that.

PS: As far as the price goes, GTX280 sales here for 590€ which is ~924 US$ :(
 
ATI will do video encoding very soon, and the architecture could be feasible for Havok.
 
Nuker said:
Is cuda all that important to you? Then I'd just recommend you getting one of those overpriced cards anyway.
Click to expand...
CUDA is supported on any 8000, 9000 and 200-series cards. A GTX280/260 is not required for CUDA. As far as OS compatibility problems with multi-vendor video support in Vista is concerned, that's yet another reason not to 'upgrade' my XP installation. :rolleyes:
 
Stoly said:
Sure you can install ATI and nvidia cards on the same pc.
Click to expand...
I tested it. You get a driver mismatch error and the 2nd card (ATI/nvidia, in either order) is disabled under Vista. Really lame limitation.
 
No need to buy an ATi card for graphics and nVidia card for CUDA. I never thought I'd see the day but ATi has agreed to support 2 nVidia APIs physx and CUDA.

http://www.overclock3d.net/news.php?/gpu_displays/ati_falls_for_physx/1

You may recall a recent article on Nvidia's offer to ATI. Nvidia gave ATI the opportunity to use PhysX if they took up CUDA. Well, the result from this offer has been agreed! ATI have decided to go with Nvidia; this is not a huge surprise as ATI could have been out of a job if it didn't take up PhysX. Nvidia could easily force it upon game developers.

You may also remember ATI showing off its R600 GPUs with a demo of falling bricks. This was designed to discourage Ageia. They claimed that the R600s could process physics better than the PhysX card itself. Now that Nvidia can run PhysX on CUDA, ATI should be able to do the same. Both sides need to work and improve this complex piece of code.

The Big Lime Greens will surely be the first with PhysX on their graphic drivers, but we are sure ATI will follow soon enough. ATI does have a good position at the moment; they have now joined with Nvidia's PhysX and Intel's Havok, a WIN-WIN situation.
Click to expand...

Even after reading that I am still shocked an skeptical, it is hard to believe that ATi actually agreed to support 2 nVidia standards. I guess I will believe it when I see it. But if it is true it is great news as we will see much faster adoption of CUDA and physx and avoid a long drawn out API standard war.
 
I reckon its bull.
I say that with a vague hope it could be true but I'm so doubtful.

Perhaps NVidia have pre-empted the day that adoption will be needed on both sides for PhysX to really take off.
It would also mean NVidia take some of ATI's own pie for themselves in royalties.

Who knows at this stage :)
 
Particleman said:
Even after reading that I am still shocked an skeptical, it is hard to believe that ATi actually agreed to support 2 nVidia standards. I guess I will believe it when I see it. But if it is true it is great news as we will see much faster adoption of CUDA and physx and avoid a long drawn out API standard war.
Click to expand...

Well, ATI doesn't have their own version of CUDA (as far as I know - they might, so someone please correct me if I am wrong), so it is much cheaper and in their interests to use what has already been developed. It will allow CUDA programs to run on their GPU, which will benefit ATI. Same with PhysX - since AMD/ATI aren't personally involved in the creation of any, they can only stand to make gains by helping with GPU driven physics, regardless of the API in question.

Nvidia wants it to create developer support. Developers now can just use CUDA and PhysX and get the benefits nvidia is preaching, so nvidia gets increased market for their technologies.

It is a win-win for both side, I don't see why people would be so skeptical.
 
ATI very recently announced something very similar to CUDA, but not compatible with it. They previously had a low level API for GPGPU but not very directly comparable to CUDA. The new ATI SDK and CUDA come from the same academic project roots, so they should be very similar conceptually, if not in the details.
 
sethk said:
The new ATI SDK and CUDA come from the same academic project roots, so they should be very similar conceptually, if not in the details.
Click to expand...
No, CUDA and CTM are very different. CTM seems to be on the way out with AMD now talking about supporting OpenCL. Not a big surprise really since CTM never attracted the support CUDA has.
 
I was talking about Brook+

http://www.hpcwire.com/features/17910894.html

"To bring GPU computing to developers, AMD has provided its own FireStream SDK. Probably the biggest piece of the kit is the Brook+ development language, an AMD-enhanced version of the Brook programming language from Stanford University. Like its NVIDIA CUDA counterpart (which also got its start from Brook), Brook+ provides a C language environment with added functionality for data-parallel operations on a GPU. In order to encourage community adoption, AMD has open-sourced the Brook+ compiler front-end."
 
Its a great hoax if this is untrue :)

I'm wondering what the performance difference will be between competing gfx cards.
NVidia are effectively using 2 wrappers, CUDA and PhysX.
If ATI's implementation translates calls to/from PhysX, it will use 3 wrappers which will likely add a small overhead.
Hopefully it will talk directly to CUDA.
 
Nenu said:
NVidia are effectively using 2 wrappers, CUDA and PhysX.
Click to expand...
CUDA is not a wrapper, and PhysX is a library, now supporting CPU, GPU (written using CUDA) and PPU devices. THG has a good introduction to GPU PhysX and CUDA if you want to learn how it works. http://www.tomshardware.com/reviews/nvidia-cuda-gpu,1954.html

CUDA has 2 APIs, a high-level and a driver-level (low-level) API. There is some management that is pretty much a requirement in the CUDA library: shaders need to be loaded, run and results read back from the GPU. That functionality would be required no matter which card the shaders are run. Otherwise, any CUDA PhysX computation code would have been compiled as shaders when generating the library. There is not much overhead required.

It will be interesting to see how the hack required to run on AMD cards was written, and if it's running correctly instead of just discarding the results. If it requires one or more hacked nvidia display driver files, don't hold your breath for official support.
 
You must log in or register to reply here.
Back
Top