Can I use two different GPUs for CUDA programing? one will be CUDA capable the other is not. I already have a Quadro Non-Cuda capable GPU, I have to buy the CUDA capable GPU (GeForce). The not-CUDA Quadro will be in charge of displaying, the GeForce CUDA will be for General Purpose Computing. Is this configuration possible?
Yes
hey man are you sure? i about to go buy it! I have read a lot and everybody was saying that there was no way of doing stuff like that, but all the others were saying that they wanted to switch between playing and between several screens. What drivers should i use, Im new at CUDA and i only know the drivers i need to make a GeForce-CUDa work, but if i have both drivers( CUDA GeForce & Quadro ) wont they crash?? have u done this before?? thanks a lot, i though no one was going to answer me.
Well, now that you mention it, what system are you running? Windows Vista/7 and various flavors of Linux can support many different drivers running side-by-side and even cards from different vendors (like an ATI card for visuals and NVIDIA CUDA-capable card for compute, it has been done here on the forum AFAIR) but I’m not sure about XP.
I may be wrong. GPU is supposed to implement all the complex mathematical functions in hardware form, and then have some means of copying the output directly to the video ram directly, bypassing the CPU, correct? But now u are calculating it in one GPU, copy from the GPU memory to SRAM, and then copying to the VRAM of another GPU, is it going to work out fast enough?
But another way of dissection is possible (PURELY my guess…possible???) - physical screen-wise. So perhaps half the screen allocated for each GPU, and thus calculation can still be partition up at the screen level, and parallelized thus. But then the parallel-loop breakup have to be done at the CPU level - if this is not done ONE-OFF, then I think not much performance benefits, perhaps worst.
Please comment…I may be completely wrong :-(.