Do Nvidia GTX 590 cards work with CUDA 4.0 RC in Linux, so that one
can use the cores on both GPUs in the card for a total of 1024 cores,
and the entire 3 GB of memory?
How about 4 of these cards plugged into the same motherboard—will
CUDA 4.0 allow use of all 4 x 1024 GPU cores?
How about 8 of these cards plugged into the same motherboard—will
CUDA 4.0 allow use of all 8 x 1024 GPU cores?
You can use multiple GPUs at once in even the oldest CUDA versions. So yes, you can use the GTX590 and use all the cores and all the memory. And yes, you can use multiple cards.
But from the way you asked the question, you might be thinking that CUDA 4.0 somehow makes multiple GPUs merge into one single super-GPU. This does not happen in any version of CUDA. You still need to modify your code to partition work between GPUs, coordinate memory transfer between them, schedule kernels on each, etc. CUDA 4.0 just makes this multi-GPU programming easier. It does not eliminate it.
Hi,
I use 270.40 drivers, included in CUDA4.0RC2, as the only drivers for GTX590 (now 270.41, but the same situation). However today I found out that the card works ONLY!! at maximum 553MHz core clock, i.e. it reaches only to the “performance level 2” not 3 (according to the “Nvidia X server settings”).
Any ideas how I can use our cards at reference speed?
Thank you in advance!