Maybe a silly question, but I was just wondering about using 2 1GB 9500GT’s in SLI in order to get 2GB of memory available for CUDA ? Cause I think I heard that in general in SLI you don’t get the full 2GB with games and such so I was wondering if I get the full 2GB for use with CUDA in SLI?
SLI doesn’t work with CUDA at all.
Well that certainly answers my question. Too bad though.
The upcoming 2.3 toolkit allows CUDA to access both SLIed GPUs.
This does not change two GPUs into one giant uberGPU with doubled memory, but at least it finally allows you to use both GPUs for CUDA without disabling SLI.
So essentially what your saying is is that if I have 2 cards then they can’t share memory between themselves to run one task with this new tool kit but I can say GPU1 do this and GPU2 do that, each with its own 1GB of memory ? And then sync them up like having two parallel GPUs working on seperate parts of the same problem ?
No. Basically it just means in CUDA 2.3 that you won’t have to disable SLI in the driver to use both GPUs in CUDA. In the current release, you do.
I think I am misunderstanding what you mean by “use both GPUs in CUDA”. Do you mean that you can just use one GPU at a time in CUDA without having to disable SLI? or can you use both simultaneously?