Running GTX 295 and Tesla C1060 together?

I am considering to buy GTX295 video card and Tesla C1060 together.
Is there anyone who ran both cards together making 2x240 (GTX295) + 240 (C1060) = 720 streaming processors?
GTX 295 has 1.5GB memory C1060 has 4GB. Can mutiprocessors in CUDA access all the memory (5.5GB) or only 1.5GB for 2x240 processors + 4GB for 240 processors?
Does anybody know the answer?

CUDA doesn’t work the way you are imagining. Each GPU is independent, with its own memory and driver interface, and each requires its own host CPU thread with its own GPU context. In the case of a GTX295 and C1060 in the same system, CUDA will see you have 3 GPUs, 2 “halves” of the GTX 295, each with 896Mb of memory, and 1 C1060 with 4Gb. You will have to program each GPU separately, or use something like the gpuworker framework if you want to use more than a single GPU for CUDA.

Here is an example of 3xGTX295 running on a single system under Windows XP, if you are interested.

It costs a lot of money!! :rolleyes:

GTX-295 + 3x Tesla work well together. In my benchmarks, the GTX-295 config with Tesla’s significantly outrun the same config with a FX5800 Quadro instead of a GTX-295. …and I think its more bandwidth related that the comparative number of cores in the two video cards!

But note that under Vista you cannot allocate the entire 4GB on the Tesla using a regular ‘cudaMalloc’. In most cases I think the max allocation is limited to 1GB.