I have a cluster with 4 nodes, each node has 2 video cards(nVidia GeForce 8800 GT). When I start MPI application each CPU must use only one video card. When I start MPI application and call cudaGetDeviceCount and cudaGetDevice. As a result cudaGetDeviceCount detects 2 devices and cudaGetDevice says, that 0 device is currently used on each node. Can 2 nodes use only one video card? Is it possible to get some unique information about CUDA device through CUDA API?
The programm gives something like that:
Running on 2 nodes Node 0 Device 0 of 2: "GeForce 8800 GTX" Node 1 Device 0 of 2: "GeForce 8800 GTX"