I have a Telsa S1070. The S1070 is attached to two linux servers. Each linux server can “see” two of the four GPUs using a single PCI 2.0 16 lane card.
I’ve been using: cudaHostAlloc((void **)&buffer, size, flags) with flags “cudaHostAllocPortable|cudaHostAllocMapped” to allocate portable and mapped pinned memory on the host.
If I allocate pinned memory and map it to GPU device 0, can I somehow map this same memory to GPU device 1 on the linux server, so the memory is mapped concurrently to both GPU devices?
Thanks in advance.