Current high performance clusters are equipped with high bandwidth/low latency networks, lots of processors and nodes, very fast storage systems, etc. However, due to economical and/or power related constraints, in general it is not feasible to provide an accelerating coprocessor -such as a graphics processor (GPU)- per node. To overcome this, in this paper we present a GPU virtualization middleware, which makes remote CUDA-compatible GPUs available to all the cluster nodes. The software is implemented on top of the sockets application programming interface, ensuring portability over commodity networks, but it can also be easily adapted to high performance networks.

Paper available at ACM.