Advanced Tools for GPU Cluster
Résumé
In this lab, attendees will experiment with some leading-edge distributed GPU technologies that could strongly enhance HPC efficiency and productivity. We'll provide exercises on the ROMEO GPU-powered cluster for getting faster GPU data communications with GPU Direct RDMA and for using many virtualized GPUs through rCUDA. GPU Direct RDMA technology enhances the efficiency of data exchange between GPU and PCI Express devices, such as other GPU or infiniband networks. rCUDA is a virtualization framework that enables the usage of remote CUDA-enabled GPU devices locally in a transparent way.