Remote direct memory access for GPU Nodes
This project contains information and scripts are used in Data Injection Report: Benchmark of communication between HPC nodes.
- ibverbs
- cuda (8)
- mpi
- Two compute node
- Nvidia - GPU Direct RDMA compatible cards (at least one per node)
- Infiniband compatible network interface cards (at least one per node)