View article

Davide Rossetti, S Team

2015/3

GPU Technology Conference

185

GPUDIRECT: INTEGRATING THE GPU WITH A NETWORK INTERFACE Page 1 DAVIDE ROSSETTI, SW COMPUTE TEAM GPUDIRECT: INTEGRATING THE GPU WITH A NETWORK INTERFACE Page 2 GPUDIRECT FAMILY1 GPUDirect Shared GPU-Sysmem for inter-node copy optimization GPUDirect P2P for intra-node, accelerated GPU-GPU memcpy GPUDirect P2P for intra-node, inter-GPU LD/ST access GPUDirect RDMA2 for inter-node copy optimization [1] developer info: https://developer.nvidia.com/gpudirect [2] http://docs.nvidia.com/cuda/gpudirect-rdma Page 3 GPUDIRECT RDMA CAPABILITIES & LIMITATIONS GPUDirect RDMA direct HCA access to GPU memory CPU still driving computing + communication Fast CPU needed Implications: power, latency, TCO Risks: limited scaling … Page 4 MOVING DATA AROUND D ata plane Control plane GPUDirect RDMA GPU CPU IOH HCA CPU prepares …