|Harnessing the massive computational power of modern-day Graphics Processing Units (GPUs) allows building extremely powerful clusters and super-computers. NVIDIA’s GPUDirect technology enables direct access to a GPU over the PCI bus eliminating unnecessary memory copies, dramatically lowering CPU overhead, and reducing latency.
With GPUDirect RDMA using Chelsio’s iWARP RDMA adapters, network access to the GPU is achieved at 10 and 40Gb speeds across a standard Ethernet network infrastructure with both high performance and high efficiency.
Enabling GPUDirect RDMA using Chelsio Adapters:
GPUDirect RDMA over 40Gbps Ethernet
GPUDirect over 40GbE iWARP RDMA
- Chelsio and NVIDIA Webinar on High-Performance GPU Clustering
- Chelsio Session on deployment of iWARP in GPU applications at NVIDIA GTC 2016
- NVIDIA GPUDirect
- GPUDirect Technology Overview
- iWARP: Ready for Data Center and Cloud Applications
- iWARP: From Clusters to Cloud RDMA
- Active development of iWARP RDMA standards continues