QLogic announced today at ISC that they are working closely with NVIDIA in order to optimize their TrueScale Infiniband products for the NVIDIA Tesla GPU platforms. They intend on tapping into the NVIDIA GPUDirect technology provided by the Tesla GPU hardware and software stack in order to reduce the overall CPU involvement in multi-node communication between GPUs.
HPC environments increasingly deploy multiple nodes with NVIDIA GPUs to speed up processing for computationally-intensive applications such as seismic processing and molecular dynamics modeling,” said Andy Keane, general manager, Tesla business at NVIDIA. “QLogic’s TrueScale InfiniBand architecture, optimized with our GPUDirect technology, natively supports this distributed computing environment with fully optimized NVIDIA-based GPU performance and industry-leading message passing performance.”
QLogic’s TrueScale InfiniBand architecture requires no application re-compilation or software patches to deliver computational performance with NVIDIA GPUs, and still maintain the scalable latency and high message rate for which QLogic InfiniBand solutions are known.
Our work with NVIDIA is another demonstration of how QLogic TrueScale InfiniBand goes beyond basic OpenFabrics implementations to fully optimize the performance of key cluster technologies,” said Jesse Parker, vice president and general manager, Network Solutions Group, QLogic. “Our software development groups continually strive to anticipate and meet the needs of HPC users who are implementing the latest technologies to maximize application performance, and our support of NVIDIA and its innovative GPUDirect technology is a great example of how those efforts benefit customers.”
For more info on the work between NVIDIA and QLogic, check out the full release here.