Accelerating HPC & Cloud-Based Scientific Applications
MVAPICH2-DPU
MVAPICH2-DPU is an optimized derivative of the widely used MVAPICH software, developed to harness the full potential of NVIDIA Bluefield DPU to accelerate HPC and cloud-based scientific applications.
Built on the strong foundation of MVAPICH Libraries, used by +3,400 organizations in 92 countries.
Up to 99% overlap of communication and computation with non-blocking collectives.
Designed to offload nonblocking for Alltoall and Allytoally to DPU, broadcast, send and receive, while also supporting sub-communicator non-blocking collective offloading and multiple outstanding non-blocking collective operations per MP rank offloading
Significantly accelerates scientific applications using any mox of MPI_Ialltoall, MPI_lalitoall, MPI_Ibcast, MPI_Isend, MPI_|rec nonblocking collective and point-to-point communications. See charts below for specific performance testing.
PERFORMANCE CASE STUDIES
Optimized Workloads for Scientific Applications
Questions on MVAPICH2-DPU?
Our team of experts are available to answer your questions and show you an MVAPICH2-DPU demo to help you optimize your scientific applications and accelerate workflows.
