20230615_A-202_Dai.PDF

《20230615_A-202_Dai.PDF》由会员分享，可在线阅读，更多相关《20230615_A-202_Dai.PDF（22页珍藏版）》请在三个皮匠报告上搜索。

1、Accelerating HPC and Deep Learning(DL)Applications with SmartNICsDonglai Daid.daix-http:/x-OutlineOverview of X-ScaleSolutionsOverview of Products MVAPICH2-DPU:High-Performance MVAPICH2 for Accelerating Applications with NVIDIAs DPU technologyX-ScaleHPL-DPU:Accelerating High-Performance Linpack Code

2、(HPL)Benchmark with DPU OffloadX-ScaleAI-DPU:Accelerating DL Training with DPU Offload2Overview of X-ScaleSolutionsStarted in 2018,bring innovative and efficient end-to-end solutions,services,support,and training to our customersCommercial support and training for the state-of-the-art communication

3、librariesPlatform-specific optimizations and tuningApplication-specific optimizations and tuningObtaining guidelines on best practicesTimely support for installation and operational issues encountered with the libraryFlexible Service Level Agreements Web portal interface to submit issues and trackin

4、g their progressInformation on major releases and periodic information on major fixes and updatesHelp with upgrading to the latest releaseWinner of multiple U.S.DOE SBIR grantsMarket these products for HPC and AI applications with commercial supportA Silver ISV member of the OpenPOWER Consortium Out

5、lineOverview of X-ScaleSolutionsOverview of Products MVAPICH2-DPU:High-Performance MVAPICH2 for Accelerating Applications with NVIDIAs DPU technologyX-ScaleHPL-DPU:Accelerating High-Performance Linpack Code(HPL)Benchmark with DPU OffloadX-ScaleAI-DPU:Accelerating DL Training with DPU Offload4MVAPICH

6、2-DPU Library 2023.05 ReleaseBased on MVAPICH2 2.3.7Supports all features available with the MVAPICH2 2.3.7 release(http:/mvapich.cse.ohio-state.edu)Novel GVMI-based framework to offload non-blocking collectives to DPUOffloads non-blocking Alltoall(MPI_Ialltoall)and non-blocking Bcast(MPI_Ibcast)to