This session will describe the development platform for NVIDIA GPUs for both HPC and AI that will be available on NCSA's Delta system. The NVIDIA HPC SDK is a suite of development tools for programming NVIDIA GPUs, including compilers for C, C++, and Fortran, with GPU offload models such as ISO standard language parallelism, OpenACC, OpenMP, and CUDA. The HPC SDK also includes a suite of GPU-accelerated math libraries and communication libraries. On the AI side, NVIDIA provides containers with fully optimized implementations of the popular machine learning frameworks for training and inference. NVIDIA will also describe the A100 and A40 GPUs and hardware features advantageous to users.
Delta is a dedicated, eXtreme Science and Engineering Science Discovery Environment (XSEDE) allocated resource designed by HPE and NCSA, delivering a highly capable GPU-focused compute environment for GPU and CPU workloads.