Cloud-based HPC will reduce the need to maintain a strong on-premise systems engineering team and take advantage of the latest technology offerings without having to upgrade the on-premise resources. However, additional efforts will be required to migrate critical workloads to the cloud and to provision suitable resources for those workloads. Our certified cloud Solution Architects work with the customers to help migrate their workloads to the cloud based resources.
Contact UsBuilding and running most HPC codes are challenging tasks that involve resolving multiple dependencies (e.g., on external frameworks such as OpenBLAS, GlobalArrays, and HDF5), and selecting optimal runtime parameters (e.g., thread affinity and threads per rank for MPI+OpenMP codes). Even for expert HPC users, these deployment steps present significant barriers to adopting available HPC codes in their workflow. Containerization technologies can address deployment challenges by providing lightweight virtualized encapsulations of applications including all executables and libraries. A containerized application instance can be deployed effortlessly on many HPC resources – from leadership-class DOE systems and commercial cloud infrastructures, to personal workstations and laptops.
As part of the DoE Exascale Computing Project, EP Analytics containerized GAMESS, a quantum chemistry code, and its GPU-accelerated version, GAMESS-LIBCCHEM. The GPU-enabled container image of GAMESS-LIBCCHEM is available on the NVIDIA GPU Compute (NGC) cloud.
