LLNL's Kripke Ported To AMD HIP With More HPC Software Seeing Radeon/Instinct Support

Lawrence Livermore National Laboratory now has their Kripke software ported to running on AMD's HIP for GPU acceleration.

Kripke as a simple, scalable, 3D Sn deterministic particle transport code solver is up to version 1.2.5 today and introduces AMD HIP support as an alternative to its existing NVIDIA CUDA support. The Heterogeneous-Computing Interface for Portability (HIP) is AMD's means of easing conversion of traditionally NVIDIA CUDA applications over to portable C++ code for execution on AMD graphics processors.

This LLNL open-source project has added AMD HIP support by way of their RAJA software library that serves as an abstraction layer for HPC software. Kripke has supported NVIDIA CUDA for a while and also OpenMP/MPI for CPU execution while finally there is AMD accelerated support with this native HIP support.

AMD HIP adoption was off to a slow start but in more recent times with AMD's increasingly competitive graphics hardware and more supercomputer wins, AMD HIP support in the upstream code-bases of more open-source HPC projects has been increasing. The AMD HIP software stack as part of ROCm continues maturing too and new projects come about like Orochi for continuing to improve the AMD GPU compute story.

Should Kripke be of interest to you, check out the new release with HIP support at GitHub.

LLNL, El Capitan

Lawrence Livermore National Laboratory continues their AMD support bring-up as part of preparing for the El Capitan supercomputer expected to go online next year with performance expected to be about 2 ExaFLOPS using AMD EPYC "Zen 4" CPUs with next-generation Radeon Instinct GPUs.