ROCm 3.0 Arrives

The latest edition of the AMD’s innovative open platform for GPU computing offers a new compiler and support for some popular workload deployment tools.

AMD has released version 3.0 of the ROCm open platform for GPU-based computing. The latest release includes new innovations to support the LLVM-based HIP-clang compiler; upstream integration with the TensorFlow and PyTorch machine-learning frameworks; expanded support for the OpenMP, LAMMPS, and NAMD programming environments; and support for workload deployment tools like Kuberneters, Singularity, SLURM, and TAU.

The ROCm platform provides a vendor-neutral format for GPU-based parallel computing, allowing the programmer to write the code once and then compile for either the AMD and NVIDIA hardware environments. Automated porting tools allow for easy porting of existing CUDA programs to ROCm’s HIP format.