Radeon ROCm 2.10 Released With SUSE 15 SP1 Support, rocBLAS For Complex GEMM

While AMD announced ROCm 3.0 earlier this month at SuperComputing 19 as the next major iteration to Radeon Open eCosystem, it looks like they aren't quite ready to ship it and instead released ROCm 2.10.

ROCm 3.0 support was noted in their SC19 press release with having new HIP-clang improvements for better running CUDA codes on Radeon GPUs, expanded HPC programming model support with the likes of OpenMP / NAMD / LAMMPS, and new support for system/workload deployment tools like Kubernetes. ROCm 3.0 sounds like it should be a great release, thus surprising today they diverted to releasing ROCm 2.10.

Whatever the case may be, ROCm 2.10 is out today. ROCm 2.10 comes with rocBLAS support for complex GEMM. The release notes explain this change to their GPU-accelerated BLAS implementation, "support is extended to the General Matrix Multiply (GEMM) routine for multiple small matrices processed simultaneously for rocBLAS in AMD Radeon Instinct MI50. Both single and double precision, CGEMM and ZGEMM, are now supported in rocBLAS."

ROCm 2.10 also comes with official support for SUSE Linux Enterprise Server 15 SP1, code marker support within the rocProfiler and rocTracer libraries, and a handful of bug fixes. There isn't any Navi support mentioned for ROCm 2.10 but we hope that will still be coming with ROCm 3.0, especially considering the Radeon Pro W5700 graphics card being RDNA/Navi based.

Download links and the list of ROCm 2.10 changes in full can be found via the RadeonOpenCompute GitHub. While the sources are available for deploying elsewhere, ROCm 2.10 binaries are available for CentOS/RHEL, Ubuntu, and SUSE Enterprise Linux.