AMD ROCm 5.5 Official: Several RDNA 3 Updates Including Better Radeon RX 7000 Support On Linux

AMD ROCm 5.3.3 Now Available With Optimizations & Bug Fixes 1

AMD has released the new ROCm 5.5 GPU compute stack for the open-source Linux community which adds improved RDNA 3 support.

This update brings changes and better support for the Radeon RX 7000 (RDNA 3) graphics cards. ROCm is AMD's compute stack, similar to what NVIDIA and Intel offer with the CUDA, oneAPI, and Level Zero options. From the README file on the AMD ROCm Platform page on GitHub:

ROCm is AMD's open-source stack for GPU computation. It is a grouping of drivers, development tools, and APIs enabling GPU programming from the low-level kernel to end-user applications.

ROCm is powered by the AMD Heterogeneous-computing Interface for Portability (HIP), an OSS C++ GPU programming environment, and its corresponding runtime. HIP allows ROCm developers to create portable applications by deploying code on various platforms, from dedicated gaming GPUs to exascale HPC clusters. ROCm supports programming models such as OpenMP and OpenCL and includes all the necessary OSS compilers, debuggers, and libraries. ROCm is fully integrated into ML frameworks such as PyTorch and TensorFlow. ROCm can be deployed in many ways, including containers such as Docker, Spack, and your build from source.

via AMD

AMD notes several new additions and fixes to this new ROCm release, including:

HIP enhancements

Enhanced Stack Size limits

HIPCC changes

Separation of HIPCC from HIP

New memory management

New module management

HIP compiler changes

New APIs

New Linux file system hierarchy

New Library changes in ROCm 5.5: hipBLAS (0.53.0 to 0.54.0) hipCUB (2.13.0 to 2.13.1) hipFFT (1.0.10 to 1.0.11) hipSOLVER (1.6.0 to 1.7.0) hipSPARSE (2.3.3 to 2.3.5) RCCL (2.13.4 to 2.15.5) rocALUTION (2.1.3 to 2.1.8) rocBLAS (2.46.0 to 2.47.0) rocFFT (1.0.21 to 1.0.22) rocPRIM (2.12.0 to 2.13.0) rocRAND (2.10.16 to 2.10.17) rocSOLVER (3.20.0 to 3.21.0) rocSPARSE (2.4.0 to 2.5.1) rocThrust (2.17.0) rocWMMA (0.9 to 1.0) Tensile (4.35.0 to 4.36.0)

hipBLAS (0.53.0 to 0.54.0)

hipCUB (2.13.0 to 2.13.1)

hipFFT (1.0.10 to 1.0.11)

hipSOLVER (1.6.0 to 1.7.0)

hipSPARSE (2.3.3 to 2.3.5)

RCCL (2.13.4 to 2.15.5)

rocALUTION (2.1.3 to 2.1.8)

rocBLAS (2.46.0 to 2.47.0)

rocFFT (1.0.21 to 1.0.22)

rocPRIM (2.12.0 to 2.13.0)

rocRAND (2.10.16 to 2.10.17)

rocSOLVER (3.20.0 to 3.21.0)

rocSPARSE (2.4.0 to 2.5.1)

rocThrust (2.17.0)

rocWMMA (0.9 to 1.0)

Tensile (4.35.0 to 4.36.0)

The above is only a small list of changes and fixes. The ROCm 5.5 compute stack is looking to be one of the most significant updates compared to previous releases, especially with enhanced support for AMD's Radeon 7000 series graphics cards and improved support for Radeon PRO W7800 and W7900 series GPUs that will launch soon. Additionally, the new library changes in ROCm 5.5 mention RDNA 3 (GFX11) support, which was missing in previous versions.

Michael Larabel of the Linux Hardware website Phoronix mentions that the improved RDNA 3 support is not fully cited in the update notes as the company focuses "on their professional and workstation graphics card and accelerator support." He notes that the support matrix is meager, with a small gathering of GPUs from architectures GFX9, CDNA, and RDNA.

The release notes for ROCm 5.5, which can be found here, mention an increase to the stack size limit, which initially was at 16K, and has now increased to 128K.

News Sources: Phoronix, AMD ROCm GitHub