Overnight the AMD ROCm 7.0 release tags began appearing within the public Git repositories. Now AMD ROCm 7.0 is officially released as a very significant step forward for AMD’s open-source GPU compute stack for better competing against NVIDIA’s CUDA ecosystem.
ROCm 7.0.0 is officially out and all of the ROCm 7.0 documentation has also been published along with the binaries being available via the AMD repository.
The key highlights of ROCm 7.0 include:
– AMD Instinct MI350X and Instinct MI355X are now officially supported.
– Ubuntu 24.04.3 LTS and Rocky Linux 9 with Linux 5.14 are now officially supported combinations.
– ROCm 7.0 supports KVM Passthrough for MI350X and MI355X GPUs.
– ROCm 7.0 supports PyTorch 2.7, integrated Fused Rope kernels in APEX, Python C++ extension support with amdclang++, TensorFlow 2.19.1 support, ONNX 1.22 support, Triton 3.3, and support for JAX 0.6.0.
– ROCm now supports Ray as a unified framework for scaling AI and Python applications.
– Official support for Llama.cpp.
– The AMD GPU kernel driver code is now distributed separately from the ROCm stack.
– HIP Runtime support for Open Compute Project FP4, FP6, and FP8 data types and APIs.
– Support for the AMD Next-Gen Fortran Compiler (llvm-clang / new-flang).
– ROCgdb debugger enhancements.
– The ROCm Compute Profiler brings an interactive command line with TUI.
Downloads and more details on the ROCm 7.0 release via rocm.docs.amd.com. Details on all of the ROCm 7.0 specific changes can also be found now via this GitHub page. I will be working on some AMD ROCm 7.0 benchmarks shortly.
