AMD's ROCm 6.0 software stack and the full source code are now available to download & come with Instinct MI300 & Radeon 7000 GPU support.
AMD's ROCm 6.0 Software Stack Now Brings Support For Additional AI Libraries, Unlocks FP8 Performance At PyTorch, Adds Instinct MI300 & Radeon 7000 GPU Support
In the company's Advancing AI event, AMD revealed plans for enhancing the company's ROCm platform through an "updated" framework, and in the initial stages, has been released on Linux.
ROCm 6.0 is AMD's attempt to give tough competition to NVIDIA's CUDA. Not only has the new software stack upgraded to support AMD's latest Instinct MI300 GPUs, but it has received support for powerful new features which is a step towards making it a viable alternative to the CUDA platform.
Starting with what ROCm 6.0 brings, it comes with support for AMD's Instinct MI300A/MIX300X AI accelerators, which were yet again shown off in the Advancing AI event. The software stack has improved a lot over time, and the performance results shared by AMD of its Instinct GPUs are boosted by the improvements made within the ROCm platform, especially with the new version. However, Phoronix discloses that ROCm 6.0 currently doesn't support the new Instinct GPUs at Red Hat Enterprise Linux 9, but eventually, it would gain support later on.
Some of the main features of the ROCm 6.0 stack include:
- Expanded support and access: support for AMD Instinct MI300A and MI300X
- Key AI-supported features: highly optimized attention algorithms, and proven collective communications libraries
- Optimized performance: FP8 support in PyTorch and hipblasLT
- Enabling developer success: prepackaged HPC and AI/ML frameworks ready for download on the AMD Infinity Hub; streamlined and improved tools
- Easy access to resources: new and improved ROCm Docs and a new home for all other developer resources on the AMD ROCm Developer Hub
Moving on to the more interesting part, ROCm 6.0 has gained support for additional libraries, such as DeepSpeed, ONNX-RT, Jax, and CuPy that provide consumers with more flexibility when it comes to AI implementation. Moreover, ROCm 6.0 will now support FP8 performance at PyTorch and hipblasLT, which will not only provide enhanced performance, but it would cater to actually contributing towards AI workloads with your respective RDNA/CDNA architecture.
AMD officials have stated that ROCm 6.0 has reached software parity with CUDA on large language model training, suggesting that Team Red has been able to close the gap when it comes to AI-focused software resources, an area where NVIDIA has been dominant in recent years. With support for ROCm 6.0 in Linux, users can now download the full source code which is publicly available here and you can also find the full list of supported Radeon 7000 and Instinct GPUs in the list below:
About the author: Muhammad Zuhair is a hardware and technology reporter for Wccftech, specializing in the semiconductor industry and the complex interplay between technology, manufacturing, and geopolitics. His coverage focuses on the corporate strategies and technological roadmaps of industry giants like TSMC, NVIDIA, Samsung, and Intel. Zuhair's expertise lies in deconstructing complex topics such as fabrication nodes (e.g., 2nm process), the economic impact of policies like the CHIPS Act, and the strategic development of AI infrastructure from NVIDIA, AMD and Intel.
Follow Wccftech on Google to get more of our news coverage in your feeds.
