Skip Navigation

InGameLoop relies on readers. We may earn commissions when you purchase through our links. Check Affiliate Disclosure

AMD ROCm 6.0 Source Code Now Available: Supports Instinct MI300, Radeon 7000 GPUs, AI Additions

In a significant move towards bolstering its AI capabilities, AMD has announced the release of...

AMD ROCm 6.0 Source Code Now Available: Supports Instinct MI300, Radeon 7000 GPUs, AI Additions

In a significant move towards bolstering its AI capabilities, AMD has announced the release of the ROCm 6.0 software stack, making the full source code available for download. This latest iteration brings support for the highly anticipated Instinct MI300 and Radeon 7000 GPUs, marking a substantial stride in AMD’s competition with NVIDIA’s CUDA platform.

Unveiled at the company’s Advancing AI event, ROCm 6.0 represents AMD’s commitment to providing developers with an updated framework. The release, initially available on the Linux platform, showcases AMD’s dedication to advancing its ROCm platform as a robust alternative to CUDA.

One of the key highlights of ROCm 6.0 is its expanded support for AMD’s Instinct MI300A and MI300X AI accelerators. These GPUs, showcased at the Advancing AI event, reflect AMD’s continuous efforts to push the boundaries of performance within the ROCm platform. However, it’s worth noting that ROCm 6.0 currently lacks support for the new Instinct GPUs at Red Hat Enterprise Linux 9, a limitation expected to be addressed in subsequent updates.

The ROCm 6.0 software stack introduces several features aimed at optimizing performance and enhancing developer accessibility. Notably, it includes highly optimized attention algorithms and proven collective communications libraries, catering to key AI-supported features. Additionally, ROCm 6.0 boasts FP8 support in PyTorch and hipblasLT, showcasing AMD’s commitment to providing optimized performance for AI workloads.

Developers will find a range of tools and frameworks packaged for high-performance computing (HPC) and AI/machine learning (ML) on the AMD Infinity Hub. The streamlined tools and improved ROCm Docs contribute to a developer-friendly environment, easing the integration of ROCm 6.0 into various projects.

In a move to broaden its compatibility and offer users more flexibility, ROCm 6.0 introduces support for additional libraries. DeepSpeed, ONNX-RT, Jax, and CuPy are among the newly supported libraries, enriching the toolkit available to developers for AI implementation. This expansion not only reflects AMD’s commitment to openness but also enables users to leverage a wider array of resources in their AI development endeavors.

AMD’s ROCm 6.0 doesn’t merely stop at introducing new features; it signifies a milestone in the ongoing competition with NVIDIA. AMD officials proudly claim that ROCm 6.0 has achieved software parity with CUDA in large language model training, a domain where NVIDIA has traditionally held a dominant position. This signals a significant breakthrough for AMD, showcasing its dedication to closing the gap in AI-focused software resources.

For users keen on exploring the capabilities of ROCm 6.0, the full source code is now available for download, accessible to the public. The transparency of open-source development aligns with AMD’s commitment to fostering collaboration and innovation within the developer community. Additionally, users can find a comprehensive list of supported Radeon 7000 and Instinct GPUs, providing clarity on the hardware compatibility of ROCm 6.0.

As AMD continues to make strides in the AI landscape, the availability of ROCm 6.0 with support for Instinct MI300, Radeon 7000 GPUs, and additional AI libraries underscores the company’s dedication to empowering developers and offering competitive solutions in the evolving world of artificial intelligence.

AMD Ryzen 7000G Desktop and Ryzen 8040 Laptop APUs May Have Been Uncovered

AMD’s next gen Threadripper Pro 7000-series chips come with up to 96 cores and I long for consumer version