AMD has started working on releasing the ROCm 7 software stack, which was being hyped up as a way to break NVIDIA's CUDA 'lock-in' ecosystem.
AMD's ROCm 7 Sees Addition Into GitHub, With Launch Expected To Occur In Upcoming Weeks
One of the major reasons NVIDIA is so dominant in the AI industry is the presence of the CUDA software stack. This is the go-to suite for AI developers, and since it's exclusive to native hardware, competitors like AMD have little room to play with. However, at the Advancing AI event, AMD announced its next-gen ROCm 7.0 software stack, which was claimed to bring in enhanced frameworks, and new algorithms to position the compute ecosystem to be an alternative to NVIDIA's CUDA, and now, efforts to release it has officially began.
According to Phoronix, AMD has put up the release tags for ROCm 7.0 at GitHub, which means that AMD is prepping for a launch soon. The tags include rocm-7.0.0 in ROCm/hip and rocm-7.0.0 in ROCm/aomp, with HIP, AOMP, and ROCm Libraries having their 7.0.0 tags up for display. This confirms that a release is being planned soon, and it could pan out as a major launch for Team Red. At the Advancing AI event, Team Red disclosed huge developments towards ROCm 7, which we'll talk ahead.
The firm claimed that the biggest focus of the upgraded compute stack was towards inferencing workloads, and AMD has brought in up to 3.5x performance uplifts in AI workloads versus ROCm 6. More importantly, you are looking at 'accelerated training performance', with the Instinct MI355X running faster than NVIDIA's Blackwell B200, achieving a 30% higher FP8 throughput performance in DeepSeek R1. Here are the major upgrades:
- Latest Algorithms & Models
- Advanced Features for Scaling AI
- MI350 series support
- Cluster Management
- Enterprise Capabilities
For now, there isn't a defined release date for ROCm 7, but based on recent developments, the software stack is expected to launch in the upcoming weeks, depending on how AMD positions it alongside AI hardware releases.
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.
