
AMD is gearing up for the launch of its ROCm 7 software stack, aiming to disrupt NVIDIA’s dominance in the AI landscape, particularly the tightly-knit CUDA ecosystem that has long confined developers to NVIDIA’s hardware.
AMD’s ROCm 7 Approaches Release on GitHub
NVIDIA’s supremacy in the artificial intelligence sector is largely attributed to its CUDA software stack, a vital asset for AI developers that is exclusive to its proprietary hardware. This situation has left rivals like AMD with a challenging task. However, during the recent Advancing AI event, AMD unveiled its next-generation ROCm 7.0, which promises enhanced frameworks and innovative algorithms intended to serve as a viable alternative to CUDA. The groundwork for its release has now officially begun.
According to a report from Phoronix, AMD has initiated the release process by adding tags for ROCm 7.0 on GitHub. This step indicates that the launch is imminent, with associated tags such as rocm-7.0.0 in ROCm/hip and rocm-7.0.0 in ROCm/aomp already displayed. This move signals an important milestone for AMD, as the company prepares to roll out what could become a pivotal update for Team Red.

The focus of the revamped ROCm stack is primarily on inferencing workloads. AMD touts an impressive performance enhancement of up to 3.5 times in AI tasks compared to ROCm 6. Notably, the Instinct MI355X now outperforms NVIDIA’s Blackwell B200 by achieving a 30% higher FP8 throughput in the DeepSeek R1 test. The key upgrades in ROCm 7.0 include:
- Cutting-edge Algorithms & Models
- Enhanced Features for AI Scaling
- Support for the MI350 Series
- Cluster Management Enhancements
- Advanced Enterprise Capabilities
While there is currently no set release date for ROCm 7, it is anticipated to debut in the coming weeks, especially as AMD aligns it with new AI hardware announcements.
Leave a Reply