Key Takeaways
1. ROCm is AMD’s open-source alternative to Nvidia’s CUDA, supporting HPC and datacenter GPUs.
2. ROCm 7.2 adds support for the latest Ryzen AI 400 APUs on both Windows and Linux platforms.
3. Version 7.1 demonstrated significant performance improvements, achieving up to 5 times better performance compared to version 6.4.4.
4. ROCm 7.2 enables enhanced performance for various AI models, including a 5.4 times increase in video generation tasks.
5. A ComfyUI build with ROCm 7.2 will be available, and users can install it alongside Adrenalin 26.1.1.
AMD is sharing several new updates to its ROCm software stack at CES 2026. ROCm, which originally meant Radeon Open Compute Platform but no longer serves as an acronym, is AMD’s open-source alternative to Nvidia’s proprietary CUDA. This software has been a key component for HPC and datacenter GPUs, such as the Instinct accelerators and Radeon Pro.
Overview of ROCm Functionality
ROCm consists of runtimes, compilers, tools, and libraries that allow for GPU kernel-level programming tailored for high-level end-user applications. It is utilized in training AI models and performing inference, as well as handling HPC tasks like scientific simulations and data analysis. Additionally, it supports general-purpose GPU (GPGPU) tasks that do not require graphics, all utilizing AMD hardware.
New Features in ROCm 7.2
The recent ROCm 7.1 version has shown remarkable performance in text-to-video generation models like LTX Video 0.9 and Wan 2.2, managing to meet the prompt requirements in merely 3 minutes on Ryzen AI Max+ APUs and just 45 seconds on Radeon AI Pro GPUs.
With the release of version 7.2, ROCm now accommodates the latest Ryzen AI 400 APUs for both Windows and Linux platforms. A ComfyUI build featuring the integrated ROCm 7.2 will be accessible from ComfyUI.org. Also, users can install this software stack alongside Adrenalin 26.1.1.
Performance Highlights
Initially, ROCm introduced Windows support with version 7.1, which also delivered a notable performance increase of 5 times compared to version 6.4.4. For Ryzen Max APUs, AMD boasts a 2.6 times boost in performance for Stable Diffusion XL (SDXL) and 5.2 times for Flux S with the new v7 in comparison to the earlier versions.
ROCm 7 can also harness the capabilities of Radeon AI Pro R9700 GPUs, achieving a performance increase of 5.4 times in the Wan 14b video generation model.