At Computex, AMD made waves with one of its most underrated unveilings: ROCm support for Strix Halo and RDNA 4 GPUs. This innovation is poised to transform AI usage at a consumer level, marking a significant shift in the tech landscape.
AMD has long been recognized for its hardware prowess, but its commitment to fostering a comprehensive ecosystem for consumers has reached new heights. Previously, there were uncertainties about ROCm’s compatibility with RDNA 4 GPUs. However, AMD surprised enthusiasts by announcing ROCm 6.4.1, extending robust support to RX 9070 series GPUs and Strix Halo APUs. This makes ROCm one of the most formidable software stacks available for everyday users.
By positioning ROCm as a viable alternative to NVIDIA’s CUDA, AMD is setting the stage for impressive benefits. With support for multiple AI libraries, consumers using Strix Halo APUs can tap into the power of the onboard XDNA 2 AI engine. This upgrade unlocks AI performance for 40 RDNA 3.5 Compute Units and 16 Zen 5 cores with AVX512, substantially boosting capabilities in both inference and training tasks.
For RDNA 4 GPUs, ROCm now empowers the use of onboard Compute Units and AI accelerators across various applications. RX 9000 series GPU users can run advanced frameworks like PyTorch 2.5/2.6 and Megatron-LM, allowing seamless execution of large language models and even local instances of Stable Diffusion. This shift from professional to consumer AI applications could spark a revolution in edge AI, should AMD continue to enhance its ROCm software stack.
An exciting development is Microsoft’s recent decision to open-source WSL (Windows Subsystem for Linux), which aligns perfectly with the new ROCm version. This collaboration ensures smooth integration of ROCm’s AI tools into Windows environments, previously concentrated solely on Linux. This compatibility enables developers to use a familiar setting for expanding the ROCm ecosystem, likely accelerating its adoption.
AMD also plans to broaden ROCm’s reach across various Linux distributions, including OpenSuSE and eventually Ubuntu by mid-2025. This move is a strategic step in AMD’s quest to challenge NVIDIA’s CUDA ecosystem, focusing keenly on the consumer market. With a strong hardware foundation, AMD is well-positioned to capture this audience, promising exciting developments ahead.
