AMD Ryzen AI Max+ "Strix Halo" Performance with ROCm 7.0 (www.phoronix.com)

Although AMD’s ROCm 7.0 release did not list the new Ryzen AI Max “Strix Halo” SoCs among supported GPUs, hands-on testing shows the stack runs cleanly on that hardware. Using a Framework Desktop with an AMD Ryzen AI Max+ 395 (Radeon 8060S), Ubuntu 24.04.3 LTS, Linux 6.14, the AMDGPU DKMS driver and ROCm 7.0, the compute stack worked as expected. Benchmarks included vLLM and Llama.cpp (with explicit comparisons between Vulkan and ROCm back-ends) and Mixbench for HIP vs OpenCL, demonstrating that common AI inference toolchains operate on the Strix Halo despite its absence from the official compatibility list. For the AI/ML community this is meaningful: it indicates practical, community-driven ROCm compatibility on Ryzen AI Max hardware and lowers the barrier to using these desktop SoCs for on-prem or edge model inference and experimentation. The technical takeaway is that a standard Linux kernel (6.14) + AMDGPU DKMS + ROCm 7.0 is sufficient to run popular inference stacks (vLLM, Llama.cpp) and cross-API tests (HIP/OpenCL) on the 8060S, while Vulkan remains an important performance contender for Llama.cpp workloads. In short, Strix Halo users can feasibly leverage ROCm 7.0 today, even as official support catches up.