AMD brings native PyTorch support to Windows with the ROCm 6.4.4 preview, unlocking AI development on consumer Radeon GPUs and Ryzen AI APUs. The rollout honors a commitment made at Computex 2025 to deliver a true cross-platform, developer-first stack, and it arrives with matching support on Linux.
What’s new in ROCm 6.4.4
– Native PyTorch support for Windows and Linux in a preview build
– Compatibility with the latest Radeon RX 9000 series (RDNA 4) and RX 7000 series (RDNA 3) GPUs
– Support for Ryzen AI 300 “Strix” and Ryzen AI Max “Strix Halo” APUs
– A foundation for ongoing performance tuning and expanded feature coverage
In practical terms, developers can now run and accelerate AI models directly on Windows PCs powered by Radeon RX 7000 and RX 9000 graphics, as well as the newest Ryzen AI APUs. That includes popular enthusiast cards like the Radeon RX 7900 XTX and RX 7900 XT, along with upcoming RX 9000-class GPUs, and professional-focused options such as Radeon PRO W7900. On the mobile and compact PC side, Ryzen AI 300 and Ryzen AI Max platforms get first-class treatment, making local inference and fine-tuning more accessible on everyday machines.
Why it matters
– Windows-first AI workflows are now viable on AMD consumer hardware
– Easier local testing, prototyping, and fine-tuning of models without switching OSes
– Broader hardware choice for AI researchers, app developers, and power users
– Cross-platform parity helps teams target Windows and Linux with one stack
Preview today, more to come
This is a preview release, designed to get tools into developers’ hands quickly and gather feedback. Expect rapid updates as AMD expands feature coverage and dials in performance on Windows. On the enterprise side, the company is already pushing ahead with ROCm 7 to drive AI and productivity workloads on Instinct accelerators and EPYC platforms.
Getting started
– Confirm your hardware is supported: Radeon RX 7000 or RX 9000 series GPU, or a Ryzen AI 300/Ryzen AI Max APU
– Install the ROCm 6.4.4 preview and the corresponding PyTorch build for Windows
– Follow AMD’s documentation for setup steps, environment configuration, and known limitations
– Validate that PyTorch recognizes your AMD device and begin running models locally
The bottom line
ROCm 6.4.4 removes a major barrier for AI developers on Windows by delivering native PyTorch support across modern Radeon GPUs and Ryzen AI APUs. It’s a meaningful step toward a unified, cross-platform AMD AI ecosystem—and a clear win for developers who want high-performance model training and inference on the systems they already use.
