Ultralytics YOLO on AMD GPUs: Train, Run, and Deploy With ROCm
AMD, Monday, October 26th, 2026
Ultralytics YOLO now officially supports AMD GPUs, with ROCm training and inference and MIGraphX for exported ONNX models.
The AMD ROCm blog announces official Ultralytics YOLO support on AMD GPUs, letting developers train and run YOLO26 models with the same Ultralytics API.
PyTorch on ROCm handles training and inference, while AMD MIGraphX runs exported ONNX models.
The integration was tested on Instinct, Radeon, and Ryzen AI hardware, with training verified on Instinct MI350X and Radeon AI PRO R9700, and installation supported on Linux x86_64 via Docker or pip.
The guide walks through training YOLO26n on coco8, running inference, and exporting to ONNX, and reports MIGraphX cut latency by up to 67% versus PyTorch on one Ryzen AI processor.