AMD Shows YOLO on Radeon GPUs and ROCm at Ultralytics YOLO Vision 2026
AMD, title sponsor of Ultralytics YOLO Vision 2026 (YV26), used the event to show how Ultralytics YOLO models can be trained and deployed on AMD GPUs with PyTorch and ROCm. The company combined a main-stage talk, two hands-on workshops and demos on the exhibition floor.
- AMD was title sponsor of Ultralytics YOLO Vision 2026, which opened on September 13.
- Alex Zhang of AMD presented how to run Ultralytics YOLO on AMD GPUs with PyTorch and ROCm.
- Two workshops covered YOLO on AMD Radeon, including a YOLO26x pipeline using ONNX, AMD MIGraphX and FP16 inference.
- Ultralytics says ROCm is a supported path for training and deploying YOLO models.
- The AMD AI Developer Program offers 100 cloud credits (100 hours) in AMD Developer Cloud.
A ROCm talk on the main stage
The centerpiece of AMD's day was a session by Alex Zhang, an AI Software Product Engineer and Senior Member of Technical Staff at AMD, who leads the AMD ROCm Ambassador Program. According to Ultralytics, his talk covered what it takes to run Ultralytics YOLO models on AMD GPUs using PyTorch and ROCm, from early experimentation and training through to real-world inference.
The session fit the conference's broader theme, "Open vision, built for the real world." Ultralytics describes Zhang's argument as treating open compute as part of what makes vision AI deployable: teams should not have to lock themselves into one vendor's stack just to run Ultralytics YOLO in production.
YV26 opened on September 13. Ultralytics Founder and CEO Glenn Jocher set the tone with the idea that vision AI creates the most value when it is open, accessible, and shaped by the people deploying it in the field.
Workshops on real AMD hardware
AMD also ran dedicated hands-on workshops that extended the talk. Small groups were walked through fine-tuning, deploying and accelerating Ultralytics YOLO on AMD Radeon, including building an end-to-end computer vision workflow on a single AMD Radeon GPU.
A second workshop was titled "Ultralytics YOLO26 on AMD Radeon: From an Ultralytics Checkpoint to an End-to-End Video AI Pipeline." It showed how YOLO26x is exported to ONNX and compiled with AMD MIGraphX for FP16 inference. The pipeline was then connected to:
- hardware video decode
- OpenCV HIP preprocessing
- GPU NMS
- hardware encode
Both sessions ran on real AMD hardware, so attendees could work through the workflows themselves rather than only watch them.
AMD AI Developer Program
The article also points readers to the AMD AI Developer Program. According to the post, registration includes:
- 100 cloud credits (100 hours) in AMD Developer Cloud credits
- early access to exclusive webinars and AMD events
- monthly sweepstakes for an AMD Radeon GPU or Ryzen AI PC
- one month of complimentary DeepLearning.AI membership
- a private Discord channel with AMD engineers and technical peers
- a monthly newsletter with curated AI insight and the AMD product roadmap
What it means for YOLO users
AMD's sessions took place on the same day Ultralytics announced the upcoming Ultralytics YOLO27 and rolled out AutoTrain, Agents, and Monitoring on Ultralytics Platform. Ultralytics argues that the AMD talk and workshops show the "built for the real world" theme is not only about new model architectures, but also about making sure models run on the compute teams already have.
For developers, the message Ultralytics took from the event is that ROCm is a supported path for training and deploying Ultralytics YOLO models, and that AMD is investing in developer-facing effort. Teams with AMD Radeon hardware can therefore consider it for YOLO workflows, from fine-tuning to video pipelines.
Ultralytics: Open Compute, Open Vision: AMD at Ultralytics YOLO Vision 2026
