Sam Witteveen · 2026-07-21 · notable
Sam Witteveen: 'AMD Ryzen AI Halo - 100% Local AI'
Sam Witteveen runs LM Studio, ComfyUI, Hermes-Agent and Unsloth locally on AMD's new Ryzen AI Halo developer platform, and walks through what a strix-halo mini-workstation actually feels like for open-model workflows.
A hands-on walkthrough of running frontier open models on AMD's Ryzen AI Halo mini-workstation.
What is it?
Sam Witteveen's July 21 video 'AMD Ryzen AI Halo - 100% Local AI' is a sponsored hands-on with AMD's new Ryzen AI Halo developer platform — the strix-halo class of mini-workstations aimed at running LLMs, image and video generation, and agentic workflows entirely on-device. Sam is an AI Engineer who covers small model releases and on-device inference on his channel.
How does it work?
The video moves through four workloads on the same box: LM Studio serving a quantized LLM with vulkan llama.cpp, ComfyUI rendering images with a Krea-style pipeline, an agent loop built on Hermes-Agent, and an Unsloth fine-tuning run. Sam shows tokens-per-second numbers on the Ryzen AI Halo APU and calls out where the unified memory helps vs. where a discrete GPU still wins.
Why does it matter?
Ryzen AI Halo is AMD's answer to Apple Silicon and NVIDIA DGX Spark for the local-AI dev crowd — a single Windows or Linux box that runs mid-size open models without cloud tokens. Sam's channel is a common first-look source for AI engineers picking hardware, and the video is one of the earlier independent walkthroughs of what these dev-kit machines feel like end-to-end.
Who is it for?
AI engineers evaluating local-AI hardware and developers running open models off-cloud
Try it
Watch: youtube.com/watch?v=ogVSqcVxv28