Sam Witteveen · 2026-07-31 · notable
Sam Witteveen — 'AMD Ryzen AI Halo'
Sam Witteveen takes AMD's Ryzen AI Halo developer kit — the $3,999 mini-workstation built around the Ryzen AI Max+ 395 and 128 GB of unified memory — through hands-on local-LLM tests aimed at models that don't fit on a single consumer GPU.
Sam Witteveen puts AMD's $3,999 Ryzen AI Halo developer platform to work as a local-LLM box.
What is it?
Sam Witteveen's new video covers AMD's Ryzen AI Halo — the $3,999 pocket-sized developer platform AMD started shipping to reviewers in July 2026, built on the Zen 5 Ryzen AI Max+ 395 processor, an XDNA 2 NPU, a 40-CU Radeon 8060S iGPU, and 128 GB of LPDDR5x-8000 unified memory that the CPU and GPU share.
How does it work?
The Ryzen AI Halo's unified 128 GB memory pool lets a developer load large language models that would not fit on any single consumer discrete GPU. AMD's documented figures for the box put a 20-billion-parameter model at around 20 tokens/second on the XDNA 2 NPU at about 35 watts, with ROCm and a pre-validated software stack handling the deployment on Windows and Linux.
Why does it matter?
For local-LLM developers, Ryzen AI Halo is the first mainstream 128 GB unified-memory workstation aimed squarely at LLM inference rather than gaming — a direct competitor to Apple's high-memory M-series machines and NVIDIA-based workstation towers, and Sam Witteveen's channel is one of the main places local-LLM builders decide which hardware to actually buy.
Who is it for?
local-LLM developers, on-prem AI engineers, indie coding-agent builders
Try it
Watch the video, then check AMD's Ryzen AI Halo product page for developer-kit availability