← All setups

Step-by-step build

Strix Halo (Ryzen AI Max+ 395, 128GB) on Ubuntu with ROCm + llama.cpp: full guide

A complete Ubuntu guide for Ryzen AI Max+ 395 machines (Framework Desktop, GMKtec EVO-X2): BIOS, kernel, ROCm via prebuilt toolbox containers, and llama.cpp with full offload, making ~115-120GB available for models.

Solo2-10 peopleWriting assistantCodingDocument search / chat with files$3k+Sensitivity: High (client/confidential)

Hardware

ItemApprox. price (as of 2026-10-16)
On sale −20% Minisforum MS-S1 MAX (Ryzen AI Max+ 395, 128GB, 2TB) (Minisforum Store, related current offer)$3,879.00 $4,849.00
On sale −6% MSI GeForce RTX 5060 Ti 16GB Ventus 3X OC (Newegg, related current offer)$579.99 $614.99
AMD Ryzen AI Max+ 395 system with 128GB (e.g. Framework Desktop, GMKtec EVO-X2)Not stated in source
Framework Desktop DIY (Ryzen AI Max+ 395, 128GB) (Framework, related current price)$3,449.00

Source prices are as stated by the source. Retailer offers: prices checked 2026-10-16, may change. Some links may be affiliate links; we may earn a commission at no extra cost to you. This never changes what we list.

Software stack

  • Ubuntu 24.04+ (tested on 25.10)
  • ROCm (kyuz0/amd-strix-halo-toolboxes containers)
  • llama.cpp

Models used

  • Llama-2-7B
  • 70B-class models (Q4)

Good at

  • Llama-2-7B ~50-52 t/s per the guide
  • 70B models (Q4) ~15-20 t/s and fit entirely in memory

Weak at

  • Requires BIOS changes and kernel parameters
  • Needs --no-mmap and careful backend setup

Rough total cost

Not stated in source.

Getting started

  1. Configure BIOS memory settings.
  2. Install Ubuntu 24.04+ with a recent kernel and set kernel parameters.
  3. Create a ROCm toolbox container (kyuz0 image).
  4. Run llama.cpp with -ngl 99 and --no-mmap.

Source & credit

Gygeek/Framework-strix-halo-llm-setup — GitHub

Summarized from the source; quotes are the source author's words. Check the original before buying.

Get new tested setups monthly

One email a month with new real-world builds. No spam.