eGPU Fail on Strix Halo: Llama-server Cross-GPU Config Log
parepeg · reddit · 2026-08-19
The author details their failed attempts to run an eGPU setup on a Framework Strix Halo for local LLM inference, highlighting specific hardware compatibility and PCIe bandwidth issues.
- Hardware: ADT-link UT3G failed to connect; F9G with OCuLink worked but dropped to PCIe Gen1, negating speed benefits.
- Software: llama-server successfully managed hybrid internal AMD and external NVIDIA GPUs. The specific command -ngl 999 -ts 6,128 --main-gpu 1 -dev Vulkan0,Vulkan1 enabled tensor splitting across devices.
More from Embodied
- Human efficiency vs. robots: Walking a mile burns half a banana — noahsolomon · 2026-08-19
- Cybercab revolution imminent, launch event likely next week — mitchdeg · 2026-08-19
- Open Source Personal AI Computer: Local Compute Blueprint — tom_doerr · 2026-08-19
- Embodied-Navigator: Point, Think, Memorize, and Align for Navigation — OmniAI-ZJU · 2026-08-19
- Podcast: OpenRoboto on training humanoid robot AI on decentralized networks — markjeffrey · 2026-08-19
- Stanford BPP: Robots Learn New Tasks from One Demo — SongShuran · 2026-08-19