ASUS PX13 Strix Halo runs 120B LLMs
Benchmarks show the 256-bit memory bus on AMD's new APU enables 120B parameter models to run at 15 tok/s on a 13-inch chassis. A significant leap for local AI workstation portability that finally bridges the gap with multi-GPU desktops.
Strix Halo is the hardware breakthrough local AI enthusiasts have waited for, bringing desktop-class unified memory bandwidth to a portable form factor. The 128GB LPDDR5x-8000 configuration provides approximately 215 GB/s bandwidth, while Vulkan (RADV) emerges as the superior backend for consumer Linux over ROCm. Performance scaling at 40W "Power Saver" mode keeps the 13-inch chassis quiet during inference, and the amd-strix-halo-toolboxes project simplifies the complex dependency chain via containerization.
DISCOVERED
4h ago
2026-04-12
PUBLISHED
6h ago
2026-04-12
RELEVANCE
AUTHOR
Willing-Toe1942