Exploring SSD Streaming for DeepSeek and Kimi Models on Strix Halo

lawanda123 · reddit · 2026-07-30

The poster is exploring the feasibility of SSD streaming large models like DeepSeek V4 and Kimi K3 on the AMD Strix Halo APU. The goal isn't fast inference speeds, but rather using a capable model overnight for complex planning and delegating tasks to smaller models during the day. The user is seeking architectural setup advice and is willing to test and report performance numbers.

Original post →

More from Infra

Infra channel →