DeepSeek V4 Flash 284B (13B active)
on 3× AMD MI50 32GB · llama.cpp
Sep 7, 2026
Use cases
coding
Summary
User reports DeepSeek V4 Flash 0731 at 15-16 t/s generation and 105-110 t/s prompt processing on 3x AMD MI50 32GB GPUs, 96GB of VRAM in total.
Setup is llama-server (llama.cpp) with the UD-IQ2_M quant. The 90.9GB model fits fully in VRAM.
The user tested it with a Rubik's cube HTML coding task and is impressed it runs locally.