llama.cpp Discussion #15021 — Performance of llama.cpp on AMD ROCm/HIP (community scoreboard)

Curated community scoreboard for the llama.cpp ROCm/HIP backend: llama-2-7b.Q4_0, pp512/tg128, build commit + reporter per row. Covers Instinct MI50-MI300X and Radeon consumer/workstation. Re-read 2026-10-17: filed the rows the earlier pass skipped — Instinct MI25 (both halves), RX Vega 64 FA, Radeon 880M (both halves), and the FA rows for the AI Max+ 395 and the Radeon 8060S.