llama.cpp Discussion #10879 — Performance of llama.cpp with Vulkan (community scoreboard)

Vulkan-backend sibling of the #15013 CUDA and #15021 ROCm scoreboards: llama-2-7b.Q4_0 pp512/tg128, maintainer-curated rows with build commit + reporter. Re-read 2026-10-17: the FA-enabled table (Llama 2 7B Q4_0, -fa 1) was transcribed for the chips that have an accelerator record and did not already have an FA pair.