inference/research
verifiedhigh · 0.90updated 2026-10-03

llama.cpp issue #25207: unexpected massive performance drop with Vulkan Flash Attention

forum

Fields

id
sources/gotcha-llamacpp-issue-25207-vulkan-fa-slowdown
URL
https://github.com/ggml-org/llama.cpp/issues/25207
Publisher
llama.cpp GitHub issues
Kind
forum
Published
2026-07-01
Accessed
2026-10-03
notes
llama-bench tables with FA on/off for two models on Strix Halo (RADV, warp size 64). Labelled bug-unconfirmed upstream.

Notes

llama-bench tables with FA on/off for two models on Strix Halo (RADV, warp size 64). Labelled bug-unconfirmed upstream.

Cross-references

This record references no other records.