geerlingguy/ai-benchmarks issue #36 — RTX A4000 llama.cpp Vulkan llama-bench

Fields

id

sources/geerling-ai-benchmarks-issue-36-rtxa4000

URL

https://github.com/geerlingguy/ai-benchmarks/issues/36

Publisher

Jeff Geerling

Kind

benchmark

Published

2025-11-28

Accessed

2026-10-17

notes

Same two-host protocol, Vulkan ngl 99. The comment that contains the tables does not print the llama.cpp build hash. gpt-oss-20b completed the full pp512/pp4096/tg128/pp4096+tg128 set on both hosts. Power is whole-system wall draw.

Notes

Same two-host protocol, Vulkan ngl 99. The comment that contains the tables does not print the llama.cpp build hash. gpt-oss-20b completed the full pp512/pp4096/tg128/pp4096+tg128 set on both hosts. Power is whole-system wall draw.

Cross-references

This record references no other records.