Skip to content
llm-speed
← raw run
wrapped·Jul 01, 2026
/r/r_q-it_4s9z4p
your run

gemma-4-31B-it-Q4_K_M.gguf

on RTX 5090
67.2
tok/sdecode
RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB
rank in tier
148/165RTX 5090 runstop 90%
best workload
chat-short
where the rig flew
slowest workload
—
single-workload run
backend
llama.cppllama.cpp is the universal backend — broad model support, modest speed ceiling.
faster than
  • starcoder2-15b-4bit on M3 Ultra64.2 tok/s
  • Codestral-22B-v0.1-4bit on M3 Ultra47.5 tok/s
  • Qwen2.5-32B-Instruct-4bit on M3 Ultra34.6 tok/s
share
67.2 tok/s on RTX 5090 running gemma-4-31B-it-Q4_K_M.gguf — top 90% of RTX 5090 runs. Signed benchmark via llm-speed.
https://llm-speed.com/wrapped/r_q-it_4s9z4p
X / TwitterBlueskyMastodonRedditraw run

This is a shareable view of one signed llm-speed suite-suite-v1 run. Numbers are not edited. The raw record has every workload, the public key, and the fingerprint hash.