Skip to content
llm-speed
← raw run
wrapped·Sep 06, 2026
/r/r_8q-uf-rq-gp
your run

gemma-4-12b-it-qat

on RTX 5090
134.2
tok/sdecode
RTX 5090 (32GB)
rank in tier
56/181RTX 5090 runstop 31%
best workload
long-context-decay
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
  • r1 on RTX 4090133.8 tok/s
  • Flint-40B-A2.5B-UD-ctx8k on RTX 3090130.8 tok/s
  • Llama-3.1-8B-Instruct-4bit on M3 Ultra130.2 tok/s
share
134.2 tok/s on RTX 5090 running gemma-4-12b-it-qat — top 31% of RTX 5090 runs. Signed benchmark via llm-speed.
https://llm-speed.com/wrapped/r_8q-uf-rq-gp
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.