Skip to content
llm-speed
← raw run
wrapped·Jul 01, 2026
/r/r_1_xl4zb5-xj
your run

gemma-2-9b-it

on RTX 5090
69.5
tok/sdecode
RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB
rank in tier
115/165RTX 5090 runstop 70%
best workload
chat-short
where the rig flew
slowest workload
agent-trace68.5 tok/s
where the rig struggled
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
69.5 tok/s on RTX 5090 running gemma-2-9b-it — top 70% of RTX 5090 runs. Signed benchmark via llm-speed.
https://llm-speed.com/wrapped/r_1_xl4zb5-xj
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.