Skip to content
llm-speed
wrapped·Nov 12, 2026
/r/r_-93wbtb83mc
your run

Qwen3.8-27B

25.8
tok/sdecode
RTX 5090 (32GB)
rank in tier
204/208RTX 5090 runstop 98%
best workload
concurrent-decode
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
  • llama3.2 on AMD64 Family 23 Model 104 Stepping 1, AuthenticAMD23.9 tok/s
  • Muse-Glimmer-30B-ctx8k on RTX A400023.3 tok/s
  • gemma-2-9b-it-4bit on M3 Pro22.0 tok/s
share
25.8 tok/s on RTX 5090 running Qwen3.8-27B — top 98% of RTX 5090 runs. Signed benchmark via llm-speed.
https://llm-speed.com/wrapped/r_-93wbtb83mc

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.