Skip to content
llm-speed
← raw run
wrapped·Oct 16, 2026
/r/r_m6j1zxwvp-f
your run

Qwen3.6-35B-A3B

on RTX 4090
160.6
tok/sdecode
RTX 4090 (24GB) + AMD EPYC 7642 48-Core Processor (48c) + 252GB
rank in tier
7/31RTX 4090 runstop 23%
best workload
chat-short
where the rig flew
slowest workload
chat-long160.4 tok/s
where the rig struggled
backend
llama.cppllama.cpp is the universal backend — broad model support, modest speed ceiling.
faster than
  • Nemotron-3.5-Bolt-74B-A9B-24GB-ctx32k on RTX 5090157.7 tok/s
  • gemma-2-9b-it on RTX 5090152.7 tok/s
  • gpt-oss-20b-MXFP4-Q4 on M3 Ultra152.7 tok/s
share
160.6 tok/s on RTX 4090 running Qwen3.6-35B-A3B — top 23% of RTX 4090 runs. Signed benchmark via llm-speed.
https://llm-speed.com/wrapped/r_m6j1zxwvp-f
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.