gpt-oss-20b

on RTX 5090

318

tok/sdecode

RTX 5090 (32GB) + AMD Ryzen 7 9850X3D 8-Core Processor (8c) + 30GB

rank in tier

3/165RTX 5090 runstop 2%

best workload

chat-short

where the rig flew

slowest workload

chat-long302 tok/s

where the rig struggled

backend

llama.cppllama.cpp is the universal backend — broad model support, modest speed ceiling.

faster than

Qwen2.5-0.5B-Instruct-4bit on M3 Pro287 tok/s
stable-code-instruct-3b-4bit on M3 Ultra192 tok/s
DeepSeek-Coder-V2-Lite-Instruct-4bit on M3 Ultra168 tok/s

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.