phi-4

on RTX 5090

139.1

tok/sdecode

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

rank in tier

49/183RTX 5090 runstop 27%

best workload

chat-short

where the rig flew

slowest workload

agent-trace124.8 tok/s

where the rig struggled

backend

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

faster than

Qwen2.5-Coder-7B-Instruct-4bit on M3 Ultra138.6 tok/s
llama3.1 on RTX 3090136.2 tok/s
glm-4.7-flash on RTX 3090134.6 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.