Llama-3.3-70B-Instruct

on A40

13.0

tok/sdecode

A40 (45GB) + Intel(R) Xeon(R) Gold 6342 CPU @ 2.80GHz (48c) + 504GB

rank in tier

1/1A40 runs

best workload

chat-short

where the rig flew

slowest workload

chat-long12.7 tok/s

where the rig struggled

backend

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

faster than

Qwen3-32B-4bit on M3 Pro7.2 tok/s
Qwen2.5-32B-Instruct-4bit on M3 Pro7.1 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.