MyColo measured local AI inference

Share card

bundle 15 · measured result · 27B

147.8 tok/s

hardware
NVIDIA GeForce RTX 5090
model
qwen3.8-27b · 22130167c4c20e20…
engine / quant / ctx
ollama · Q4_K_M · ctx 4096
workload
wl-v1-qwen38-27b-ollama
trust
measured · lone · test · 2026-09-21 00:14:09

Throughput comparison

147.8 tok/s · baseline 142.8 · +3.5% · n=10

27B · qwen3.8-27b · your prior signed/measured runs (same recipe)

Same handle, hardware class, size track, and recipe version. Current run excluded.

hash
f444579903388c4edb6e6387f9889b45dced321c4a18aea63aea3c9c7c9f713d

Hardware class

{
  "cpu": "AMD",
  "driver": "580.126.18",
  "gpu_or_soc": "NVIDIA GeForce RTX 5090",
  "os_name": "Linux",
  "stack": "cuda",
  "vram_or_unified_mib": 32607
}

Sign-offs

hardware receipt · JSON · measure