MyColo measured local AI inference

Share card

bundle 1 · measured result · 27B

40.0 tok/s

hardware
NVIDIA GeForce RTX 5090
model
qwen3.8-27b
engine / quant / ctx
llama.cpp · Q4_K_M · ctx 65536
workload
—
trust
signed · lone · test · 2026-08-26 22:09:14

Throughput comparison

40.0 tok/s · baseline n/a · n=0

Not enough comparable prior runs yet (same handle · hardware · track · recipe).

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

Retracted — typed practice score — not a bench. Kept in the archive.

hash
91ae010b8679cc21b48981cb372cd07d2a835581929eb406ef9f9602bb6e0303

Hardware class

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

Sign-offs

hardware receipt · JSON · measure