#2383 bench: RTX 5080, Ryzen 7 9800X3D, Qwen3.8-Flash-Next IQ3_XXS at 131K, engine 0.1.43.1 (Windows)

说明

Results-only community benchmark, no engine changes: adds bench/results/2026-10-13-community-rtx-5080-9800x3d-iq3-xxs/ (README.md, results.json, engine-log-excerpt.txt), written by tools/strata_bench.py (suite 1, 3 runs per test, thinking off).

Hardware: RTX 5080 16 GB, Ryzen 7 9800X3D, 63.6 GiB RAM, 4 TB NVMe, Windows 11 (10.0.26100), driver 610.62, CUDA 13.3.
Software: Strata/engine 0.1.43.1 (a0116b27), IQ3_XXS, 131,072-token context, KV int8. Install's own configuration, unchanged; experimental speed projection on.

Medians: short-chat decode 114.1 tok/s, 4K prompt 3,272.8 tok/s, 16K prompt 3,862.5 tok/s, agent follow-up decode 138.1 tok/s at 68.7% cache hit.