#1967 Community benchmark: RTX 5080, Ryzen 7 7800X3D, 64 GB RAM (Windows)

Beschreibung

Results-only community benchmark on a Windows desktop (see bench/results/2026-10-11-community-rtx-5080/).

Hardware: RTX 5080 16 GB, Ryzen 7 7800X3D (16 logical CPUs), 63 GiB RAM.
Software: Strata 0.1.42, locally built source (CUDA 13.4, sm_120), Windows 11 build 26200, NVIDIA driver 616.64.
Model: ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-GGUF at the installer's pinned revision, IQ3_XXS, 262,144-token context.

Measured: 4,096 / 32,768 / 128,000 prompt tokens, three runs each, 256-token output cap.
Decode 91.9 / 94.8 / 95.5 tok/s, prompt 3,060.9 / 3,461.9 / 3,607.0 tok/s, TTFT 1.38 / 9.52 / 35.57 s.
All nine runs with zero reused tokens; needle check 6/6.

Not tested: vision, thinking/sampled decoding, coding-task quality, concurrency, long soak.
One machine, one configuration; no cross-engine baseline.