#2112 Community benchmark: RTX 4060 Ti 16 GB, Coder IQ1_M

说明

Adds bench/results/2026-10-12-community-rtx-4060-ti-qwen3-8-flash-next-coder-iq1-m/.

Hardware: NVIDIA GeForce RTX 4060 Ti 16 GB (driver 616.92, CUDA 13.4, PCIe gen4 x8, 165 W power limit), Intel i7-14700K, 31.7 GiB RAM, Samsung 990 EVO Plus 2TB NVMe, Windows 11 (10.0.26200).

Model/config: qwen3.8-flash-next-coder-iq1_m (IQ1_M), 65,536-token context, KV int8, expert-cache auto, spec 4, mtp rt, pool-workers 10; install's own configuration unchanged, thinking off, temperature 0.

Method: `python tools/strata_bench.py --long` (suite 1, 3 runs each, medians with ranges), engine's own timings, client TTFT.

Results: prompt 975.8 / 1223.8 / 1245.3 tok/s at 4K / 16K / 32K fresh; short-chat decode 30.6 tok/s; agent cache hit 68.5%. 0 failures, 261 s.

Limitations: speed only (no correctness check); background load not recorded; power draw in the report is the post-suite nvidia-smi sample.

Generated with the tool patch in #2111, which is why the README fills CUDA, storage, GPU detail and OS where an unpatched run left them "not read".