#3063 bench: community report, RTX 5070 12 GB + 32 GB RAM on Windows 10, Flash-Next IQ2_XS low-RAM mode (0.1.45)

本文

## Title
Issue: none (results-only report)

## Summary
A results-only community report from an RTX 5070 12 GB with 31.8 GiB of RAM on Windows 10: the original Flash-Next IQ2_XS in low-RAM mode (`--mmap-experts`), Strata 0.1.45, measured with `tools/strata_bench.py`. As far as I can see there is no single-GPU RTX 5070 (non-Ti) report yet (it appears only as one card of the three-GPU layer split in #2086), and this PC has a slower NVMe drive than the 31.6 GiB RTX 5070 Ti report in #2791.

## What changed
- `bench/results/2026-10-18-community-rtx-5070-qwen3-8-flash-next-iq2-xs/README.md` and `results.json` (new)

No engine or code changes. I left `docs/COMMUNITY_BENCHMARKS.md` alone so you can decide where to list it; a one-line entry is ready if you want it in this PR.

## Extra Notes
- Two workarounds were needed to run the 0.1.45 bench tool on this PC, both already reported: the server was started separately and the tool used `--url` (#2864), and the tool ran with `PYTHONUTF8=1` on a Chinese-locale Windows (#2792).
- The daily configuration was used except that the optional control-vector file was left out, so the numbers compare with other reports.
- Speed only, one machine, one model, 32K context.
- I also have opt-in changes that speed up the prompt path in this setup; I will describe them in a separate issue.