#2012 Community benchmark: RTX 5090 + RTX 3090 (IQ3_S, 131K ctx, engine 0.1.43)

Descrição

## Summary
Adds a community benchmark report for a layer-split RTX 5090 + RTX 3090 rig (56 GiB VRAM total), Ryzen 9 7900X, 191 GiB RAM, Windows 11, official Windows engine 0.1.43.

- Model: original Flash-Next IQ3_S, 131,072-token context, install-default settings (--spec 4, --mtp, int8 KV)
- Short chat decode: 79.1 tok/s (TTFT 0.38 s)
- Agent session decode: 115.8 tok/s at 68.8% conversation-cache hit
- Fresh 4K / 16K prompt read: 1,945 / 2,282 tok/s
- Three runs per test, medians with ranges; speed suite only, no correctness checks

Files: `bench/results/2026-10-12-community-2x-rtx-5090-qwen3-8-flash-next-iq3-s/` (README.md + results.json) and an index entry in `docs/COMMUNITY_BENCHMARKS.md`. Results-only submission, no engine changes.

## Limitations
- No storage details, background load, or power limits recorded
- No needle/correctness checks run
- Parallel slots: 1 (default)