Qwen3.8-27B Dual 3090 Benchmark: Detailed Perf Data and Configs
No-Statement-0001 · reddit · 2026-08-15
User shared detailed benchmarks of running Qwen3.8-27B (Q8KXL) on dual RTX 3090s, generating a retro GeoCities page packed into a Go binary via a single prompt.
- Output Quality: Code worked perfectly without adjustments, demonstrating the model's capability in complex instruction following and full-stack generation.
- Performance Metrics:
- Total duration 475s (with extensive thinking), processing 20k tokens.
- Prompt processing speed varied (10-464 t/s), generation speed hovered around 40-70 t/s.
- Preserving reasoning tokens significantly improved results.
- Key Configuration:
- Model: Qwen3.8-27B-UD-Q8KXL.gguf
- Context: 238000 (due to VRAM limits)
- Speculative Decoding: Enabled draft-mtp (--spec-type draft-mtp), max drafts 3.
- Params: temp 0.6, topp 0.95, preservethinking true.
More from coding & agent
- Terminal Night 3 talk slides released, missing demo video to follow — 4310sy · 2026-10-16
- Agent developers are sitting on unused API credits that can exceed actual usage — EnvironmentalDay8864 · 2026-10-16
- Notion's bet on agents: abstract away deployment so users and agents just write code — simonguozirui · 2026-10-16
- Four Gemini 4 Argon agents beat Minecraft's Ender Dragon autonomously in 27 hours — Maxence Faldor · 2026-10-16
- Tencent's Terminal-RSI builds recursively self-improving terminal agents, hitting 19.84% on Terminal-Bench 4.0 — tencent · 2026-10-16
- Tsinghua's AutoSFT uses multi-agent collaboration to automate SFT data research — TsinghuaNLP · 2026-10-16