Qwen3.8-27B on 16GB VRAM: 50 tok/s with 85k Context, Config Included
brainExploded99 · reddit · 2026-08-19
A Reddit user shares running Qwen3.8-27B on an Nvidia 5070Ti 16GB, achieving 85k context and 50 tok/s by removing the MTP layer and tuning config. Includes full config and tips like q8 cache and ubatch-size ablation.
More from coding & agent
- Six-month prediction: hiring for generic agent harnesses built from scratch will cool — yi111 · 2026-10-16
- Omarchy ISO wouldn't boot on NVIDIA, so he had Claude patch it — indragie · 2026-10-16
- Dev uses Grok to clean up 3D printer firmware, catches it sneaking in nonsense code — yacineMTB · 2026-10-16
- A rude AI agent took 89% of a shared resource in 102-run multi-agent personality experiment — Brave-History-6502 · 2026-10-16
- Guardrails can make risky delegation work — but it takes a full-time team, says iandanforth — iandanforth · 2026-10-16
- What can you truly delegate? Only tasks with unambiguous, verifiable results — iandanforth · 2026-10-16