Qwen3.8-27B hits 268 tok/s on single Blackwell GPU with full context
EAccelerate_42 · x · 2026-08-24
A developer released an optimization recipe for Qwen3.8-27B on a single RTX PRO 5000 Blackwell (48GB). Using SGLang, DFlash2 block-16, and Triton attention, it achieves 267.8 tok/s (310.7 peak) with the full 262K context, doubling the EAGLE baseline. Reproducible scripts are provided.
More from coding & agent
- raylib 6.2 ships with 300+ improvements and 100 new contributors, 20 new API functions — jsuarez · 2026-10-16
- "Vibecoding" is as misleading a label as "AI" and "SI," dev complains — eschadiol · 2026-10-16
- Doc2Harness: New benchmark finds best coding agent passes only 53% across 329 cases — _akhaliq · 2026-10-16
- Perplexity open-sources SPACE, the Firecracker microVM sandbox behind Perplexity Computer — perplexity_ai · 2026-10-16
- Blackstone goes all-in on Factory for autonomous software development — matanSF · 2026-10-16
- Longtime Windsurf user cancels $10/month legacy plan, citing AI subscription fatigue — tristanbob · 2026-10-16