SGLang Hits 873 tok/s on 4×GB300 With DeepSeek V4.1 Flash Day-0 Support
hsu_byron · x · 2026-09-12
SGLang shipped Day-0 support for DeepSeek V4.1 Flash and pushed performance to 873 tok/s at BS=1 on 4×GB300 within 24 hours, via FP8 GEMM fast paths, kernel fusion and overlap, DSpark optimization, and MoE TP4.
V4.1 Flash introduces major architectural changes — a causal encoder-decoder design, CSA2 shared KV, Engram, mHC, and DSpark — all supported from day zero. The team published a hands-on engineering guide with step-by-step reproduction instructions; DeepGEMM and FlashMLA optimizations are coming next.
Related event: SGLang delivers Day 0 support for DeepSeek V4.1 Flash, hitting 873 tok/s(2 posts)→
More from Infra
- GPU price index tracks 114 rates from 25 vendors, free MCP server for agents — StasBekman · 2026-10-31
- CMU paper unifies pretraining, SFT and RL into one likelihood-based training algorithm — rohanpaul_ai · 2026-10-31
- Boom Supersonic airliner in doubt as company pivots to AI data center turbines — Ars Technica AI · 2026-10-31
- AWS adds AgentCore Session endpoints to preview coding agent web apps live in browser — AWS ML Blog · 2026-10-31
- Data center and AI infrastructure investment to hit $10.3T by 2032, per WSJ — Beth_Kindig · 2026-10-31
- Altman says RAM belongs in data centers, not your PC, as OpenAI's buying spree squeezes memory supply — tomwarren · 2026-10-31