Qwen3.8 gets Day-0 support from LightSeek, boosting inference performance by 30%+
Alibaba_Qwen · x · 2026-08-15
Alibaba's Qwen team announced Day-0 support for Qwen3.8 (2.4T params) with LightSeek on NVIDIA Blackwell. LightSeek optimized DP/EP scaling across nodes, delivering 30%+ faster performance than TP16, plus DSpark speculative decoding with single CUDA graph optimization.
More from Infra
- Macro Analysis: Is the AI Dip a Buy? Key Levels to Watch — Beth_Kindig · 2026-08-15
- Sanders Threatens AI Pause, Nvidia Partners with Wall St on Compute-Backed Securities — PeterDiamandis · 2026-08-15
- Fireship: How edge ML cameras built a warrantless tracking network — Fireship · 2026-08-15
- Qwen Official Guide: Tuning reasoning depth and extending context to 1M tokens — solyarisoftware · 2026-08-15
- DeepSeek V4 Pro Requires Linux/WSL to Function Properly — teortaxesTex · 2026-08-15
- Tiny-Qwen update: PyTorch-native build for Qwen 3.8 27B model — No-Compote-6794 · 2026-08-15