Qwen3.8-Flash NVFP4 wows on V100s: insanely fast prefill with Strata
Scobleizer · x · 2026-11-14
Developer @unbug shares a strong first impression of NVIDIA's Qwen3.8-Flash NVFP4 quantized model: running on V100s with the Strata inference framework, prefill is "insanely fast" while output detail consistency and coherence actually improve.
He also found Strata works even better with FP4-local, NVFP4, and NVFP4+Q8 Mix (GPTQ), calling the up-to-3K prefill performance "wild" — though he plans to validate on real-world workloads next.
More from Infra
- Cloud Reports Measure the Wrong Thing: Hybrid Cloud Is Often Legacy, Not Strategy — DavidLinthicum · 2026-11-14
- Will Depue claps back at Teortaxes over claim that 2-3K Blackwell runs aren't "data-center scale" — teortaxesTex · 2026-11-14
- Local LLM stack cheat sheet: seven open models mapped to chat, coding, TTS and more — alex_verem · 2026-11-14
- How a $1M 'fake chip' saved Nvidia in 1997 — the RIVA 128 bet that made Jensen Huang — bigaiguy · 2026-11-14
- SGLang Summit 2026 swag pins turn insider jokes like CUDA OOM and KV Cash into collectibles — ying11231 · 2026-11-14
- Nvidia's $500B GPU lending consortium stalls: zero loans closed, one lender may exit — SumitGup · 2026-11-14