Running Ripple 65B Preview NVFP4 on SGLang with 2x RTX PRO 6000: full recipe
HankYeomans · x · 2026-11-09
The author shared a complete open-source recipe for serving Ricursive AI's Ripple 65B Preview on SGLang (and vLLM).
Key setup:
- NVFP4 checkpoint: MLP layers in FP4 while attention, embeddings and LM head stay BF16
- SGLang 0.5.19, tensor parallel = 2 across two RTX PRO 6000 Blackwell cards (96GB each); the author says it can fit on a single GPU if desired
- DFLASH speculative decoding with 32 draft tokens
- Repo ships start/stop/bench/doctor scripts and profiles covering context settings, TP and KV quantization
He previewed the 65B model on coding tasks and found it competitive with much larger models.
Related event: Open Recipe for Running Ripple 65B on a Single GPU(2 posts)→
More from Infra
- Nvidia is the second-largest position in SPY, yet the largest position inside a SPY — dr1337 · 2026-11-09
- OpenAI CFO says datacenter expansion won't slow, stressing local community ties — i_dg23 · 2026-11-09
- 92T tokens over 1.54M DAU: the agent-era math feud over token burn — teortaxesTex · 2026-11-09
- Gary Marcus backs Ed Zitron: US data center capacity additions halved in Q4 2025 — GaryMarcus · 2026-11-09
- Analysts: Nvidia could stabilize near $155 if Huang is cleared, but an exit over security violations would hurt far more — GaryMarcus · 2026-11-09
- Texas gubernatorial candidate vows moratorium on all data center approvals — Traditional-Chip8339 · 2026-11-09