NVIDIA's first Rubin NVL72 Kimi K3 submission hits 850 tok/s/user at 2.8T params
scaling01 · x · 2026-10-28
Per SemiAnalysis, NVIDIA submitted its first Rubin NVL72 Kimi K3 inference submission, reaching over 850 tok/s/user even at 2.8 trillion parameters, with 3x+ higher throughput on vLLM. Rubin Software 1.0 GA is coming soon. scaling01 muses what Astra- or Fable-class models at 400 tokens/s would feel like.
More from Infra
- Statistically Lossless Qwen3.8 27B GGUF Quants Hit ~99% EAR at 23GB, Feels FP8-like — Dany0 · 2026-10-28
- Tonawanda Seneca Nation opposes data center on wetland at People's AI Assembly — AINowInstitute · 2026-10-28
- Kimi K3.1 teased to adopt DSA like DeepSeek V3.2 for better inference economics — zephyr_z9 · 2026-10-28
- VLM Run Gateway Launches: One API for Open-Weight OCR, VLM and Vision Models — spillai · 2026-10-28
- Hyperscalers issue $228B in bonds YTD, nearly 14x the 2024 total — Beth_Kindig · 2026-10-28
- Bittensor subnet Targon burns ~$2.1M of tokens from organic GPU revenue — markjeffrey · 2026-10-28