Nemotron 3 Ultra's 550B params mogged by models 20x smaller, SemiAnalysis blasts NVIDIA's frontier training
teortaxesTex · x · 2026-09-03
SemiAnalysis argues that while NVIDIA Research produces excellent architecture work (LatentMoE used in Kimi K3, GatedDeltaNets in Qwen), NVIDIA's bureaucratic culture undermines end-to-end frontier training: Nemotron 3 Ultra, with 550B total params (55B active), is outperformed by all Chinese models, including Qwen3.8 27B with 20x fewer parameters. teortaxesTex adds that architecture R&D output tends to be anticorrelated with model quality and questions the dogged criticism of Nemotron 3.
More from Companies & People
- Nyne, under 15 people with tens of millions in monthly revenue, hires SPLs and RL engineers — soleio · 2026-10-21
- OpenAI: new drugs take 10–15 years, AI agents can fix fragmented workflows — Dr_Singularity · 2026-10-21
- OpenAI acquires Orchestra Bio to speed drug development with AI agents — Dr_Singularity · 2026-10-21
- UNESCO and ICOM Release 2026 Report on AI Use in Museums Across 90 Countries — LuizaJarovsky · 2026-10-21
- AI app-layer acquisitions are now mostly about founders and customers, not tech — signulll · 2026-10-21
- Vanta CEO on the 'boring and complex' playbook, eating AI costs, and 10X comp — soleio · 2026-10-21