SemiAnalysis 发文称 NVIDIA Research 产出过 LatentMoE（用于 Kimi K3）和 GatedDeltaNets（用于 Qwen）等优秀研究，但在端到端前沿训练上，NVIDIA 的官僚文化导致 Nemotron3 Ultra 表现尴尬。Nemotron3 Ultra 总参数 550B（55B 激活），被包括 Qwen3.8 27B（参数量约少 20 倍）在内的中国模型全面超越。
SemiAnalysis@SemiAnalysis_SemiAnalysis 发文称 NVIDIA Research 产出过 LatentMoE（用于 Kimi K3）和 GatedDeltaNets（用于 Qwen）等优秀研究，但在端到端前沿训练上，NVIDIA 的官僚文化导致 Nemotron3 Ultra 表现尴尬。Nemotron3 Ultra 总参数 550B（55B 激活），被包括 Qwen3.8 27B（参数量约少 20 倍）在内的中国模型全面超越。
NVIDIA Research 🚀 has produced some great research, like LatentMoE (used in Kimi K3) and GatedDeltaNets (used in Qwen). But for e2e frontier training, NVIDIA's bureaucratic culture has produced embarrassing models like Nemotron3 Ultra.
Despite NVIDIA Research having amazing talent, Nemotron3 Ultra, with 550B total params (55B active), is getting mogged by all the Chinese models, including even Qwen3.8 27B parameters, which has ~20x fewer parameters.
来源：SemiAnalysis· x.com