Unverified90% confidenceFactTime unknown
Meta的Llama 3训练使用了超过16,000块H100 GPU
1
Sources
90%
Confidence
Long-term
Relevance
6/1/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedMeta此前已大量投资NVIDIA GPU用于模型训练72% similarVerifiedH100 GPU采用英伟达Hopper架构，单卡峰值算力达3,958 TFLOPS（FP16精度）70% similarUnverifiedGrok 4.5基于全新的V9基础模型，参数达到1.5万亿，是上一代V8的三倍，使用数万张英伟达GB300 GPU训练70% similarVerifiedGrok 4.5在数万台英伟达GB300 GPU上完成训练，支持最高50万Token超长上下文，并原生支持视觉输入69% similarUnverifiedGrok 4.5 是在数万块 NVIDIA GB300 GPU 上训练的66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/21220API
curl https://kongchang.com/api/v1/knowledge/claims/21220MCP
get_claim(id=21220)