Unverified90% confidenceFactTime unknown
GB200 NVL72在FP8精度下的AI算力可达1.44 ExaFLOPS
1
Sources
90%
Confidence
Long-term
Relevance
6/1/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedA100拥有40/80GB显存和19.5 TFLOPS的FP32性能74% similarUnverifiedUP主使用FP8精度部署千问3.8 27B在48GB显卡上，跑4个并发request占用约46GB70% similarUnverified35B模型在FP16精度下需要约70GB显存，远超RTX 5090的32GB容量，因此单卡部署需量化70% similarUnverifiedB200的FP4算力峰值可达约9PFLOPS（稀疏模式），相比FP16提升可达8倍70% similarUnverifiedLLaMA-2-7B使用FP16精度时每个token的KV Cache约占0.5MB69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/16835API
curl https://kongchang.com/api/v1/knowledge/claims/16835MCP
get_claim(id=16835)