Unverified50% confidenceBenchmarkExact time
Puget Systems 测算显示 R9700 的 FP32 约为 B70 的两倍，矩阵吞吐约为其 5 倍
1
Sources
50%
Confidence
Long-term
Relevance
10/5/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedFP8数据量比FP16压缩一半，H100的FP8 Tensor Core算力达到FP16的两倍（约1979 TFLOPS）74% similarUnverified序列长度为4096时，N×N注意力矩阵在FP16精度下仅此一项就消耗约64MB显存69% similarVerifiedLLaMA-2-7B使用FP16精度时每个token的KV Cache约占0.5MB69% similarUnverified2.4万亿参数模型在FP32下约需9.6TB显存，BF16约需4.8TB，INT8约需2.4TB，INT4约需1.2TB69% similarUnverifiedGB200 NVL72在FP8精度下的AI算力可达1.44 ExaFLOPS69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/977976API
curl https://kongchang.com/api/v1/knowledge/claims/977976MCP
get_claim(id=977976)