Unverified50% confidenceFactExact time
NVIDIA B200于2024年发布，FP16算力达到4.5 PFLOPS级别
1
Sources
50%
Confidence
Long-term
Relevance
8/5/2026
First Seen
Sources
Related Claims
UnverifiedNVIDIA A100的理论FP16峰值算力约为312 TFLOPS，但朴素LLM推理场景下GPU利用率常不足30%73% similarVerifiedNVIDIA H100单卡FP64双精度浮点性能达到约34 TFLOPS72% similarUnverifiedB200的FP4算力峰值可达约9PFLOPS（稀疏模式），相比FP16提升可达8倍71% similarUnverified2025年发布的NVIDIA 5090将显存提升至32GB并引入了FP4推理支持70% similarVerifiedNVIDIA A100单卡拥有80GB HBM2e显存和312 TFLOPS的FP16算力68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/688643API
curl https://kongchang.com/api/v1/knowledge/claims/688643MCP
get_claim(id=688643)