Unverified50% confidenceBenchmarkExact time

NVIDIA 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%

1

Sources

50%

Confidence

Long-term

Relevance

7/11/2026

First Seen

Sources

vLLM推理框架详解：吞吐优化核心原理与面试攻略

bilibiliAI大模型升升6/9/2026

Related Claims

VerifiedNVIDIA H100单卡FP64双精度浮点性能达到约34 TFLOPS82% similar VerifiedNVIDIA H20 的 BF16 峰值算力约为 296 TFLOPS，相比 H100 的 989 TFLOPS 大幅削减，但保留了 141GB HBM3 显存与 4.0 TB/s 显存带宽76% similar UnverifiedT4 GPU拥有16GB显存和8.1 TFLOPS的FP32算力74% similar UnverifiedNVIDIA B200于2024年发布，FP16算力达到4.5 PFLOPS级别73% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/488591

API

curl https://kongchang.com/api/v1/knowledge/claims/488591

MCP

get_claim(id=488591)