Unverified50% confidenceBenchmarkExact time
The NVIDIA H100's FP16 peak can reach 1000 TFLOPS, but when training Transformer models the MFU is typically only 30%–60%.
1
Sources
50%
Confidence
Long-term
Relevance
9/7/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe NVIDIA A100's theoretical FP16 peak compute is approximately 312 TFLOPS, but GPU utilization in naive LLM inference scenarios is often below 30%.79% similarVerifiedThe NVIDIA H100 single-card FP64 double-precision floating-point performance reaches approximately 34 TFLOPS.79% similarVerifiedThe ridge point of the NVIDIA H100 is approximately 597 FLOP/Byte (approximately 2000 TFLOPS ÷ 3.35 TB/s).70% similarUnverifiedThe typical generation speed of an NVIDIA RTX 4090 running a similarly scaled Q4 quantized model is in the range of 40–60 TPS68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/870406API
curl https://kongchang.com/api/v1/knowledge/claims/870406MCP
get_claim(id=870406)