Unverified50% confidenceFactExact time
大语言模型自回归生成阶段属于内存带宽受限任务，单batch推理算术强度约1-2 FLOP/Byte，远低于RTX 4090的计算/带宽比约160 FLOP/Byte
1
Sources
50%
Confidence
Long-term
Relevance
8/24/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedLarge model inference is typically memory-bandwidth-bound; the computational intensity of llama-2-70B inference is approximately 1–2 FLOP/Byte, far below the H100 peak computational intensity threshold of approximately 120 FLOP/Byte.78% similarVerifiedThe speed bottleneck in large model inference is memory bandwidth rather than insufficient computing power, as generating each token requires loading hundreds of GB of model weights from memory.72% similarUnverified单Token生成的算术强度约为1 FLOP/Byte，导致GPU计算单元利用率不到1%69% similarUnverifiedAutoregressive text generation is a memory-bandwidth-bound task, where memory bandwidth—not peak compute—is the true performance bottleneck.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/793632API
curl https://kongchang.com/api/v1/knowledge/claims/793632MCP
get_claim(id=793632)