Unverified50% confidenceFactExact time
自回归生成中每生成一个token需读取全部权重一次，内存带宽而非算力峰值才是真正的性能瓶颈
1
Sources
50%
Confidence
Long-term
Relevance
7/17/2026
First Seen
Sources
Related Claims
UnverifiedAutoregressive text generation is a memory-bandwidth-bound task, where memory bandwidth—not peak compute—is the true performance bottleneck.84% 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.75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/547159API
curl https://kongchang.com/api/v1/knowledge/claims/547159MCP
get_claim(id=547159)