Unverified50% confidenceFactExact time
LLM inference requests are divided into two phases—Prefill (input) and Decode (output); Prefill can read inputs in parallel with extremely high throughput, while Decode requires token-by-token serial inference and occupies significantly more GPU time.
1
Sources
50%
Confidence
Long-term
Relevance
9/2/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAt the hardware level, the Prefill (input) stage can read all tokens in parallel and compute the self-attention matrix simultaneously, whereas the Decode (output) stage requires serial autoregressive token-by-token inference, causing GPU utilization to drop sharply.77% similarVerifiedLarge language model inference is divided into two phases, prefill and decode: prefill is compute-intensive and sensitive to GPU processing power, while decode is a memory-bandwidth-bound task.77% similarUnverifiedThe Prefill phase is a compute-intensive operation, while the Decode phase is a memory-bandwidth-intensive operation.77% similarUnverifiedThe decode phase is a memory-bandwidth-intensive operation, and overall throughput is constrained by VRAM bandwidth rather than compute.74% similarUnverified输出Token定价高于输入Token是因为Prefill阶段可并行矩阵运算而Decode阶段必须串行自回归解码，Decode阶段GPU吞吐量约为Prefill阶段的1/6到1/1073% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/842836API
curl https://kongchang.com/api/v1/knowledge/claims/842836MCP
get_claim(id=842836)