// ⚡ Inference Acceleration & Serving
Prefill–Decode Disaggregation
Runs prompt prefill and token decoding on separate GPU pools.
预填充/解码分离 2024 Advanced extended
IN PLAIN WORDS
Prefill is compute-bound and decode bandwidth-bound; co-locating them causes interference. Separating them lets each scale to meet latency targets with higher goodput.
Used in: Large serving clusters
01Original paper
DistServe: Disaggregating Prefill and Decoding for Goodput-optimized Large Language Model Serving
Yinmin Zhong, Shengyu Liu, Junda Chen, et al. · 2024
arXiv ↗ ✓ Verified · arXiv API
02Where it sits
Learn the left column first; the right column builds on this one.
Then you can learn
- Nothing in the atlas builds on it yet
Related:PagedAttention (vLLM)
Left: prerequisites · right: builds on this · dashed: related. Full map →
03Learning resources
Official docs, explainers, courses and reference code. All links checked.
04Where the course teaches it
For AI / data
- Verification: verified, checked 2026-10-21.
- JSON: /api/learn/algorithms.json (id:
pd-disaggregation)