Is CachedAttention superseded?

CachedAttention (KV-cache compression): cited as a baseline — critiqued by newer work, not yet beaten on a benchmark here. 1 paper(s) critique it, 0 beat it on benchmarks — #125 of 234 most-superseded. Sub-problem: cluster led by KVFlow. Newer alternatives in the same sub-problem include PBKV, ShadowServe.

KV-cache compression

CachedAttention

Superseded baseline#125 of 234 most-superseded

Cited as a baseline — critiqued by newer work, not yet beaten on a benchmark here

1 papers critique it · 0 beat it on benchmarks

What papers say

Verbatim critique sentences, each from a paper that cites CachedAttention as a baseline.

CachedAttention overlaps layer-wise pre-loading with computation. However, it needs a dedicated, potentially unbounded pre-loading buffer in the GPU, while ShadowServe has a bounded GPU memory usage.

— ShadowServe: Interference-Free KV Cache Fetching for Distributed Prefix Caching

What to use instead

Recent methods in the same sub-problem, not yet superseded in the knowledge base — arXiv benchmark leaders, not vetted production recommendations.

PBKVEfficient Serving for Dynamic Agent Workflows with Prediction-based KV-Cache Management
May 7, 2026
ShadowServeShadowServe: Interference-Free KV Cache Fetching for Distributed Prefix Caching
Sep 21, 2025