Unverified50% confidenceBenchmarkExact time
以LLaMA-2 70B为例，处理4096 token上下文、批量大小32时，KV Cache可消耗超过80GB显存
1
Sources
50%
Confidence
Long-term
Relevance
7/15/2026
First Seen
Sources
Related Claims
UnverifiedLLaMA-2-70B在128K上下文长度下单请求KV Cache可占用超过160GB显存82% similarUnverified100万Token在FP16精度下仅KV Cache一项即可消耗数百GB显存，远超单张A100/H100的80GB容量上限82% similarUnverified当上下文长度扩展至百万Token时，KV Cache本身就会占用数十GB显存81% similarUnverified以LLaMA-2 70B模型为例，2048 token的序列需约4GB显存仅用于KV缓存80% similarUnverified对于 LLaMA-2 70B 模型，处理 4096 token 的上下文时，FP16 精度下 KV Cache 约需 20GB 系统内存79% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/512128API
curl https://kongchang.com/api/v1/knowledge/claims/512128MCP
get_claim(id=512128)