Unverified50% confidenceTradeoffExact time
Keeping model weights fully resident in GPU memory avoids memory swap-in/swap-out overhead, whereas models that exceed GPU memory capacity require data to be moved between CPU and GPU, causing latency to rise significantly.
1
Sources
50%
Confidence
Long-term
Relevance
9/30/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedGPU的SM能同时驻留的Warp数量决定了GPU隐藏内存访问延迟的能力，制约Occupancy的核心资源是寄存器文件与共享内存80% similarUnverifiedWhether a model fits entirely in VRAM is more important than the absolute size of the model file; once CPU offloading occurs, the data transfer overhead between GPU and CPU negates the space advantage of a smaller model.79% similarUnverifiedThe KV cache introduced by extremely long contexts grows linearly with context length, and Unsloth's optimized offload planner prioritizes keeping the KV cache on the GPU.77% similarUnverifiedGPU上下文切换需保存完整GPU状态（包括寄存器文件、共享内存和L1缓存），数据量可达数百MB，而CPU上下文切换仅需保存几KB77% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/962966API
curl https://kongchang.com/api/v1/knowledge/claims/962966MCP
get_claim(id=962966)