Unverified50% confidenceFactExact time

GPU显存并非100%可用，因为GPU驱动、CUDA运行时上下文、显示输出和操作系统图形合成器都会预先占用一部分显存

1

Sources

50%

Confidence

Long-term

Relevance

8/12/2026

First Seen

Sources

Running Local LLMs on 24GB VRAM? Your Actual Usable Memory Is Less Than You Think

Related Claims

UnverifiedGPU设备的状态存储于驱动私有的内核数据结构与显存中，绕过了Linux标准的进程状态模型，CRIU无法处理GPU工作负载79% similar UnverifiedEven when using Docker, if the NVIDIA driver version and CUDA runtime version are not fixed, GPU computation results may still be not fully consistent due to differences in the underlying drivers.76% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/735387

API

curl https://kongchang.com/api/v1/knowledge/claims/735387

MCP

get_claim(id=735387)