mlock and load time

[Syhong330]

Hello, I'm using llama.cpp to run llama2 in Windows.
When I set '--mlock' option on, the load time seems to increase by about 2 seconds. As I know it's stored in the committed area of RAM, but I'd like to know why the load time increases so much.

[aviallon]

@Syhong330 it is simply because you are asking the kernel to load the entire weights into RAM at once, instead of lazy-loading them. Said differently, it is simply the time it takes for your computer to move the data from storage to RAM.