LLM should fit in video memory, but the RAM is also heavily loaded.

[by-noob]

I'm using "llama-b7472-bin-win-vulkan-x64" with a Radeon Rx 470 Series graphics card with 8GB of video memory and LLM models that fit completely.

When loading the model, the task manager shows that the use of memory allocated to the GPU increases, and at the same time, the use of RAM also increases by the same amount in "router mode", and in normal mode, the RAM increases by about two-thirds of the GPU memory.

When I try to see which process is consuming RAM, I understand that it's llama cpp, but when loading a model, its usage only increases by 600MB, while RAM usage increases by the full 6GB.

p.s. I hope this isn't because of another Windows 10 "update" that has broken part of the system again.