Open source developers recently launched a tool called NBD-VRAM, which can divide part of the video memory of consumer-grade NVIDIA GeForce GPU into swap space under Linux system to make up for the lack of system memory, especially for laptop users with independent RTX graphics cards whose memory is soldered and difficult to expand.
According to reports, NBD-VRAM was announced yesterday. It works by running a small daemon in the background, applying for a piece of video memory space on the graphics card through the NVIDIA CUDA driver API, and then exposing the video memory as a block device through the Unix socket through the Linux Network Block Device (NBD, Network Block Device) protocol. Finally, it appears in the system as a regular Linux swap partition (swap device) and can be used directly.
Developers clearly position NBD-VRAM as a solution for NVIDIA consumer GPUs to provide an additional "memory overflow buffer channel" in scenarios where the official peer-to-peer (P2P) API and other alternative mechanisms cannot be used. Users need to enable NVIDIA's official Linux graphics driver and CUDA support in the system. Open source drivers such as Nouveau or Nova cannot be used, otherwise the tool will not work properly.
The project is released as open source under the MIT license. Interested users and developers can obtain the source code and learn more implementation details on GitHub. The project is currently under continuous improvement:
https://github.com/c0dejedi/nbd-vram