I have a 128-d vector, and I need to compare it with a large database of more than 15GB. I want to perform this operation using Pytorch GPU, but it is not possible to fit the entire data in VRAM of one single GPU. I have a similar unused server with same configuration, and wanted to use it for the searching. Essentially, dividing the 15GB data into chunks of smaller data and load into GPUs for multiple devices.
Is there a way or a software that can do this?