lupinemachines/lupine ? reverse-engineered prompt

Build me a GPU over IP tool that lets a machine with no local GPU use a GPU on another machine as if it were attached locally.

I want a client and server setup where the server runs on the GPU host, the client runs on the CPU only host, and normal CUDA apps can work through it, including things like nvidia-smi, cuBLAS, and simple Python examples. It should be easy to run with Docker, and it would be great if the client can connect to one server or a list of servers. Please include a simple demo mode or example config, plus basic logging so I can see what’s happening when connections are made.

If you need to, look up current docs online for any CUDA or container details, but keep the experience simple for someone who just wants to point the client at a remote GPU and start using it.