RGPU – Use a remote Nvidia GPU from PyTorch on a machine that has none

rGPU / Product documentation

Your GPU, anywhere.

Keep Python on your laptop. Run PyTorch operations and hold tensors on a remote GPU, including from a Mac with no CUDA installation.

Local code · remote GPU

Python stays with you. Tensors stay on the GPU.

train.py · local Python

import rgpu, torch

x = torch.randn(1024, 1024, device="rgpu")
y = (x @ x).relu().sum()
print(y.item())

Local Python → SSH tunnel → GPU server

Train with PyTorch #

Move a model, run autograd, and ship compiled graphs.

Use the CUDA shim #

Keep device="cuda" with the Linux driver-remoting path.

Understand the cost #

Measure host waits, transfers, and completed GPU work.

Built for trusted GPU hosts. Read the deployment guidance before connecting.