rgpu

What is rgpu?

rgpu allows developers to run GPU computations on a remote NVIDIA machine while keeping their Python or C++ application running locally. It supports both macOS and Linux clients, making it possible to use powerful GPUs without installing CUDA drivers on the local machine.

The tool provides two integration methods. The PyTorch device path lets Python programs specify an rgpu device for tensor operations over a TCP connection. The CUDA shim path intercepts standard CUDA calls like libcuda and cuDNN, allowing existing Linux binaries to run remotely.

Setup involves installing the package via pip, deploying a remote server, and using an SSH tunnel to secure the connection. Documentation includes quickstart guides, a nanoGPT training example, configuration references, and performance benchmarks. The project is licensed under Apache 2.0.

rgpu features

• Run PyTorch tensor operations on a remote NVIDIA GPU
• Intercept CUDA calls to route existing binaries to remote hardware
• Secure connections through SSH tunnels
• Compatible with Apple Silicon laptops lacking CUDA drivers
• Includes documentation, quickstart guides, and benchmark reports
• Provides a Python package installable via pip

Who uses rgpu and for what

• Train deep learning models on a cloud GPU while coding on a MacBook
• Run legacy CUDA applications on a remote server without local driver setup
• Develop PyTorch scripts that automatically offload heavy computation to a cluster node
• Test machine learning workloads across different GPU configurations remotely

rgpu FAQ

Comments (0)

Comments and upvotes open when this product launches on ProductPerch.