- Category
- Developer Tools
- Rank
- No. 3718Tools index
Previous survey · No. 3537 ·
- Pricing
- Open Source
- Type
- TOOL
- Use case
- Models: Train & Run
- GitHub
- 80 stars
- Date
About
rGPU runs GPU work on a remote NVIDIA machine while the application stays on the client. It lets PyTorch programs hold tensors and run operations on a remote GPU, including from a Mac with no CUDA installation. It also offers a CUDA shim path for existing Linux CUDA programs.
At a glance
rGPU lets a program on one machine send GPU operations to a remote NVIDIA host. It is worth evaluating if you want to keep the application on your current machine and can check whether the network overhead fits your workload.
Good fit
Developers with access to a remote NVIDIA machine who can test either a PyTorch device integration or a Linux CUDA shim.
Consider something else when
Skip it if your workload needs unsupported CUDA calls or is sensitive to many small network round trips.
Start here
Read the PyTorch smoke test and use it as the first check in an isolated setup. Once a simple tensor operation works, time one representative batch against running the same code on the GPU host.
Cost and requirements
Apache-2.0 project. The documented setup needs a client, a remote NVIDIA GPU host and SSH access. Rented hardware and network usage can add costs.
Limits to know
The native PyTorch path and CUDA shim have different compatibility surfaces. The protocols do not authenticate or encrypt traffic. Use the documented SSH tunnel or a controlled private network, and consider what the GPU host can see.
Sources checked 2026-11-05. Source-based assessment.
Tags
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.