How to Install PyTorch with CUDA Using conda-forge (Now that the Official Conda Channel Is Deprecated)

Learn the current way to install PyTorch with CUDA using conda: get the GPU build from conda-forge now that the official conda channel is deprecated. See how.

Installing PyTorch with conda used to be a straightforward path from the official website. Today that option is gone from the installer page, yet many teams still rely on conda-based workflows and want a CUDA-enabled build. The good news: PyTorch is available on conda-forge, including a GPU build. The nuance: it’s a community-maintained channel, and the official PyTorch conda channel has been deprecated.

What changed and why it matters

The official PyTorch site no longer provides a conda installation option. At the same time, the package remains available on conda-forge, which is maintained by the community. The PyTorch developers have deprecated their official conda channel, which shifts the center of gravity to conda-forge for users who prefer conda environments.

The practical ask

If your requirement is to install PyTorch via conda with CUDA enabled, there is still a single-step path using the conda-forge channel.

Solution

Use the following command to install the CUDA-enabled build from conda-forge:

conda install pytorch-gpu -c conda-forge

This installs the GPU-enabled variant of PyTorch using the conda-forge channel.

What to expect and what not to assume

The conda-forge packages are a community effort rather than an officially maintained distribution from the PyTorch developers. Whether this is the “recommended” way depends on your project policies and trust in community-maintained channels. The fact that the official conda channel is deprecated is the key context for making that decision.

Why knowing this saves time

Teams often standardize on conda for reproducibility and environment isolation. Knowing that the official route is deprecated and that conda-forge provides a CUDA-enabled build helps avoid chasing outdated instructions and keeps CI/CD pipelines and developer onboarding clean.

Takeaway

If you want PyTorch with CUDA via conda, install it from conda-forge using the single command shown above. Keep in mind the package is community-maintained, and align that with your organization’s packaging and provenance requirements.