Using RunCellpose with a GPU#

If you want to use a GPU to run the model (this is recommended for speed), you’ll need a compatible version of PyTorch and a supported GPU. General instructions are available at this link.

1. Your GPU should be visible in Device Manager under Display Adaptors. If your GPU isn’t there, you likely need to install drivers. Here is where you can find NVIDIA GPU drivers if you need to install them.

2. To test whether the GPU is configured correctly:

• Run python on the command line (i.e., in Command Prompt or Terminal) to start an interactive session

• Then run the following

import torch
torch.cuda.is_available()

• If this returns True, you’re all set

• If this returns False, you likely need to install/reinstall torch. See here for your exact command.

• Exit the session with exit() then install torch if necessary

pip3 install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu113

If you have a previous version of torch installed, make sure to run pip uninstall torch first.

NOTE: You might get a warning like this:

W tensorflow/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'cudart64_110.dll'; dlerror: cudart64_110.dll not found
2022-05-26 20:24:21.906286: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine.

If you don’t have a GPU, this is not a problem. If you do, your configuration is incorrect and you need to try reinstalling drivers and the correct version of CUDA for your system.