🧵 Troubleshooting DeepLabCut installation: "TensorFlow was not built with CUDA kernel binaries compatible with compute capability" - articulateinstruments/AAA-DeepLabCut-Resources GitHub Wiki

DeepLabCut can use your GPU or your CPU. Normally it tries to use your GPU and if this goes wrong for any reason then it tries to use your CPU instead. Sometimes DeepLabCut gives errors and refuses to use your CPU if it has specific problems with your GPU.

You might get an error that looks like this:

"TensorFlow was not built with CUDA kernel binaries compatible with compute capability" / "CUDA kernels will be jit-compiled from PTX"

^{W tensorflow/core/common_runtime/gpu/gpu_device.cc:2027] TensorFlow was not built with CUDA kernel binaries compatible with compute capability 12.0. CUDA kernels will be jit-compiled from PTX, which could take 30 minutes or longer.}

This error happens if your GPU is too modern to work with the version of DeepLabCut used by AAA. AAA does not use the latest version of DeepLabCut, so some GPUs are too new for it to understand.

^{(We plan to update AAA in the future to use the latest version of DeepLabCut. This help page was written in 2026. If you are reading this in subsequent years please check if we have a more recent version of AAA which might support your GPU.)}

You might be able to overcome this issue by forcing DeepLabCut to use your CPU. You can do this by telling Windows to temporarily ignore your GPU for CUDA purposes.

Please follow these steps:

• Press your Windows key + R.
• Type in sysdm.cpl and run it.
• Go to Advanced.
• Click Environment Variables...
• In the section titled User Variables click the button to add a new variable.
• In Variable Name: type CUDA_VISIBLE_DEVICES. In Value put -1.
• Click OK.
• Close the window and restart your computer.

^{(Many thanks to Jeannene Lang Matthews for this clever solution!)}

If you want your computer to use your GPU again, please follow these steps:

• Press your Windows key + R.
• Type in sysdm.cpl and run it.
• Go to Advanced.
• Click Environment Variables...
• In the section titled User Variables look through the list to find the Variable you created in the solution above.
• Click the button to Delete the variable.
• Click OK.
• Close the window and restart your computer.