Apple M1 Max vs CUDA on chest analysis pack

[kirion (Klaus)]

Hi, I am running the chest platform on a Mac M1 Max, but it seems too slow as it looks for CUDA instead of the MPS. Does anyone knows how to set it for apple M1 Max processors, please?

[jamesobutler (James Butler)]

You are using SlicerTotalSegmentator?

I have detailed some development work that is required to get full MPS support working and needs a developer to help champion this work:

GPU not detected Total Segmentation

TotalSegmentator appears to have added the option for MPS where your Mac Pro AMD Radeon Pro Vega II card supports the mps backend for PyTorch. It does seem to be a bit on the untested and new side as the following place in the readme hasn’t been updated to reflect the mps option although you can see that --device does allow the “mps” option. The SlicerTotalSegmentator utilizes the TotalSegmentator python package, but will need to be updated to handle the MPS option as right now if it doesn’t…

Also note that Slicer only provides x86_64 binaries, so on macOS you won’t be able to utilize latest PyTorch since the torch developers dropped support for x86_64. This means things may still not be optimized for your system. Building Slicer native arm64 builds for macOS is work still in progress by the Slicer developers.

[kirion (Klaus)]

Hi James,

I am not sure if you have received my prior reply to your message.

Yes I am trying to use totalsegmentator but it is too slow on my MAC and I would be really keen to get it adapted, as I do not want to change my computer, which should be fast enough if not relying on CUDA.

Kind regards

Klaus