CUDA 12.8 and Newer Cards Workaround [READ FOR SOLUTION IF YOUR SLICER DEFAULTS TO USING CPU TORCH]

[lukepolson]

Slicer doesn't currently seem to support versions of Torch with CUDA>=12.8 using the pytorch-utils extension. There is a workaround by opening the python terminal within Slicer and typing

import sys
import subprocess
subprocess.check_call([
    sys.executable, "-m", "pip", "install",
    "torch==2.10.0+cu128",
    "torchvision==0.25.0+cu128",
    "--index-url", "https://download.pytorch.org/whl/cu128"
])

This will give GPU-accelerated capabilities to systems with newer cards (I was a little frustrated that my new 5090 was defaulting to CPU when pytorch-utils extension default installed the CPU version!)