@EricThomson Yup in general the same requirements as Caiman if you're using those modules.
CNMFE is more resource intensive than CNMF. If your imaging resolution is ~512x512 and your movies are shorter than 10 mins you might be able to get away with 32GB or less. For longer movies I think it's recommended to use CNMFE online anyways. Faster RAM and quad channel RAM also seems to help for both CNMFE and motion correction (quad channel is usually only availalbe on Xeon and Threadripper I think).
For certain downstream things like computing Earth Mover's Distances like in our papers, very high thread count (64+) is necessary for large datasets (thousands of traces) and the computation will take days. Similar for kshape clustering.
Use motion correction kwargsbox and manually add in
pw_rigid=False, and then just set the rigid correction params I want (and ignore the Elastic correction params)?