I’ve been playing with NU-refinement parameters recently, and noticed that when using a particle set with a box size of 384, if I enable higher-order CTF the job will eventually chew through all of the ram on my system - 96 GB in this case. Presumably this is a bug?
The only modified parameter was clicking the higher-order CTF job - everything else was left on the default setting. The same job runs fine with all setting on default, just a bit slow compared to homogeneous refinement.
We have been able to reproduce this and are working on fixing the memory over-allocation issues. We should be able to get it back to using as much or less memory as before CTF refinement was implemented.
Fantastic, thanks for the quick response!