Automatically downsampling boxes during late NU-refinement

Dear colleagues,

we are currently performing non-uniform refinement on a large volume (about 350 A in each direction), with a full box size of 550 cubed voxels; we have about 370 thousand particles extracted with 550 x 550 px (1.1 A per px).

After four or five cycles of refinement, as we approach a GSFSC resolution of 3.1 A, the following message appears repeatedly:

*[CPU: 50.68 GB] Optimizing FSC Mask… *
[CPU: 50.68 GB] Using full box size 550, downsampled box size 288, with low memory mode disabled.
[CPU: 50.68 GB] Computing FFTs on GPU.
[CPU: 51.92 GB] Using full box size 550, downsampled box size 288, with low memory mode disabled.
… (several more times)
[CPU: 51.92 GB] Using full box size 550, downsampled box size 288, with low memory mode disabled.
[CPU: 51.92 GB] Computing FFTs on GPU.
[CPU: 51.92 GB] Done in 150.816s

Then, NU-refinement consistently ends at around 3-ish A GSFSC, regardless of further significant improvements of earlier steps, from motion correction to CTF fit and so on.

Does this mean there that we are being limited by our hardware, thus preventing achieving higher resolution, or have we potentially reached the dataset limits?

Happy to provide further information.

I thank you all beforehand for your attention.

No - it is not doing quite what it seems, it is a confusing message. See here for details:

Thank you for your response and for pointing out an earlier discussion, which I overlooked. Things are clearer now.

Just to add on to the above, final resolutions will never be artificially limited by memory availability, unless you explicitly set smaller box sizes in jobs that permit doing so.

Best,
Michael