Probable CPU time per iteration step

Dear Igor,
Technically the I/O could be parallelized with an MPI library called Parallel-HDF5 but you won't really notice the difference unless you use multiple nodes (not just more cores but more nodes). The fort.* files would also have to be in HDF5 format. This isn't yet implemented in MRCC.

You might benefit from using node local storage or switching to a solid-state drive.

With best wishes!
Nike

Dear Nike,

Thank you for sharing your experience!

To be honest, I am not entirely interested in energies. I am interested in the calculation of electron density using CC methods. You have raised a theme of extrapolating schemes, and as I know, they are not available for densities: the results are significantly different in the different basis sets.
And also, about using CCSDT(Q): for this method, derivatives are not available, so the CC density can not be computed.

By the way, Ar2(36e, 336o) with R=5.0A, CCSDT(no frozen core)/ACCQ has one CC iteration step per hour on my machine. Really cool result!

Best regards,
Igor

Dear Igor,
Now that you have said this, I can understand your problem better. Your original post in this thread didn't say anything about needing densities, and I only checked to see if you'd asked a different question elsewhere because Mihaly somehow knew you were using 160 orbitals! I found your question about the neon dimer with 160 orbitals but did not look to deep into what you wrote there, so I didn't see you were interested in densities.

With best wishes!
Nike