Acceleration in CCSDT calculation

Dear MRCC developers,

I am running CCSDT calculations on small transition metal complexes using MRCC2019. I found that one iteration of CCSDT can take up to 8 hours for a system of 121 basis function. I know that CCSDT calculations are expensive but I am still wondering if I did something wrong to make it this slow. For example, if I could allocate more memory to reduce the file I/O on the disk? Below I attach the essential inputs of mine as well as the output file. Thanks in advance for your help.

Best,
Chenru

======input=======
calc=ccsdt
ovirt=off
rest=2
cialg=direct2
core=frozen
scftype=rohf
rohftype=semicanonical
dens=0
mem=64GB
usedist=1
mult=2
charge=2
===================

=======output========
Executing goldstone...

Generation of MRCC equations in terms of H and T...
Generation of antisymmetrized Goldstone diagrams...
Number of diagrams in T^1 equations: 15
Number of diagrams in T^2 equations: 37
Number of diagrams in T^3 equations: 47
Translation of diagrams to factorized equations...
Number of restricted diagrams in T^1 equations: 30
Number of restricted diagrams in T^2 equations: 126
Number of restricted diagrams in T^3 equations: 274
Optimizing intermediate calculation...
Number of floating-point operations per iteration step: 2.2488E+14
Probable CPU time per iteration step (hours): 2248.77
Required memory (Mbytes): 21353.1
Number of intermediates: 210
Number of intermediates to be stored: 100
Length of intermediate file (Mbytes): 49946.4

************************ 2020-02-16 12:47:24 *************************
Executing xmrcc...

**********************************************************************
MRCC( 3 ) calculation


Allocation of****** Mbytes of memory...
Number of spinorbitals: 246
Number of alpha electrons: 16
Number of beta electrons: 15
Number of active particles: 108
Number of active holes: 0
Spin multiplicity: 2
z-component of spin: 0.5
Spatial symmetry: 1
Convergence criterion: 1.0E-06
Construction of occupation graphs...
Number of 0 -fold excitations: 1
Number of 1 -fold excitations: 3332
Number of 2 -fold excitations: 4060650
Number of 3 -fold excitations: 2345138260
Total number of configurations: 2349202243
Calculation of coupling coefficients...
Length of intermediate file (Mbytes): 36278.4

======================================================================

Memory requirements /Mbyte/:
Minimal Optimal
Real*8: 36145.9237 36145.9237
Integer: 58.1929
Total: 36204.1166 36204.1166


************************ 2020-02-16 12:47:35 *************************
Executing mrcc...

**********************************************************************
CCSDT calculation


OpenMP parallel version is running.
Number of CPUs: 16
Allocation of36145.9 Mbytes of memory...
Number of spinorbitals: 246
Number of alpha electrons: 16
Number of beta electrons: 15
Number of active particles: 108
Number of active holes: 0
Spin multiplicity: 2
z-component of spin: 0.5
Spatial symmetry: 1
Convergence criterion: 1.0E-06
Construction of occupation graphs...
Number of 0-fold excitations: 1
Number of 1-fold excitations: 3332
Number of 2-fold excitations: 4060650
Number of 3-fold excitations: 2345138260
Total number of determinants: 2349202243
Calculation of coupling coefficients...
Initial cluster amplitudes are read from unit 16.
Length of intermediate file (Mbytes): 36278.4
Reading integral list from unit 55...
Sorting integrals...
Sorting integrals...
Sorting integrals...
Sorting integrals...
Energy of reference determinant [au]: -1557.177195858343
Calculation of MP denominators...

Starting CC iteration...
==========================

Dear Chenru,
This is an expensive calculation as your molecule has no symmetry (at least, mrcc recognizes it as C1; you should provide more symmetrical coordinates if your molecule do have some symmetry). The only thing you can do to accelerate the calculation is to run it on a machine with larger memory.

Dear Mihaly,

Thank you so much for your speedy reply. I will try playing around with the symmetry. In terms of using a machine of larger memory, I am wondering that now it seems only using 36204.1166 MB memory (I allocated 64 GB) and writing 36278.4 MB intermediate file. Do you think it will be accelerated if I use a machine with >73GB memory? Sorry I do not know whether I read the out file correctly and will be appreciated for your further help.

Best,
Chenru

Dear Chenru,
I can tell you only on the basis of the output. Please post it.

Here is the out file. Thank you so much!

Chenru

At least 145GB would be required for optimal performance.