9
$\begingroup$

Similar to this question which asks for each answer to provide one software that can do CASSCF: I am interested in software that can do integrals, SCF (Hartree-Fock), and the integral transformation, with more than one CPU. Do any of these programs allow also for use with more than one node?

The software can be commercial or open-source.

$\endgroup$
5
  • 3
    $\begingroup$ While not a perfect indication, pretty much any of the programs in this list that can do HF with GPUs can also do it with multiple nodes. There will be a few false positives/negatives, but its a good starting point. $\endgroup$
    – Tyberius
    May 16 '20 at 22:47
  • 1
    $\begingroup$ @Tyberius which programs in that list can do HF with GPUs? The list tells us if a program can do HF, and if it has some GPU capability, but not that it can do SCF with GPUs: some of them use GPUs for MD, which GPUs are known to be especially good for. Also, when you say that the ones that can use GPUs also can use multiple nodes, do you mean multiple cores? Many of these programs are multi-threaded using OpenMP but not with MPI. $\endgroup$ May 16 '20 at 23:06
  • 2
    $\begingroup$ Here's a list of GPU accelerated molecular modeling software: en.m.wikipedia.org/wiki/Molecular_modeling_on_GPUs $\endgroup$ May 17 '20 at 2:25
  • $\begingroup$ The MRCC manual says: "Hybrid MPI + OpenMP parallelization of the Hartree-Fock and Kohn-Sham SCF, the hand-optimized CCSD(T), and the high-order CC implementations." Peter Nagy is a member here and a developer of MRCC, so I'll ask if he's able to answer. $\endgroup$ May 21 '20 at 17:21
  • $\begingroup$ Today I just noticed that the installation instructions of MRChem say "To build the code with hybrid MPI/OpenMP parallelization...." meaning that MRChem may be able to do SCF with Hybrid MPI + OpenMP. I'll ask Luca Frediani if he might like to answer. $\endgroup$ May 22 '20 at 5:40
13
$\begingroup$

Psi4 and PySCF

Psi4 and PySCF are free, open source programs that can do both parallel integrals and SCF. MPI parallellization is quite rare in quantum chemistry codes, since the algorithms don't tend to parallellize easily across nodes..

MPQC

If you really want to do massively parallel calculations, then the Massively Parallel Quantum Chemistry (MPQC) package is your choice. This is a program aimed for molecules and periodic solids that can also run massively parallel coupled-cluster calculations.

ORCA

The ORCA program is MPI parallel, allowing it to run in parallel both on shared-memory and distributed-memory setups.

$\endgroup$
2
  • 2
    $\begingroup$ Unfortunately both package only do multi threading calculations they does not have massively parallel capability. $\endgroup$
    – QMlab
    Jun 15 '20 at 5:34
  • 1
    $\begingroup$ @QMlab .. which was not even the question. roflmao $\endgroup$ Jun 15 '20 at 9:07
7
$\begingroup$

MRCC

There are more programs with parallel SCF capability, but it is quite hard to scale SCF (with HF exchange) to large number of cores. MRCC indeed has a recent hybrid multi-node (MPI) & multi-threaded (OpenMP) parallel HF/hybrid DFT code relying on an integral direct density fitting algorithm, but you should not expect great scaling above few hundred cores, depending on system size. GPUs are not supported there. It is however free for academics, open source, and for large systems (cca. 100+ atoms in 3D) the HF exchange can be further accelerated via local approximations. Largest (local) HF calc I performed with it had about 1000 atoms and 50000 atomic orbitals.

More details:

https://www.mrcc.hu/index.php

J. Chem. Phys. 152, 074107 (2020); https://doi.org/10.1063/1.5142048

$\endgroup$
0
4
$\begingroup$

NWChem

NWChem is an ideal package for massively parallel needs.

JCP recently published a collection of articles focused on the special topic "Electronic Structure Software". Not only NWCHem, GAMESS, and ACES4, but various other packages have the capability to do massively parallel calculations.

$\endgroup$
1
  • $\begingroup$ Would you list some “various other packages”? $\endgroup$
    – Greg
    Oct 3 '20 at 3:16
2
$\begingroup$

GAMESS

Most parts of GAMESS are parallelised with MPI (including SCF), and can run on multiple CPUs. The parallel SCF runs entirely on replicated memory, however other parallel algorithms such as MP2, CCSD etc. use distributed memory. GAMESS is free, and you can also access the source code. Some parts of the GAMESS are also parallelised with OpenMP like the recently implemented RI-CCSD module. GAMESS can run on multiple nodes.

$\endgroup$
2
  • 1
    $\begingroup$ +1 But please follow this format: mattermodeling.meta.stackexchange.com/q/125/5, the other answer by Susi Lehtola was written before that policy was decided! $\endgroup$ Feb 5 at 21:03
  • 1
    $\begingroup$ @NikeDattani Thanks! I didn't know about this policy. $\endgroup$
    – S R Maiti
    Feb 5 at 21:07

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.