Does MRCC compute CCSDT/CCSDTQ effectively on multiple cores?

  • marcingronowski
  • Topic Author
  • Offline
  • New Member
  • New Member
More
1 year 15 hours ago #1421 by marcingronowski
Dear MRCC experts,

What is your experience with CCSDT or CCSDTQ calculation run via OpenMPI parallelization? Usually, I use 4-6 cores, and MRCC works fine. I'm facing the calculation with many electrons and a big basis set. I expect to use ~40 CPUs and ~300 GB of RAM. I'd like to know if MRCC can handle more than 10 CPUs effectively.
Any suggestions are welcome.

All the best
Marcin

Please Log in or Create an account to join the conversation.

  • nagypeter
  • Offline
  • Premium Member
  • Premium Member
  • MRCC developer
More
1 year 12 hours ago #1422 by nagypeter
Dear Marcin,

There is indeed only little info in Sect II.L of the MRCC release paper about
CCSDT and CCSDTQ: doi.org/10.1063/1.5142048 .

CCSDT and CCSDTQ are very data intensive so you cannot expect good MPI parallel efficiency for tens of CPUs, but you can certainly use more than 4-6 cores (1 CPU?). You can try to help the MPI scaling by providing local, separate, and fast I/O for the MPI tasks.

The MPI scaling of CCSDT(Q) is good as shown in the paper.

Alternatively, for many electrons and big basis sets, it is often not needed to treat all of them at this very high theoretical level. You may want to try our FNO, LNO, active orbital or embedding approaches in MRCC to accelerate the computation or use a composite method for basis set correction, etc.

All the best,
Peter

Please Log in or Create an account to join the conversation.

Time to create page: 0.037 seconds
Powered by Kunena Forum