RT-TDDFT optimal computing resources

[Carles]

Dear All:

I am trying to estimate the optimal computer resources to apply for a project in a supercomputing center. I have not had the opportunity however to do any previous tests in such systems.

I am doing rt-tddft using single x-ray and trains of x-ray pulses, so that the total propagation can go from 50 fs to 500 fs, with time steps of 0.1 as. 

My question is, how many cores in a distributed memory system are the expected/most optimal for rt-tddft calculations with NWChem, considering modern processors (such as e.g. 2.6 GHz Xeon processors)?. Also, which execution time is expected with such a number of cores for a typical XAS calculation, considering for example a propagation of 50000 steps (tmax=200, dt=0.005 a.u.)?. 

An approximate estimation or example would be very helpful! 

Thank you,

Best regards,

Carles

[Niri Govind]

Dear Carles,

This can be estimated from a ground state calculation itself. Typically, the cost per time step in a RT run is roughly ~2x the cost of a step in a ground state SCF cycle. In addition, RT calculations can also be restarted from the last time step. 

Best,

-Niri

Niri Govind

PNNL

niri....@pnnl.gov

--
You received this message because you are subscribed to the Google Groups "NWChem Forum" group.
To unsubscribe from this group and stop receiving emails from it, send an email to nwchem-forum...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/nwchem-forum/278137c0-95e0-4b29-8e2f-c1fe3f6c6d3en%40googlegroups.com.

[Carles]

Dear Niri,

thanks a lot for the hint. But this will change depending on the number of cpu's that are used in the simulation, and this is what I have no opportunity to test, since I have no access yet to a supercomputing center, I am applying for it. So, it is 1000 cpu's per calculation, or less is better, I mean, which is the most common number to use in a parallel NWChem calculation.

Best wishes,

Carles 

El dia divendres, 12 de març de 2021 a les 15:37:32 UTC+1, niri....@gmail.com va escriure:

[Niri Govind]

Dear Carles,

We have benchmarks on our github link, which will allow you to estimate your needs based on what I mentioned earlier. 

Please use the zeolite fragment benchmark at the bottom of the page. It's the most recent one.

https://github.com/nwchemgit/nwchem/wiki/Benchmarks

Best regards,

-Niri

Niri Govind

PNNL

niri....@pnnl.gov

To view this discussion on the web visit https://groups.google.com/d/msgid/nwchem-forum/d64b8dd9-5f28-46b9-8d80-445602b2c47bn%40googlegroups.com.

[Carles]

Dear Niri,

thanks a lot! now I understand and can make the estimation.

Best regards,

Carles 

El dia dilluns, 15 de març de 2021 a les 5:39:54 UTC+1, niri....@gmail.com va escriure: