[CP2K:4289] AIMD with CP2K (estimation of CPU time)
ibet... at epcc.ed.ac.uk
Tue Feb 12 13:45:52 CET 2013
I support the CP2K installation on HECToR. While the answer to your question is really - do some benchmarks on the machine to enable a good estimate - I can give you some numbers that may help. (Note that you can get HECToR class 2a allocations immediately for benchmarking calculations to prepare for a full grant application - email me off list if you have any questions about how to do this).
Based on some crude assumptions:
* 0.5 fs timestep = 200,000 MD steps
* Assume each MD step requires 4 SCF iterations for convergence (a big assumption!)
* Assume your system costs approximately the same as 4096 waters (typical DFT parameters see cp2k/tests/QS/benchmark/H20-4096.inp)
* Then each MD step will take ~30 mins on 2048 cores of HECToR (based on prior benchmark I ran)
* Total cost of the simulation is ~205 million core hours = 820 million AU i.e. very, very expensive
Of course any of the above assumptions may be incorrect, so please do test, but this sounds like a very big system for long-timescale AIMD. Using a semi-empirical method would likely be somewhat cheaper but I don't have any timings to estimate how much.
Project Manager, EPCC
Email: ibet... at epcc.ed.ac.uk
Tel/Fax: +44 (0)131 650 5201/6555
Mob: +44 (0)7598317015
Addr: 2404 JCMB, The King's Buildings, Mayfield Road, Edinburgh, EH9 3JZ
On 11 Feb 2013, at 13:25, Rasoul wrote:
> Dear All,
> I'm trying to estimate required time for doing of ab initio MD simulation using CP2K since I need this info for making an application of supercomputer time (HECToR).
> My interest system are about 10,000 atoms (hydrocarbon molecules) and detail info for HECToR machine is as following:
> 1. HECToR is a connection of 2816 nodes.
> 2. Each node has 32 cores and 32 GBytes of RAM, which it is shared by the 32 cores.
> 3. Each core is of type AMD Opteron 2.3GHz Interlagos (The latest AMD Opteron family).
> 4. So total cores are: 32 cores per node * 2816 nodes = 90112 cores.
> 5. And total distributed RAM memory is: 32 GBytes per node * 2816 nodes = 90122 GBytes distributed RAM Memory.
> 6. The speed connection between nodes has an MPI point-to-point bandwidth of 5 GB/s or more (the best available on the market).
> I just wanted to know how long CPU time would be necessary for performing ab initio MD simulations (with PM6 and DFT), approximately.
> Will it change significantly, if I can use FMO-MD technique?
> Time of simulation is considered to be about 100 ps.
> I would be appreciated if anybody help me for this estimation.
> Best wishes
> Rasoul Nasiri, Ph.D
> Post-Doc Research Fellow
> Sir Harry Ricardo Laboratories
> School of Computing, Engineering and Mathematics
> University of Brighton
> Cockcroft Building, room C214, Lewes Road
> Brighton BN2 4GJ, UNITED KINGDOM
> Tel: +44(0)1273 642330
> e-mail : R.Na... at brighton.ac.uk
> Website: http://www.brighton.ac.uk/shrl/staff/nasiri.ph
> You received this message because you are subscribed to the Google Groups "cp2k" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns... at googlegroups.com.
> To post to this group, send email to cp... at googlegroups.com.
> Visit this group at http://groups.google.com/group/cp2k?hl=en.
> For more options, visit https://groups.google.com/groups/opt_out.
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.
More information about the CP2K-user