QuantumATK Forum
QuantumATK => General Questions and Answers => Topic started by: liximao on August 16, 2010, 03:11
-
I want to know how long I can save when I calculate one 800-atom system by ATK 10.8 with 8 cores instead of 4 cores.
40% or 50% or others? ???
Thanks! :D
-
It depends strongly on what kind of system the 800 atoms is.
It might be as much 50% time saved in time, but if you tell what kind of system it is, then I can give you a better estimate.
-
Thanks! ;D
1. How about the graphene?
2. How about the Ag double electrode device with a small molecule?
-
1) If the graphene has many k-points, then it is 50% of the calculation time.
2) If you skip the equvialent bulk calculation for the molecular junction and you have a bias, then it is close 50% also.
-
The performance improvement in parallel depends on very many factors. In your case, it matters a lot also whether the cores are distributed among several MPI nodes or all on the same machine. In the latter case you would also have a strong memory overhead, and in fact probably there is no way you can run 800 atoms on a single node, unless you have 32 Gb of RAM or so.
-
I get it, thanks a lot!! ;D
-
We just published a more detailed guide on running ATK in parallel: http://quantumwise.com/documents/tutorials