Hi, There are some keywords that reduce the memory usage in parallel calculations, like ON.LowerMemory and a couple of others, check with the manual. Besides these, what you can do is increase the number of nodes (pretty obvious), shrink the basis set (obvious, too), use basis orbitals with a narrower localiation range and a lower O(N) cutoff distance (to sparsen the matrices and reduce their dimensions), and finally, reduce the grid cutoff. That's all I can think of by now.
Regards, Vasilii 2007/1/10, Cherry Y. Yates <[EMAIL PROTECTED]>:
I also found something wired. I was optimize a nanostructure, and if I do a grep "max": siesta: iscf Eharris(eV) E_KS(eV) FreeEng(eV) dDmax Ef(eV) Max 0.077525 Max 0.077525 constrained * Maximum dynamic memory allocated = 188 MB siesta: iscf Eharris(eV) E_KS(eV) FreeEng(eV) dDmax Ef(eV) Max 0.063488 Max 0.063488 constrained * Maximum dynamic memory allocated = 188 MB siesta: iscf Eharris(eV) E_KS(eV) FreeEng(eV) dDmax Ef(eV) Max 0.167946 Max 0.167946 constrained * Maximum dynamic memory allocated = 188 MB siesta: iscf Eharris(eV) E_KS(eV) FreeEng(eV) dDmax Ef(eV) Max 0.067086 Max 0.067086 constrained * Maximum dynamic memory allocated = 2146 MB siesta: iscf Eharris(eV) E_KS(eV) FreeEng(eV) dDmax Ef(eV) Max 0.055845 Max 0.055845 constrained * Maximum dynamic memory allocated = 2146 MB siesta: iscf Eharris(eV) E_KS(eV) FreeEng(eV) dDmax Ef(eV) Max 0.044563 Max 0.044563 constrained * Maximum dynamic memory allocated = 2146 MB * Maximum dynamic memory allocated : Node 0 = 2146 MB * Maximum memory occured during rdiag As you see, the maximum dynamic memory allocated jumps from 188 MB to 2146 MB. Is it all right? Thanks, Cherry --- "Cherry Y. Yates" <[EMAIL PROTECTED]> wrote: > Dear SIESTA developers > > Actually I tested the memory usage with the SZP > basis > against the SZ basis. For the bulk Si, SZP > calculations require twice as much memory as the SZ > calculations. However for the nanowire (2600 atoms), > the memory usage jumps from 3.4 GB to more than > 14GB! > Is it normal? Please give me some hints! > > Thanks, > > Cherry > > > --- "Cherry Y. Yates" <[EMAIL PROTECTED]> wrote: > > > Dear SIESTA developers, > > > > I am running SIESTA for some silicon nanowires, > the > > memory of each node in my cluster is 4GB. when the > > atom is less than 1000, the vmem and mem is > roughly > > the same, for example: > > > > resources_used.mem = 2213804kb > > resources_used.vmem = 2258976kb > > > > Then after that, the vmem increases quickly, e.g., > > for > > a 1800-atom system, > > > > resources_used.mem = 3894052kb > > resources_used.vmem = 6427792kb > > > > It looks like SIESTA knows 4GB is the limit for my > > machine, and it uses swap. Is it right? Do you > know > > how to cut some memory usages to fit my machine? > It > > really slows down a lot... > > > > Thanks, > > > > Cherry > > > > > > __________________________________________________ > > Do You Yahoo!? > > Tired of spam? Yahoo! Mail has the best spam > > protection around > > http://mail.yahoo.com > > > > > __________________________________________________ > Do You Yahoo!? > Tired of spam? Yahoo! Mail has the best spam > protection around > http://mail.yahoo.com > ____________________________________________________________________________________ Any questions? Get answers on any topic at www.Answers.yahoo.com. Try it now.
