[Pw_forum] ifc/linux allocation of more than 2 GB dynamical memory

[Giovanni Cangiani]

Dear Nicola,
	I hope I am wrong but I think the memory is limited
to 2Gb due to the 32bit addressing (don't ask me why 2Gb and 
not 8Gb=2^33) . 
AFAIK there is no way to solve it unless you buy 64 bit CPU 
like intel itanium or AMD opteron.
I am presently testing the program on an opteron based 
cluster with Portland compiler (there is no i386_64 
version of the intel compiler) but I still have quite a 
lot troubles.

	giovanni


	as far as I know the limit of 2Gb is due to the
32bit addressing. 
On Tue, 2004-05-04 at 13:57, Nicola Marzari wrote:
> Dear All,
> 
> we have recently realized that there is an issue running large jobs
> on Linux machines, and I was wondering if anyone had some suggestions.
> 
> This is the problem: if a fortran code running under Linux
> (Red Hat 9) and using ifc (7.1) and mkl (6.0) tries to allocate
> a number of arrays that total over 2 GB, it ends with a runtime error.
> 
> This happens no matter the amount of dynamical memory physically
> available (we tested this on a machine having ~4GB RAM), and happens
> exactly at the point in which the array that is being allocated
> overshoots 2GB .
> 
> The Intel webpage has a comment somewhere mentioning this limitation,
> suggesting it is related to where the kernel allocates space for the
> linked libraries, but we have had no success trying their suggestions
> (e.g. compiling -static).
> 
> In practice, this limitation is even more serious, since a running
> code (e.g. CP) that allocates ~< 2GB will only use, in runtime, an
> average value that is smaller (~1.25GB).
> 
> Any helpful suggestions would be very welcome,
> 
> 
> 			nicola
> 
> ---------------------------------------------------------------------
> Prof Nicola Marzari   Department of Materials Science and Engineering
> 13-5066   MIT   77 Massachusetts Avenue   Cambridge MA 02139-4307 USA
> tel 617.4522758  fax 617.2586534  marzari at mit.edu  http://nnn.mit.edu
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum