[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
## Re: [Help-gsl] Maximum size of matrices

**From**: |
Analabha Roy |

**Subject**: |
Re: [Help-gsl] Maximum size of matrices |

**Date**: |
Fri, 27 Jan 2006 05:17:37 -0600 |

Well. Assuming ur datatypes are "floats" then 128 bits per float (right?),
with 16000*3000 floats means abt 1 GIG of RAM
+ ull need space in RAM for temporary work areas for the diagonalization
process that's typically goes as N^2 or some such power law, so waaay too
much I think (my numerical analysis fundas are somewhat sketchy, but I think
if ur diagonalizing matrices using Householder-type Transformations+Jacobi
rotations then space requirements go as some power law of the matrix order)
But I think that most distributive teragrids can take the load (if that's
where ur doing it).
If ur using "doubles" then out of the question in standard boxen.
On 1/26/06, Emmanuel Eckard <address@hidden> wrote:
>
>* Greetings !*
>
>* Is there any known maximum size of matrices for the library to still be*
>* useful ? I am trying thinks with matrices of the order of magnitude of 16*
>* 000*
>* per 3000, and it fails to allocate the blocks. Is this dependant on the*
>* memory of the computer, or is this an intrinsic limitation of the library*
>* ?*
>
>* Thank you very much in advance.*
>* --*
>* Emmanuel Eckard*
>* Ingénieur Physicien de l'École Polytechnique de Lausanne*
>* Laboratoire d'Intelligence Artificielle, EPFL*
>* LIA/IC 1014 Ecublens, Suisse*
>* +41 21 693 66 97*
>
>
>* _______________________________________________*
>* Help-gsl mailing list*
>* address@hidden*
>* http://lists.gnu.org/mailman/listinfo/help-gsl*
>