[R-SIG-Mac] How to Speed up R on the G5

[Jean Thioulouse]

At 2:38 -0600 8/02/05, Douglas Bates wrote:
>Regarding the question of whether veclib is parallelized, I'm 
>reasonably sure that it is.  A few months ago I compiled a version 
>of R using veclib for the BLAS and ran a simple timing test on a 
>matrix crossproduct.
>
>mm <- matrix(rnorm(9e6), nc = 3000)
>system.time(crossprod(mm), gc = TRUE)
>system.time(crossprod(mm), gc = TRUE)
>system.time(crossprod(mm), gc = TRUE)
>
>I found on a dual processor system that the elapsed time was about 
>half the user CPU time.

Yes indeed. Here is what I get on a dual 2.5 GHz G5 :

> mm <- matrix(rnorm(9e6), nc = 3000)
> system.time(crossprod(mm), gc = TRUE)
[1] 5.36 0.23 3.21 0.00 0.00
> system.time(crossprod(mm), gc = TRUE)
[1] 5.43 0.26 3.22 0.00 0.00
> system.time(crossprod(mm), gc = TRUE)
[1] 5.51 0.27 3.25 0.00 0.00

The fans hardly have enough time to speed up !

Jean

-- 
Jean Thioulouse - Labo de Biometrie et Biologie Evolutive, UMR CNRS 5558
Universite Lyon 1,  43 Boulevard du 11 Novembre 1918, Batiment G. Mendel
69622 Villeurbanne Cedex,  France.          Tel/Fax : (33) 4 72 43 27 56                       
                 http://pbil.univ-lyon1.fr/JTHome.html