On Oct 27, 2009, at 11:11 , Luca Scrucca wrote:
Dear Simon,
first of all I would like to thank you for providing the Mac binary
package for the latest version of R.
It works ok on my two years old MBP with OSX 10.5.8.
As mentioned in the web page the GUI installed are two, 32-bit GUI
(R.app) and 64-bit GUI (R64.app).
I noticed that running a simple loop in the two GUI I got different
values. For example, having defined the function
fun <- function(n)
{ for(i in 1:n)
median(rnorm(100000))
}
I obtained the following typical timings:
[R.app GUI 1.30 (5511) i386-apple-darwin9.8.0]
> system.time(fun(1000))
user system elapsed
30.131 2.460 32.506
[R.app GUI 1.30 (5511) x86_64-apple-darwin9.8.0]
> system.time(fun(1000))
user system elapsed
24.927 2.151 26.957
The second is always smaller, although my MBP is not 64 bit. Can you
comment on that? I'm just curious.
Your MBP *is* 64-bit capable otherwise you couldn't run the 64-bit GUI
at all (there are very few Apple machines that had the early ICD/ICS
without 64-bit support - they are rather rare, all other machines
starting with IC2D are 64-bit capable).
On Intel machines the 64-bit instruction set is different form the 32-
bit instruction set since it didn't have to follow the legacy baggage
from 16-bit times of x86. As a result code using the 64-bit
instruction set can be more efficient than its 32-bit (ix86)
counterpart. This is counterbalanced by the fact that all pointers
have double the size in 64-bit and thus you have to use double the
memory and associated speed penalty, so depending on the task (yours
is trivial enough to not involve too many pointer operations) the net
result is speed-up or slow-down.
I also noticed that running R64.app the X11 environment is also
started (the graphical device is the default, i.e. quartz). This
does not happen with R.app.
At a first glance I see no difference between those and neither of
them starts X11 on my machine ...
Aside, there is a broken link in the web page http://cran.r-project.org/bin/macosx/
, namely the link to the tools directory for R-2.10.0.pkg (there is
a missing 's' in the URL).
Thanks, that was reported and fixed already.
Cheers,
Simon
_______________________________________________
R-SIG-Mac mailing list
R-SIG-Mac@stat.math.ethz.ch
https://stat.ethz.ch/mailman/listinfo/r-sig-mac
