Uwe Ligges <[EMAIL PROTECTED]> writes: > Alex Gracian wrote: > > > Hi, > > Any plans to realse in Cran binaries for R base & packages, for > > Windows > > using 64bit, now that the AMD 64 is coming out? > > No plans, AFAIK. > Are you volunteering to compile R base (there may be need for some > adjustments in the code, as well as in the Makefiles) for CRAN? > The maintainer would need both, an AMD 64 machine, and a 64-bit > Windows in order to compile. > I'm volunteering to make the packages, if you'll buy an AMD 64 machine > + 64-bit Windows for my office. ;-)
You may regret that statement, once you realize that you also need the entire 64bit toolchain... > > It seems a perfect match for cost effective high performane number > > crunching?? > > I don't know whether it will be much faster to use the 64bit mode, > because the internal structure of x86 processors for number crunching > already has been 80 bit for floating point arithmetics. Additionally, > you will need more RAM for the same problems which slows down some > calculations. > In principle, you can extend the 2GB limit of the R process, though. Yes. And we do actually bump into that when we try to do something sufficiently "stupid". Recent case: attempting to fit a polr() to weekly data for 2500 persons for one year each, including 2500 person levels, giving a design matrix of ~2.5GB. Some times you do want to try that sort of thing by brute force and ignorance (of course, raising the memory limit might just have revealed further obstacles...) I would say that the perfect match currently is the SuSE Linux product for AMD-64. I don't think people quite realize how much extra work goes into maintaining the Windows platform (about as much as the Unix/Linux variants combined, I'd say), and we basically only do it because of its market penetration. -- O__ ---- Peter Dalgaard Blegdamsvej 3 c/ /'_ --- Dept. of Biostatistics 2200 Cph. N (*) \(*) -- University of Copenhagen Denmark Ph: (+45) 35327918 ~~~~~~~~~~ - ([EMAIL PROTECTED]) FAX: (+45) 35327907 ______________________________________________ [EMAIL PROTECTED] mailing list https://www.stat.math.ethz.ch/mailman/listinfo/r-devel
