On Tue, 18 Dec 2007, Prof Brian Ripley wrote: > On Tue, 18 Dec 2007, S Ellison wrote: > >> Hiding in the windows faq is the observation that "R's computation is >> single-threaded, and so it cannot use more than one CPU". So multi-core >> should make no difference other than allowing R to run with less >> interruption from other tasks. That is often a significant advantage, >> though. > > Yes, but that is Windows-specific. > > On most other platforms you can benefit from using a multi-threaded BLAS, > such as ATLAS, ACML or Dr Goto's. The speedup for linear algebra can be > substantial (although sometimes it will slow things down). Luke Tierney > has an experimental package to make use of parallel threads for some basic > R computations which may appear in R 2.7.0.
There are two experimental packages available in http://www.stat.uiowa.edu/~luke/R/experimental: pnmath, based on OpenMP, and pnmath0, based on basic pthreads. These packages provide parallelized versions of many of the R vectorized math functions. The README files in these packages give more details. OpenMP is I think the way we want to go in the longer term; there are a few configu issues that need sorting out and so in the interim a non OpenMP version might be useful. Best, luke > > It should be possible to use a multi-threaded BLAS under Windows, but I > know no one who has done it. There is a viable pthreads implementation > for Windows, and I've tested Luke's experimental package using it. > > Some compilers' runtimes will be able to use parallel threads for other > tasks. Since all the examples I am aware of are expensive commercial > compilers, I suspect R will make limited use of them. (In particular, > base R does not use the Fortran 9x vector operations at which many of > these features are targeted: we probably would if we routinely used such > compilers.) > > I've had dual-CPU desktops for more than ten years. Given how little > speedup you are likely to get via parallel processing (only under ideal > conditions do the optimized BLASes run >1.5x faster using two CPUs), the > most effective way to make use of multiple CPUs has been to run multiple > jobs: I typically run 3-4 at once to keep the CPUs fully used. > > One way to run multiple R processes to cooperate on a single task is to > use a package such as snow to distribute the load. > > > >>>>> Andrew Perrin <[EMAIL PROTECTED]> 18/12/2007 01:13 >>> >> On Mon, 17 Dec 2007, Kitty Lee wrote: >> >>> Dear R-users, >>> >>> I use R to run spatial stuff and it takes up a lot of ram. Runs can >> take hours or days. I am thinking of getting a new desktop. Can R take >> advantage of the dual-core system? >>> >>> I have a dual-core computer at work. But it seems that right now R is >> using only one processor. >>> >>> The new computers feature quad core with 3GB of RAM. Can R take >> advantage of the 4 chips? Or am I better off getting a dual core with >> faster processing speed per chip? >>> >>> Thanks! Any advice would be really appreciated! >>> >>> K. >> >> If I have my information right, R will use dual- or quad-cores if it's >> doing two (or four) things at once. The second core will help a little >> bit >> insofar as whatever else your machine is doing won't interfere with the >> one core on which it's running, but generally things that take a single >> thread will remain on a single core. >> >> As for RAM, if you're doing memory-bound work you should certainly be >> using a 64-bit machine and OS so you can utilize the larger memory >> space. > > They only have 3GB of RAM, which 32-bit OSes can address. The benefits > really come with more than that. > > -- Luke Tierney Chair, Statistics and Actuarial Science Ralph E. Wareham Professor of Mathematical Sciences University of Iowa Phone: 319-335-3386 Department of Statistics and Fax: 319-335-3017 Actuarial Science 241 Schaeffer Hall email: [EMAIL PROTECTED] Iowa City, IA 52242 WWW: http://www.stat.uiowa.edu ______________________________________________ [email protected] mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
