[R-SIG-Mac] Does 64-bit R use multiple cores?
simon.urbanek at r-project.org
Tue Nov 25 19:17:15 CET 2008
On Nov 25, 2008, at 11:28 , John C. Tull wrote:
> I thought Altivec was a ppc-only optimization. I'm pretty sure this
> does not apply to Intel systems, but maybe I'm wrong.
The Altivec part was incorrect but the multithreaded BLAS part was
correct. Replace Altivec with your favorite three-letter code for
Intel CPUs ;). [Not that Altivec matters in any way, since it supports
spfp only which R doesn't use ...].
For more parallelization, see snow and pnmath by Luke Tierney.
> On Nov 24, 2008, at 10:23 PM, Kasper Daniel Hansen wrote:
>> It depends on the computations you want to do.
>> R for Mac OS X uses Altivec which is a multithreaded version of
>> BLAS and lapack. What this means is that for any computations
>> involving matrix algebra, you do utilize multiple cores. Depending
>> on what you are trying to compute and how that computation is
>> implemented, this may prove to be just what you want or it may have
>> little to no effect.
>> On Nov 25, 2008, at 15:24 , John C. Tull wrote:
>>> Dear R-mac Users:
>>> I'm wanting to leverage an 8-core Intel Mac Pro for all of its
>>> computational glory. Looking over the mail list, it appears that
>>> the conversation on 64-bit builds of R has been about gaining
>>> access to memory beyond the 32-bit limit of 3.5 GB and not about
>>> multiple processors/cores.
>>> Does running 64-bit R only take advantage of one processor core at
>>> a time like the standard 32-bit R, or does it do symmetric
>>> multiprocessing (is this the right terminology?). If no, is there
>>> any possibility of doing this using X-grid or otherwise that
>>> someone can detail? I do have 10.5 Server if I have to go down
>>> that road.
>>> John C. Tull, Ph.D.
>>> Conservation Director
>>> Nevada Wilderness Project
>>> 8550 White Fir Street
>>> Reno, NV 89523 USA
>>> 775.746.7851 (office)
>>> 775.224.2947 (mobile)
>>> R-SIG-Mac mailing list
>>> R-SIG-Mac at stat.math.ethz.ch
> R-SIG-Mac mailing list
> R-SIG-Mac at stat.math.ethz.ch
More information about the R-SIG-Mac