On 14/01/2021 11:32, Jean Thioulouse wrote:
Is there a large speed difference for linear algebra operations between the
Rosetta x86_64 builds and the native arm64 builds ?
On a Mac mini M1 8/256 I get only a slight speed increase (15% faster) with the
arm version (experimental R-devel build).
Speed differences were covered in an earlier report so please look in
the list archive. Speeds are really dependent (for both Intel and ARM)
on the load on the machine and the size of the task: typically the ARM
version is 40% faster, with a very wide range.
Thanks
Jean
Le 14 janv. 2021 à 08:41, Prof Brian Ripley <rip...@stats.ox.ac.uk> a écrit :
For native builds - nothing to report for x86_64 builds running under Rosetta
which almost all the time 'just work' (and work fast).
The goal remains to release a native binary distribution with R 4.1.0 ca April:
all but the intrepid are advised to use x86_64 until then.
- There is an experimental R-devel build of the R framework at
https://mac.r-project.org/ . That page reports 'failed' but usually the
framework is complete.
- R.app will need to be partially re-written as it uses Objective C features
which are no longer supported in Xcode 12. (I guess this is why the page is
reporting a failure.)
- There are most parts of a toolchain and a Fortran compiler at
https://mac.r-project.org/libs-arm64/ (not the earlier versions at libs-arm).
These install into /opt/R/arm64, and the R-admin manual has been re-written to
reflect that and point out some pitfalls.
- Binary packages are planned but the vast majority of packages install from
source. I currently have 240 CRAN check failures with 35 failing to install
and 98 others requiring those (or BioC ones). If you are interested in a
particular package, https://www.stats.ox.ac.uk/pub/bdr/M1mac/ has recent logs
for those failing directly, listed as 'additional issues' on their CRAN check
pages.
--
Brian D. Ripley, rip...@stats.ox.ac.uk
Emeritus Professor of Applied Statistics, University of Oxford
_______________________________________________
R-SIG-Mac mailing list
R-SIG-Mac@r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-mac
--
Brian D. Ripley, rip...@stats.ox.ac.uk
Emeritus Professor of Applied Statistics, University of Oxford
_______________________________________________
R-SIG-Mac mailing list
R-SIG-Mac@r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-mac
