Do you have a (good) idea of where diminishing returns cuts in ? I imagine the number of cores (threads) doesn't help much above a fairly small number due to sequential dependencies ? More memory is always "nice to have" but does it really help HERE ?
Did you set the atlas_architecture environment variable to cut down on tuning time ? On Jan 20, 8:25 am, Jeroen Demeyer <jdeme...@cage.ugent.be> wrote: > Some observations from building sage-4.8 on a very fast system (32 > cores, 512G RAM) using "make -j64 -l32". > > First of all, apply the patch from #12329 to prune many unneeded > dependencies of the Sage library (needs_review by the way...): > > Total time for "make build" was 27 minutes, total time for "make doc" 14 > minutes, for a grand total of 41 minutes. > > The critical path to build all of Sage is as follows (every package > depends on the one just above it): > > (base) > patch > iconv > libgpg_error > libgcrypt > opencdk > gnutls > python > fortran > lapack > atlas > r + rpy2 > > The 10 slowest packages to build (slowest first): > r > atlas > scipy > polybori > maxima > ntl > singular > ecl > sage > libm4ri -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
