On Friday, 6 July 2012 18:18:54 UTC+8, Volker Braun wrote: > > A 20x speedup seems to be a bit too good to be true. It could be that the > binary has a botchered (mis-tuned) ATLAS install.
What is tuned on one host can be mis-tuned on another. Nothing strange about this. Perhaps a binary install of Sage should come with Atlas "dumbed down", or we should advise the users to re-build Atlas spkg after the installation. > I guess its possible to have that kind of speedup, but only with very > specific problems that immensely profit from SSE3/4/AVX. > > I understand that the problem in question is very heavy on linear algebra. Dima > > On Friday, July 6, 2012 9:50:37 AM UTC+1, Jan Groenewald wrote: >> >> Hi >> >> On 5 July 2012 12:40, Jan Groenewald <j...@aims.ac.za> wrote: >> >>> As an interesting aside, >>> >>> This problem, when running succesfully, on Ubuntu 11.04, took this long >>> >>> from upstream sage binary: 264 seconds >>> from sage source compiled: 11 seconds >> >> >> I'd like to demonstrate to students why this difference exists. >> Can someone suggest a short easily readable program that shows >> why certain CPU flags/capabilities can cause such a speedup >> between generically compiled binaries for minimum 64bit specs, >> and a CPU specific from source build. >> >> Regards, >> Jan >> -- >> .~. >> /V\ Jan Groenewald >> /( )\ www.aims.ac.za >> ^^-^^ >> >> -- -- 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
