https://bugzilla.redhat.com/show_bug.cgi?id=830869
Dave Love <[email protected]> changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |[email protected] --- Comment #9 from Dave Love <[email protected]> --- I was going to comment on this. I can't give a formal review, and I'm about to go on holiday, but there's an amended spec at https://loveshack.fedorapeople.org/review/hpl.spec in case it's useful. It builds on at least epel-6-x86_64 and fedora-20-x86_64 and passes rpmlint. Obviously a serial version makes no sense as it's a distributed benchmark. If you use the packaged atlas rather than openblas on x86_64 -- at least on sandybridge -- the results are a factor of 3-ish(?) pessimistic. I can't remember the actual figure. Openblas/openmpi give you essentially the same results as the Intel proprietary stuff -- hooray for free software. A devel package doesn't make sense for the benchmark, surely. [Fedora badly needs the Debian model for BLAS to avoid this sort of issue.] -- You are receiving this mail because: You are on the CC list for the bug. You are always notified about changes to this product and component _______________________________________________ package-review mailing list [email protected] https://admin.fedoraproject.org/mailman/listinfo/package-review
