[Bug 1112864] Review Request: elpa - High-performance library for parallel solution of eigenvalue problems

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



https://bugzilla.redhat.com/show_bug.cgi?id=1112864



--- Comment #19 from Dave Love <d.love@xxxxxxxxxxxxxxx> ---
(In reply to Dominik 'Rathann' Mierzejewski from comment #15)

> It is build against atlas now. openblas doesn't support ARM yet. Do you
> think it's worth it to build against openblas on x86/x86_64 and atlas on ARM?

I don't know what difference it makes in this case, but I'd expect openblas to
be maybe twice as fast on sandybridge for DGEMM-based things, at least with the
packaged versions.  However, when I tried to compare atlas recently, I got
strange, non-reproducible results that I haven't tried to understand.

If the tests in the ELPA source make a reasonable benchmark, I could maybe
try, but I think they didn't run for long.

cp2k built with openblas, fftw, and openmpi actually seemed to out-perform a
version
built with all-intel on a similar cluster on the one case I had a report for,
but
I haven't tried very serious comparisons, and I've just discovered that the
Fedora
fftw isn't avx enabled.  I'd forgotten to file an issue suggesting it should
use
openblas.

-- 
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
package-review@xxxxxxxxxxxxxxxxxxxxxxx
https://admin.fedoraproject.org/mailman/listinfo/package-review





[Index of Archives]     [Fedora Legacy]     [Fedora Desktop]     [Fedora SELinux]     [Yosemite News]     [KDE Users]     [Fedora Tools]