[petsc-dev] Fwd: How badly Elemental beats ScaLAPACK

[Matthew Knepley]

Since this graphic is from a paper, I suppose I can reshare.

  Matt

---------- Forwarded message ----------
From: Robert van de Geijn <rvdg at cs.utexas.edu>
Date: Sun, Sep 23, 2012 at 11:16 PM
Subject: How badly Elemental beats ScaLAPACK
To: FLAME mailing list <flame_cs at cs.utexas.edu>


Folks,

We recently used this graph in a paper:




DxTer here means Elemental, but run through Bryan's DxTer tool to generate
optimized implementations that always match and sometimes surpass Jack's
hand coded optimizations.  For both packages, extensive experiments with
block sizes were performed.  This is on 8192 cores.

Bryan may correct me on some of my claims, since he knows what experiments
were really run.

Pretty impressive. imho.

Jack, you may want to send this to the Elemental forum.

Robert





-- 
What most experimenters take for granted before they begin their
experiments is infinitely more interesting than any results to which their
experiments lead.
-- Norbert Wiener
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20120924/1a89a802/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: PastedGraphic-11.pdf
Type: application/pdf
Size: 33092 bytes
Desc: not available
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20120924/1a89a802/attachment.pdf>