Anyone have information on latest LSU beowulf?

Craig Tierney ctierney at hpti.com
Thu Oct 10 11:54:00 EDT 2002


On Wed, Oct 09, 2002 at 12:39:03PM -0400, Patrick Geoffray wrote:
> On Wed, 2002-10-09 at 12:07, Craig Tierney wrote:
> 
> > This says that NB=40 is good for the PIII which has a larger
> > L1 cache than a P4 (16k data vs. 8k).  NB should be a multiple
> > of 32 for the P4.  I would like to try it out on a PIII, I would
> > think that 44 is a better value based on cache size.  I tried
> > all these tricks on an Alpha was 16k L1 cache and found 88 (44*2)
> > best.
> 
> Which value is used by ATLAS ? Stick with it. It may be 40 or 32 or
> whatever, but it will be the granularity of DGEMM used in HPL.
> If 32 is the block size used by ATLAS, try ro tun with NB as 32, 64, 96
> and 128.

It seems that I was wrong about the NB size.  I thought I had tested it, but
not on 500 processors.  Here are my results from 2 runs (250 dual xeon 2.2Ghz).
These were on different systems.

In my ATLAS summary, NB=48 is being used.

Run 1:
W01R2L6       125000   160    20    25            1329.60          9.793e+02
W01R2L6       125000    80    20    25            1343.96          9.689e+02
W01R2L6       125000    96    20    25            1372.81          9.485e+02
W01R2L6       125000   192    20    25            1412.41          9.219e+02
W01R2L6       125000    64    20    25            1415.00          9.202e+02
W01R2L6       125000   128    20    25            1575.68          8.264e+02

Run 2:
W01R2L6       125000   160    20    25            1345.82          9.675e+02
W01R2L6       125000    80    20    25            1387.60          9.384e+02
W01R2L6       125000    96    20    25            1415.53          9.199e+02
W01R2L6       125000    64    20    25            1422.12          9.156e+02
W01R2L6       125000   192    20    25            1442.42          9.027e+02
W01R2L6       125000   128    20    25            1596.93          8.154e+02

The NB=192 case failed to give the correct result on both runs.

80 and 160 give the best results on this problem.  I will continue to test with
these numbers.

Craig




> 
> Patrick
> -- 
> ----------------------------------------------------------
> |   Patrick Geoffray, Ph.D.      patrick at myri.com 
> |   Myricom, Inc.                http://www.myri.com
> |   Cell:  865-389-8852          685 Emory Valley Rd (B)
> |   Phone: 626-821-5555          Oak Ridge, TN 37830
> ----------------------------------------------------------

-- 
Craig Tierney (ctierney at hpti.com)
_______________________________________________
Beowulf mailing list, Beowulf at beowulf.org
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf



More information about the Beowulf mailing list