Thanx... + ATHLON

[Kostas Gewrgiou]

 Pantos apo diafora mail stin beowulf mailing list fenetai oti se polous
kodikes o K7 exei pano apo tin diplasia taxytita apo enan PIII sta idia Mhz

gia paradeimga:
http://beowulf.gsfc.nasa.gov/listarchives/beowulf/1999/11/0120.html
http://beowulf.gsfc.nasa.gov/listarchives/beowulf/1999/11/0122.html
http://beowulf.gsfc.nasa.gov/listarchives/beowulf/1999/11/0147.html
http://beowulf.gsfc.nasa.gov/listarchives/beowulf/1999/11/0194.html

apo to teleutaio:
-----
 /src/PROJ1/ion44 mpirun n0-19 sse1
Start: -- Server: using 19 nodes. Time: 16 42 32
K7: -- Received field 129. time: 17 6 45 - 24.13min
K7: -- Received field 129.5 time: 17 30 58 - 24.13min
PIII at 500 -- Received field 128. time: 17 46 12 - 63.40min
PIII at 500 -- Received field 128.5 time: 17 46 12 - 63.40min
PIII at 450 -- Received field 120. time: 17 53 43 - 71.11min
PIII at 450 -- Received field 127.5 time: 17 53 43 - 71.11min
PIII at 450 -- Received field 125.5 time: 17 53 43 - 71.11min

The K7 is 2.5 times faster than a PIII at this particular problem... It is a
small inner loop, so I believe this is because of the 128KB level one cache
of the Athlon. For a large problem I have seen a more realistic 30% increase
in performance when going to the Athlon. But still, those 2.5 times is a very
real performance boost, and not a useless benchmark!
-----

Kostas Gewrgiou


--
====================================================================
Gia boithia (h na diagrafhte) e-mail sto majordomo at hellug.gr
Ta archives tis listas einai sto http://lists.hellug.gr/archives
prin steilete kapoia erothsh psakte mipos exei hdh apanththei.
Gia opoiodipote problima stilte e-mail ston owner-linux-greek-users at hellug.gr
====================================================================