Mark Hahn wrote:
numbers. It seems that the fortran version of the program with
pathscale 2.3 (the numbers I reported above) reports different numbers
than the C version of the program with pathscale 2.3:
my numbers were from a random stream binary I had around,
so I don't really even remember which compiler for sure
(probably pathscale 2.2.1, may well have been C rather than fortran).
recompiling with a fresh copy of the source gives me 9.4 GB/s
on my dual-socket single-core opteron 2.6 cluster, and 8.8
or so on the aforementioned 2x2 (275).
Interesting. I just ran the same stream binary I was using on the
Woodcrest systems on a few Opterons here, all of which should be running
DDR400:
1) dual-socket, single-core 2.4 GHz Opteron
Function Rate (MB/s) Avg time Min time Max time
Copy: 7776.2299 0.0041 0.0041 0.0042
Scale: 7808.8043 0.0041 0.0041 0.0041
Add: 7363.0030 0.0065 0.0065 0.0065
Triad: 7342.8621 0.0065 0.0065 0.0066
2) quad-socket, single-core 2.6 GHz Opteron
Function Rate (MB/s) Avg time Min time Max time
Copy: 18464.4006 0.0017 0.0017 0.0017
Scale: 17838.6135 0.0018 0.0018 0.0018
Add: 17937.1518 0.0027 0.0027 0.0027
Triad: 17910.0251 0.0027 0.0027 0.0027
3) quad-socket, dual-core 2.4 GHz Opteron
Function Rate (MB/s) Avg time Min time Max time
Copy: 16048.9929 0.0020 0.0020 0.0020
Scale: 16087.4659 0.0020 0.0020 0.0020
Add: 15650.3881 0.0031 0.0031 0.0031
Triad: 15574.1156 0.0031 0.0031 0.0031
So I think my binary is probably OK. I used "pathcc -Ofast -mp stream.c
-static -o stream_omp.x86_64".
Thanks,
--
Jason Holmes
_______________________________________________
Beowulf mailing list, Beowulf@beowulf.org
To change your subscription (digest mode or unsubscribe) visit
http://www.beowulf.org/mailman/listinfo/beowulf
