streams

From: Leslie A. Barnes (lbarnes@hal.com)
Date: Thu May 22 1997 - 10:34:11 CDT


 John,

 I have some new results for Streams performance on our HAL 375 and
385 machines - but, I am not sure who to send them to. You?

 Also, there seems to be an error in the Streams benchmark, where you
compute the RMS time. You divide by ntimes, but you sum the squares
from 2 to ntimes (ntimes-1). So, if the times are all the same, or
close for each repetition, the RMS time is less than the Min time (see
results below).

 Thanks

 Les Barnes

Fortran compiler options: -Kfast,V8PLUS,GREG,eval,gs,prefetch -Kunroll=10 -dn

 1) HALStation Model 385 (161 Mhz), 768Mb memory

Function Rate (MB/s) RMS time Min time Max time
Assignment: 460.9474 0.1320 0.1388 0.1396
Scaling : 476.0017 0.1279 0.1345 0.1355
Summing : 521.7283 0.1751 0.1840 0.1864
SAXPYing : 523.2701 0.1745 0.1835 0.1849

 2) HALStation Model 375 (141 Mhz), 768Mb memory

Function Rate (MB/s) RMS time Min time Max time
Assignment: 420.6846 0.1444 0.1521 0.1527
Scaling : 428.4301 0.1427 0.1494 0.1547
Summing : 467.8021 0.1953 0.2052 0.2078
SAXPYing : 473.8022 0.1925 0.2026 0.2036



This archive was generated by hypermail 2b29 : Tue Apr 18 2000 - 05:23:07 CDT