Dear Dr. McCalpin,
This morning I caught the PChallenge in a completely idle state
and obtained the following results with stream_wall.f:
***************************************************************
LOG.1:
******
--------------------------------------
Double precision appears to have 16 digits of accuracy
Assuming 8 bytes per DOUBLEPRECISION word
--------------------------------------
Timing calibration ; time = 63.35089206695557 hundredths of a second
Increase the size of the arrays if this is <30
and your clock precision is =<1/100 second
---------------------------------------------------
Function Rate (MB/s) RMS time Min time Max time
Assignment: 124.5029 0.2584 0.2570 0.2608
Scaling : 122.9766 0.2614 0.2602 0.2637
Summing : 132.7125 0.3629 0.3617 0.3661
SAXPYing : 132.9470 0.3621 0.3610 0.3635
Sum of a is : 1.1533007812800417E+18
Sum of b is : 2.3066015625365702E+17
Sum of c is : 3.0754687500612557E+17
LOG.2:
******
--------------------------------------
Double precision appears to have 16 digits of accuracy
Assuming 8 bytes per DOUBLEPRECISION word
--------------------------------------
Timing calibration ; time = 104.3899059295654 hundredths of a second
Increase the size of the arrays if this is <30
and your clock precision is =<1/100 second
---------------------------------------------------
Function Rate (MB/s) RMS time Min time Max time
Assignment: 236.3071 0.1401 0.1354 0.1621
Scaling : 234.2262 0.1395 0.1366 0.1496
Summing : 253.7440 0.1918 0.1892 0.1993
SAXPYing : 254.4098 0.1900 0.1887 0.1932
Sum of a is : 1.1533007812480837E+18
Sum of b is : 2.3066015625131405E+17
Sum of c is : 3.0754687500425120E+17
LOG.3:
******
--------------------------------------
Double precision appears to have 16 digits of accuracy
Assuming 8 bytes per DOUBLEPRECISION word
--------------------------------------
Timing calibration ; time = 110.2923035621643 hundredths of a second
Increase the size of the arrays if this is <30
and your clock precision is =<1/100 second
---------------------------------------------------
Function Rate (MB/s) RMS time Min time Max time
Assignment: 350.1632 0.0967 0.0914 0.1084
Scaling : 334.0989 0.0974 0.0958 0.1020
Summing : 363.7465 0.1341 0.1320 0.1412
SAXPYing : 369.0663 0.1376 0.1301 0.1618
Sum of a is : 1.1533007812480940E+18
Sum of b is : 2.3066015624897107E+17
Sum of c is : 3.0754687500237683E+17
LOG.4:
******
--------------------------------------
Double precision appears to have 16 digits of accuracy
Assuming 8 bytes per DOUBLEPRECISION word
--------------------------------------
Timing calibration ; time = 116.0651922225952 hundredths of a second
Increase the size of the arrays if this is <30
and your clock precision is =<1/100 second
---------------------------------------------------
Function Rate (MB/s) RMS time Min time Max time
Assignment: 461.1683 0.0764 0.0694 0.0866
Scaling : 440.0444 0.0762 0.0727 0.0839
Summing : 473.8356 0.1051 0.1013 0.1116
SAXPYing : 480.3171 0.1069 0.0999 0.1252
Sum of a is : 1.1533007812481252E+18
Sum of b is : 2.3066015624862480E+17
Sum of c is : 3.0754687500050240E+17
LOG.5:
******
--------------------------------------
Double precision appears to have 16 digits of accuracy
Assuming 8 bytes per DOUBLEPRECISION word
--------------------------------------
Timing calibration ; time = 117.0679926872253 hundredths of a second
Increase the size of the arrays if this is <30
and your clock precision is =<1/100 second
---------------------------------------------------
Function Rate (MB/s) RMS time Min time Max time
Assignment: 589.6233 0.0613 0.0543 0.0773
Scaling : 554.3244 0.0614 0.0577 0.0656
Summing : 582.1430 0.0845 0.0825 0.0887
SAXPYing : 594.9209 0.0863 0.0807 0.0929
Sum of a is : 1.1533007812481564E+18
Sum of b is : 2.3066015624878099E+17
Sum of c is : 3.0754687500000000E+17
LOG.6:
******
--------------------------------------
Double precision appears to have 16 digits of accuracy
Assuming 8 bytes per DOUBLEPRECISION word
--------------------------------------
Timing calibration ; time = 110.3252053260803 hundredths of a second
Increase the size of the arrays if this is <30
and your clock precision is =<1/100 second
---------------------------------------------------
Function Rate (MB/s) RMS time Min time Max time
Assignment: 749.0476 0.0515 0.0427 0.0638
Scaling : 673.6130 0.0500 0.0475 0.0556
Summing : 675.2572 0.0736 0.0711 0.0779
SAXPYing : 718.0683 0.0720 0.0668 0.0806
Sum of a is : 1.1533007812481876E+18
Sum of b is : 2.3066015624893718E+17
Sum of c is : 3.0754687500000006E+17
LOG.7:
******
--------------------------------------
Double precision appears to have 16 digits of accuracy
Assuming 8 bytes per DOUBLEPRECISION word
--------------------------------------
Timing calibration ; time = 117.9732084274292 hundredths of a second
Increase the size of the arrays if this is <30
and your clock precision is =<1/100 second
---------------------------------------------------
Function Rate (MB/s) RMS time Min time Max time
Assignment: 927.3727 0.0395 0.0345 0.0588
Scaling : 819.0000 0.0441 0.0391 0.0616
Summing : 816.1448 0.0649 0.0588 0.0898
SAXPYing : 845.4731 0.0608 0.0568 0.0792
Sum of a is : 1.1533007812482189E+18
Sum of b is : 2.3066015624909341E+17
Sum of c is : 3.0754687500000000E+17
LOG.8:
******
--------------------------------------
Double precision appears to have 16 digits of accuracy
Assuming 8 bytes per DOUBLEPRECISION word
--------------------------------------
Timing calibration ; time = 126.2425065040588 hundredths of a second
Increase the size of the arrays if this is <30
and your clock precision is =<1/100 second
---------------------------------------------------
Function Rate (MB/s) RMS time Min time Max time
Assignment: 567.5264 0.0684 0.0564 0.0823
Scaling : 495.9317 0.1224 0.0645 0.2393
Summing : 547.6889 0.1373 0.0876 0.2351
SAXPYing : 847.4269 0.1048 0.0566 0.1611
Sum of a is : 1.1533007812482501E+18
Sum of b is : 2.3066015624924957E+17
Sum of c is : 3.0754687500000000E+17
LOG.9:
******
--------------------------------------
Double precision appears to have 16 digits of accuracy
Assuming 8 bytes per DOUBLEPRECISION word
--------------------------------------
Timing calibration ; time = 129.0195107460022 hundredths of a second
Increase the size of the arrays if this is <30
and your clock precision is =<1/100 second
---------------------------------------------------
Function Rate (MB/s) RMS time Min time Max time
Assignment: 577.1490 0.0734 0.0554 0.0843
Scaling : 562.3711 0.0810 0.0569 0.1109
Summing : 631.2465 0.1036 0.0760 0.1523
SAXPYing : 696.0249 0.0918 0.0690 0.1177
Sum of a is : 1.1533007812482813E+18
Sum of b is : 2.3066015624940576E+17
Sum of c is : 3.0754687500000000E+17
LOG.10:
*******
--------------------------------------
Double precision appears to have 16 digits of accuracy
Assuming 8 bytes per DOUBLEPRECISION word
--------------------------------------
Timing calibration ; time = 130.4086923599243 hundredths of a second
Increase the size of the arrays if this is <30
and your clock precision is =<1/100 second
---------------------------------------------------
Function Rate (MB/s) RMS time Min time Max time
Assignment: 579.6678 0.0753 0.0552 0.1055
Scaling : 658.3819 0.0748 0.0486 0.1030
Summing : 693.2018 0.1073 0.0692 0.1440
SAXPYing : 703.9871 0.0966 0.0682 0.1199
Sum of a is : 1.1533007812483126E+18
Sum of b is : 2.3066015624956195E+17
Sum of c is : 3.0754687500000000E+17
***********************************************************
I find two aspects very interesting:
-- Without interference by other users the machine scales better
than you could determine so far.
-- If you create the extra threads beyond the number of processors
yourself, the performance goes down (cf. i = 8, 9 and 10 for
our 7 processors).
Comments?
Could you also comment the dependence of the numerical results for
the sums on the number of processors? Isn't that a very strange
and puzzling behaviour?
Best regards,
Wolf Weyrich.
This archive was generated by hypermail 2b29 : Tue Apr 18 2000 - 05:23:04 CDT