The good news is that there are some nice improvements in performance on the G5 platform.
But, as you can see, my Intel MacBook Pro laptop is significantly faster than both the Absoft and IBM compiled versions of FDS5_RC6 on the Dual G5.
I ran the room_fire.fds Verification case long enough to develop a trend then stopped it.
I don't expect different results if I run the case to completion.
== G5 ABSOFT Results at Iteration 100 ==
Time Step 100 July 2, 2007 17:14:45
----------------------------------------------
CPU/step: 1.154 s, Total CPU: 1.88 min
Time step: 0.03403 s, Total time: 3.91 s
Max CFL number: 0.85E+00 at ( 26, 52, 12)
Max divergence: 0.25E+01 at ( 26, 52, 8)
Min divergence: -0.16E+01 at ( 26, 52, 12)
No. of Lagrangian Particles: 50
Total Heat Release Rate: 10.214 kW
Radiation Loss to Boundaries: 3.622 kW
== G5 XLF Results at Iteration 100 ==
Time Step 100 July 2, 2007 16:57:27
----------------------------------------------
CPU/step: 0.629 s, Total CPU: 1.02 min
Time step: 0.03394 s, Total time: 3.91 s
Max CFL number: 0.85E+00 at ( 26, 52, 12)
Max divergence: 0.25E+01 at ( 26, 52, 8)
Min divergence: -0.15E+01 at ( 26, 52, 12)
No. of Lagrangian Particles: 52
Total Heat Release Rate: 10.173 kW
Radiation Loss to Boundaries: 3.611 kW
== Intel MacBook Pro Results at Iteration 100 ==
Time Step 100 July 2, 2007 16:33:51
----------------------------------------------
CPU/step: 0.303 s, Total CPU: 29.13 s
Time step: 0.03375 s, Total time: 3.88 s
Max CFL number: 0.84E+00 at ( 26, 52, 12)
Max divergence: 0.25E+01 at ( 26, 52, 8)
Min divergence: -0.14E+01 at ( 26, 52, 12)
No. of Lagrangian Particles: 47
Total Heat Release Rate: 10.238 kW
Radiation Loss to Boundaries: 3.631 kW
So far it looks like the XLF compiled version of RC6 is almost twice as fast as the Absoft version. But, the Intel compiler on the Intel Mac spanks them both with twice the speed of the XLF version and almost four times the speed of the Absoft version.
At least this confirms that I need to switch to XLF for G5 builds and it also means that Glenn should be able to build G4/G5 versions of Smokeview again. :)
-Bryan
Time step: 0.03403 s, Total time: 3.91 s
Max CFL number: 0.85E+00 at ( 26, 52, 12)
Max divergence: 0.25E+01 at ( 26, 52, 8)
Min divergence: -0.16E+01 at ( 26, 52, 12)
No. of Lagrangian Particles: 50
Total Heat Release Rate: 10.214 kW
Radiation Loss to Boundaries: 3.622 kW
== G5 XLF Results at Iteration 100 ==
Time Step 100 July 2, 2007 16:57:27
------------------------------
CPU/step: 0.629 s, Total CPU: 1.02 min
Time step: 0.03394 s, Total time: 3.91 s
Max CFL number: 0.85E+00 at ( 26, 52, 12)
Max divergence: 0.25E+01 at ( 26, 52, 8)
Min divergence: -0.15E+01 at ( 26, 52, 12)
No. of Lagrangian Particles: 52
Total Heat Release Rate: 10.173 kW
Radiation Loss to Boundaries: 3.611 kW
== Intel MacBook Pro Results at Iteration 100 ==
Time Step 100 July 2, 2007 16:33:51
------------------------------
CPU/step: 0.303 s, Total CPU: 29.13 s
Time step: 0.03375 s, Total time: 3.88 s
Max CFL number: 0.84E+00 at ( 26, 52, 12)
Max divergence: 0.25E+01 at ( 26, 52, 8)
Min divergence: -0.14E+01 at ( 26, 52, 12)
No. of Lagrangian Particles: 47
Total Heat Release Rate: 10.238 kW
Radiation Loss to Boundaries: 3.631 kW
So far it looks like the XLF compiled version of RC6 is almost twice as fast as the Absoft version. But, the Intel compiler on the Intel Mac spanks them both with twice the speed of the XLF version and almost four times the speed of the Absoft version.
At least this confirms that I need to switch to XLF for G5 builds and it also means that Glenn should be able to build G4/G5 versions of Smokeview again. :)
-Bryan
