I've been doing benchmark tests on my new Dual 1.25GHz Power Mac G4.
Here are my SETI@home results.
The notation xx:yy:zz means xx hours, yy minutes, zz seconds. I averaged times from multiple work units to get more accurate results. One work unit took an hour less than all the others (sometimes SETI determines that a work unit is useless and aborts its computations early), so I omitted that run from the results.
1. The minimum "Average CPU time per work unit" was achieved running the CLI (command line interface) version of SETI, with the energy saver features turned off and no other processes running.
CPU time: 5:40:48 (baseline)
Elapsed time: 5:40:50 (baseline)
CPU utilization: 100% (baseline, meaning one processor fully occupied)
Explanation: This is the fastest that this Mac can run. Nothing but Mac OS X 10.2, the Terminal application, and SETI were running. Only one CPU was used. Mac OS overhead is extremely minimal (a few seconds in over 5 hours).
2. CLI version, energy saver on:
CPU time: 5:48:09 (2.2% over baseline)
Elapsed time: 5:49:30 (2.5% over baseline)
CPU utilization: 98%
Explanation: A little more Mac OS overhead, but essentially the same.
3. Screensaver (graphical) version, energy saver on, graphics display timeout after 30 minutes:
CPU time: 5:52:11 (3.3% over baseline)
Elapsed time: not measured
CPU utilization: 97%
Explanation: This is a typical way you might use SETI@home. The graphical display (for 30 minutes) adds a little overhead, but not much, since one CPU can handle the graphics while the other is still number crunching the work unit.
4. CLI version, energy saver off, with 2 copies of SETI@home running at once:
CPU time: 11:47:57 total
5:53:59 per unit (3.9% over baseline)
Elapsed time: 5:56:55 total (4.7% over 1 unit baseline, but for 2 units!)
2:58:28 per unit
CPU utilization: 191%
Explanation: I got twice the amount of number crunching done (2 units completed) in only a few minutes more elapsed time than when I did 1 unit. The second CPU is in almost constant use. Good for you, Dual Mac!
5. CLI version, energy saver off, with 3 copies of SETI@home running at once:
CPU time: 17:45:20 total
5:55:07 per unit (4.2% over baseline)
Elapsed time: 8:56:37 total (57% over 1 unit baseline, but for 3 units)
2:58:52 per unit
CPU utilization: 191%
Explanation: You can run 3 (or more) units at essentially the same rate as 2 units, i.e., nnn units will take nnn/2 as much elapsed time as 2 units. Both CPUs will be almost fully occupied. Process switching adds neglible overhead.
Bottom line:
* To benefit from dual CPUs, run 2 or more SETI@home processes at the same time in CLI mode.
* Energy saver settings have minimal effect.
* With one SETI@home processing running, it is OK to leave the graphics display on. But you might as well set it to time out if you won't be watching it.
* Having dual processors lets you complete units at twice the elapsed time rate of a single processor of the same speed by running two or more SETI processes at once. In this case, set the screen to go dark (no graphics) if you use the screensaver version, since both CPUs will already be occupied.
* The measure of "Average CPU time per work unit" (as shown on the SETI summary pages) will be slightly less when running multiple processes at once. The results when you run dual processes on dual processors will appear the same as if you run two separate single-processor computers that have the same CPU speed.
And that's enough benchmarking for one week!
Here are my SETI@home results.
The notation xx:yy:zz means xx hours, yy minutes, zz seconds. I averaged times from multiple work units to get more accurate results. One work unit took an hour less than all the others (sometimes SETI determines that a work unit is useless and aborts its computations early), so I omitted that run from the results.
1. The minimum "Average CPU time per work unit" was achieved running the CLI (command line interface) version of SETI, with the energy saver features turned off and no other processes running.
CPU time: 5:40:48 (baseline)
Elapsed time: 5:40:50 (baseline)
CPU utilization: 100% (baseline, meaning one processor fully occupied)
Explanation: This is the fastest that this Mac can run. Nothing but Mac OS X 10.2, the Terminal application, and SETI were running. Only one CPU was used. Mac OS overhead is extremely minimal (a few seconds in over 5 hours).
2. CLI version, energy saver on:
CPU time: 5:48:09 (2.2% over baseline)
Elapsed time: 5:49:30 (2.5% over baseline)
CPU utilization: 98%
Explanation: A little more Mac OS overhead, but essentially the same.
3. Screensaver (graphical) version, energy saver on, graphics display timeout after 30 minutes:
CPU time: 5:52:11 (3.3% over baseline)
Elapsed time: not measured
CPU utilization: 97%
Explanation: This is a typical way you might use SETI@home. The graphical display (for 30 minutes) adds a little overhead, but not much, since one CPU can handle the graphics while the other is still number crunching the work unit.
4. CLI version, energy saver off, with 2 copies of SETI@home running at once:
CPU time: 11:47:57 total
5:53:59 per unit (3.9% over baseline)
Elapsed time: 5:56:55 total (4.7% over 1 unit baseline, but for 2 units!)
2:58:28 per unit
CPU utilization: 191%
Explanation: I got twice the amount of number crunching done (2 units completed) in only a few minutes more elapsed time than when I did 1 unit. The second CPU is in almost constant use. Good for you, Dual Mac!
5. CLI version, energy saver off, with 3 copies of SETI@home running at once:
CPU time: 17:45:20 total
5:55:07 per unit (4.2% over baseline)
Elapsed time: 8:56:37 total (57% over 1 unit baseline, but for 3 units)
2:58:52 per unit
CPU utilization: 191%
Explanation: You can run 3 (or more) units at essentially the same rate as 2 units, i.e., nnn units will take nnn/2 as much elapsed time as 2 units. Both CPUs will be almost fully occupied. Process switching adds neglible overhead.
Bottom line:
* To benefit from dual CPUs, run 2 or more SETI@home processes at the same time in CLI mode.
* Energy saver settings have minimal effect.
* With one SETI@home processing running, it is OK to leave the graphics display on. But you might as well set it to time out if you won't be watching it.
* Having dual processors lets you complete units at twice the elapsed time rate of a single processor of the same speed by running two or more SETI processes at once. In this case, set the screen to go dark (no graphics) if you use the screensaver version, since both CPUs will already be occupied.
* The measure of "Average CPU time per work unit" (as shown on the SETI summary pages) will be slightly less when running multiple processes at once. The results when you run dual processes on dual processors will appear the same as if you run two separate single-processor computers that have the same CPU speed.
And that's enough benchmarking for one week!
