My testing methodology was as follows:

I booted up the test system, went to the motherboard manufacturer's website, and downloaded the latest hardware monitoring utility for the motherboard in question. I installed the utility, and rebooted.

I then let the system idle for a minute or two, allowing Windows to finish up its voodoo boot sequence. I started the hardware monitoring utility, and took note of the readings for the +3.3V, +5V, and +12V rails. The voltage readings that I made note of were the most stable ones that I could see. For instance, if a certain reading stayed at 11.86V throughout the entire 3-minute test, but only dipped to 11.84V for a second, I made note of 11.86V. In the case of certain readings that alternated between something like 11.90V and 11.91V with every read cycle of the multimeter or hardware monitor, I made note of 11.905, which I think is a fair estimate.

The multimeter, which was set to the 0 - 20 V DC range, was then connected to my choice of power leads from the power supply. The claimed accuracy on the multimeter is 1% which means in a 20V range it can be off by 0.2V

With all of the systems that used the OCZ power supply, I used the additional BTX connector plug for testing the +3.3V, +5V, and +12V rails, as it conveniently provided them along with a GND.

Test System #3, which utilized the AOpen power supply, I tested by making use of a regular Molex connector for the +5V and +12V, and connected the multimeter leads directly to an orange and black wire on the ATX plug itself for the +3.3V results. This was done by pushing the sharp points into the spacing in between the wire and plastic cells of the ATX connector, allowing the connector to be plugged into the motherboard.

Finally, Test System #5 was tested in a similar fashion to that of #3, with the exception that I used an AUX connector to test the +3.3V rail rather than the ATX connector.

After getting the whole set of results for the idle system down, I fired up a game (DOOM 3 on Test System #3, Half-Life 2 on the others), turned up the graphic detail, and ran a demo loop. Both games were run at a resolution of 1152x864, windowed, with the Windows desktop resolution at 1280x960. This allowed me to run the game while also watching the readings of the hardware monitoring utility in the corner of the screen.

I ran the demos a total of six times on each test system, every time watching one of the rails on either the multimeter or the hardware monitor. As mentioned above, I then recorded the most stable readings I could pick out.