PSU voltage with multimeter

My PSU started making some noises even when the fans were stopped. I also had USB problems (two USB devices burned) so I checked the PSU voltages with a multimeter.

I removed the PSU and connected it to a fan to see if it was on. I connected the green wire to a black wire and tested the voltages of the color cables on my 20 rails cable.

I noticed that my yellow (12V) rail was only showing 11.07V. Testing on the other yellow cables had the same result.

As this is bigger than the 5% margin of error I thought that I had a faulty PSU.

However after connecting the PSU to the motherboard, the values went up and my 12V rail went to 11.95.

Am I doing something wrong? Should I always test the PSU when it is connected to the PC?


Solution 1:

Working properly without load is something not every ATX PSU does. To enhance the efficiency on real loads (or in case of bad PSU, on max load...) manufacturers do compromises, and unless you get really highy quality (and expensive) PSU, these compromises might mean the PSU does not work very well under 15% of max load.

On the other hand, the main problem with testing with motherboard is that load issued by the motherboard is very far from constant or controllable; thus your best bet is to add some 20-30W worth predictable +12V load when testing the PSU without motherboard.

And yes, I understand this can be hard to accomplish. In general, swapping in known-to-work PSU from other machine and seeing if that fixes the problem is the easiest solution, and the one I personally use. I even have a spare quality PSU just for this purpose..

Also, multimeter approach has the problem of missing small sags in power. For example, when CPU goes from zero load to full load, which happens quite a lot, the power draw from PSU to motherboard power circuitry can triple in milliseconds. With bad PSU voltages can drop here way below the nominal for few milliseconds, with is impossible to notice with multimeter but can still make the machine reboot or behave abnormally.

Oscilloscope would notice that, but they tend to cost real money.