All of this information should be available on the side of a PSU. So if you want to make the best purchasing decision, try and get a look at this. Etailers such as overclockers.co.uk tend to put such information on display should you need to study them.
The final thing to consider is efficiency. Power supplies inevitably use a certain amount of energy in the conversion process - this energy is lost as heat. So even though your PC may only be using 500W, the power draw at the plug socket may be more like 600W. Obviously this wasted energy wants to be kept to a minimum - especially with electricity bills ever increasing and the plight of the planet on everyone's minds.
The ATX12V 2.2 standard states that all power supplies must be at least 70 per cent efficient (computer = 700W, plug socket = 1000W), which isn't bad but it's still a fairly large chunk of energy to be throwing away. Any half decent supply nowadays will exceed 80 per cent efficiency and it's perfectly possible to reach 90+ per cent. However, this is generally considered to be prohibitively expensive.
How We Tested
Although the actual load placed on a power supply during general use will vary constantly, the only fair and consistent way to test them is to use specific constant loads. Unfortunately, the equipment required to provide these loads is very specialist and consequently very expensive so it's not something we have in the office. Instead we gathered up our power supplies and headed out to the UK headquarters of power supply manufacturer Enermax.
The test procedure is simultaneously very simple and yet quite complicated. Essentially the criteria we look for in a power supply are that it conforms to the 12VATX standard at a number of different load levels (i.e. current draw) and that on the way it maintains efficiency, stays reasonably cool and quiet, and doesn't, well, blow up...
To conform to ATX12V standard the voltage of each rail must stay within the following bounds (below), while providing the correct number of amps to the test load. Any power supply that strays outside these bounds automatically fails as excessive voltage fluctuations could lead to your computer becoming unstable.
The really complicated bit comes from the fact not all power supplies are built equally, or rated to the same maximum power, so testing and properly loading each one requires a bit of forethought. Nominally, we try and load each power supply to 50 per cent, 70 per cent, and 100 per cent of its rated maximum. So, for a 1000W power supply we test at a total of 500W, 750W and 1000W. However, as mentioned, each section may have its own limit so, for instance, all the 12V rails combined my only be able to draw 750W, the 3.3V + 5V rails only 200W total, and the -12V and 5Vsb rails only 50W. The trick is finding a combination that adds up to the total power draw you're currently looking for.
Once the numbers are decided, though, you just set the load levels on the test equipment, stand back, and flip the switch. If there's not a loud bang, it's a good start, and you can get on with reading those voltage levels and checking temperatures and noise levels.
One caveat that must be noted is that the test equipment, by its very nature, is quite noisy so our test room was by no means silent. However, we feel noise levels were still well within acceptable limits to make a judgement call on the relative noise levels of the power supplies themselves.
Our final assessment then takes into account the ergonomics of the power supply including the design of the modular cable system (if the supply has one), the layout and number of cables, and any other miscellaneous observations. Oh, and of course, price does come into it as well.