TechPowerUp PSU Review Methodology
Author: crmaris Date: 2011-04-15 17:45:53 Conclusion »

Measurements

If you spend some time reading the ATX specification then you will easily find out what are the most important tests that a PSU reviewer should conduct. However in order to save you from the trouble we will make a short reference to the most significant of them.

Voltage Regulation

A PSU should be able to keep all of its rails within some predefined voltage ranges at all cases/loads. In the table below you will find these ranges (ATX 2.31 specification).

Voltage Name	Range	Minimum	Nominal	Maximum
+12V1	±5%	11.40 V	12.00 V	12.60 V
+12V2	±5%	11.40 V	12.00 V	12.60 V
5V	±5%	4.75 V	5.00 V	5.25 V
3.3V	±5%	3.14 V	3.30 V	3.47 V
-12V	±10%	-10.80 V	-12.00 V	-13.20 V
+5VSB	±5%	4.75 V	5.00 V	5.25 V

Efficiency: Normal & Light Loads, Standby Mode

The 80 PLUS certification measures efficiency at 20, 50, and 100% load of the PSU's maximum rated capacity.
If a PSU has an 80 PLUS certification (plain 80 PLUS, Bronze, Silver, Gold or the new Platinum) then it must have the equivalent efficiency. However, as we already stated, 80 PLUS measures at merely 23°C ambient while most PSU reviewers measure efficiency at way higher ambient, so it happens sometimes that a PSU that is certified to a certain efficiency category, fails to deliver the same efficiency at higher temperature. Also many PSUs are tuned to deliver high efficiency above 20% load (the minimum that 80 PLUS measures) but at loads lower than 20% their efficiency is pretty low. Since you will probably run your system for large periods in low energy consumption modes (Internet surfing, Office productivity etc.), efficiency at light loads is of high importance to you.
In my reviews I measure efficiency at three different light loads (40, 65 and 90W).

The ATX specification also states that efficiency of 5VSB should be measured, too. In the table below you will find the minimum efficiency levels that should be attained by 5VSB.

Load	Efficiency
100 mA	>= 50%
500 mA	>= 60%
>= 1 A	>= 70%

Finally, in 2010 the European Union released a guideline on Energy Related Products, ErP Lot 6, which states that every electronic device should have below 1 W power consumption in standby mode. In 2013 this limit will be further reduced down to 0.5 W.

Ripple Voltage

Ripple are the AC fluctuations (periodic) and noise (random) found in the DC rails of a PSU. Ripple significantly decreases the lifespan of capacitors since it increases their temperature (10°C more decreases the lifespan of a capacitor by 50%). Also ripple plays an important role in overall system stability especially when the latter is overclocked.

The ripple limits, according to ATX specification are 120 mV for the +12V and -12V rails and 50 mV for the rest (5V, 3.3V and 5VSB). However, in modern PSUs we expect to find way lower ripple and just a small fraction in some high-end ones (e.g. Seasonic, Delta made).
Below you will find a schematic which analyzes a ripple waveform.

In the above schematic four AC components can be identified:

Low frequency ripple associated with AC mains frequency
High frequency ripple due to PWM of the main switches
Switching noise that has the same frequency with switching PWM
Non-periodic random noise that is not related to any of the above

Transient Tests

In real life a PSU is always working with loads that change all the time, depending on whether the CPU or graphics cards is busy. So a decent PSU review should contain some tests with dynamic or transient loads. At TPU we run two different transient tests.

While the PSU is working at a 20% load state a transient load is applied to the PSU for 50 ms (10A at +12V, 5A at 5V and 6A at 3.3V).
While working at 50% load the PSU is hit by the same transient load with the exception now that load at 3.3V is increased by only 4A.

In both tests, we measure the voltage drops that the sudden load change causes. In any case voltages should remain within the regulation limits specified by the ATX specification (see above).

We also conduct three turn on transient tests, where we measure the response of the PSU in simpler scenarios of transient loads, during the turn on phase of the PSU. In the first test we turn off the PSU, dial 2A load at 5VSB and then switch on the PSU. In the second test, while the PSU is in standby, we dial the maximum load that +12V can handle and we start the PSU. In the last test, while the PSU is completely switched off (we cut off power or switch off the PSU's ON/OFF switch), we dial the maximum load that +12V can handle, then we switch on the PSU from the loader and we restore power. The ATX specification states that recorded spikes on all rails should not exceed 10% of their nominal values (e.g. +10% for 12V is 13.2V and for 5V is 5.5V).

Acoustics Tests

The ATX specification provides some information about PSU acoustics tests. In my opinion it's very hard to conduct proper acoustics measurements to a PSU since you must totally isolate it from the loaders, because the latter are way too noisy and will interfere with the noise measurements. However if you move the loaders far away from the test PSU then you will need long connection cables, which by their turn will have increased resistance so power losses on them will be significant. Finally a room with low external noise levels is needed. I hope that some day I will be able to construct a suitable environment for acoustics tests. I invested on a CEM-8852 Sound DB Meter, which can be connected via USB to a PC (for real-time measurements) and also has the ability to log up to 32700 readings.