Choosing a power supply for a specific computer configuration is a kind of an eternal problem, especially when the configuration is too advanced for the typical 300-400W PSU bundled with the system case to cope with. You can take it easy and buy a PSU with a wattage rating of some 1000W, but this is likely to be a waste of money. Unfortunately, there are often no comprehensive data on the power requirements of specific components. Graphics card and CPU makers want to be on the safe side and declare overstated specifications. Various power consumption calculators use obscure numbers. And the measurement of real-life consumption, even though mastered by most computer-related media, is often deficient.
When you open the Power Consumption section in a hardware review, you will usually find power draw data as measured from the wall outlet. That is, how much power is consumed from the 220V (or 110V if outside Europe) mains by the power supply which is loaded by the tested computer. It is easy to perform this kind of measurement: a consumer wattmeter is a small device with a single connector selling for less than $50, which is very cheap in comparison with serious measurement tools.
Such a wattmeter usually has a rather high accuracy, especially at loads of hundreds of watts, and copes with nonlinear loads (any computer PSU, particularly without active PFC, is an example of such load): there is a special microcontroller inside it for integrating the current and voltage over time, allowing to calculate the amount of active power consumed in the load.
Nearly every computer-related test lab has such a device.
We’ve got one, too. However, we use it only when we need to quickly estimate the power consumption of a computer or some other device (a consumer wattmeter is most handy then because it needs no kind of preparation) but not for serious tests.
The fact is, although this measurement method is simple, it provides an impractical result:
- The PSU’s efficiency is not taken into account. For example, at a load of 500W an 80% efficient PSU will consume 500/0.8=625W from the mains. So, if the “wall outlet” measurement yields 625W, you don’t have to run for a 650W PSU because a 550W model will suffice, too. Of course, the efficiency can be accounted for or you can even recalculate the result by having first tested the PSU and measured its efficiency under different loads, but this is inconvenient and affects the resulting measurement accuracy.
- This measurement gives you an average rather than maximum value. Modern CPUs and graphics cards can change their power consumption very quickly but short spikes are smoothed out by the PSU’s capacitors and you cannot see those spikes while measuring the consumption between the PSU and the wall outlet.
- If we measure the PSU’s consumption from the wall socket, we get no clue as to how the load is distributed, i.e. how heavy it is on the 5V, 12V and 3.3V rails. This information is both interesting and important.
- Finally and most importantly, the “wall outlet” measurement won’t show how much power is consumed by the graphics card and how much by the CPU. You can only see the overall consumption of the computer. But you want to know how much juice the specific CPU or graphics card need if you are reading a review about it.
The obvious, although technically more difficult, alternative is to measure the current consumed by the load proper from the PSU. It is possible, though, and I have even tested the Gigabyte Odin GT power supply with a built-in wattmeter of that kind.
In fact, the Odin GT can make a complete power-measuring testbed in its own right (and I don’t understand why other media don’t use this PSU for such measurements and why Gigabyte does not use this as a promotional opportunity) but I want to build a testbed that would be more universal and flexible in terms of how the load is connected to it.