Skip to main content

Picking The Right Power Supply: What You Should Know

Efficiency, Efficiency, Efficiency!

“How Much Do I Get Out When I Put This Much In?”

While this is a valid question, we should probably rephrase it a little. Usually, you call the ratio between the amount of power drawn (from the wall wart) and the amount of power that is put out (to the computer) efficiency. The lower the amount of power a PSU has to draw in order to output a specific target power, the higher its efficiency.

While we’re at it, we’d like to clear up a very common misunderstanding regarding efficiency. If you have a 500W power supply with an efficiency of 75 percent, that doesn’t mean it can only output 375W to the PC. Instead, it has to draw 666W from the wall in order to provide 500W to the computer. So, the correct version of our question is, “How much power does my PC draw from the wall when it requires a certain amount of power?”


Let’s assume we’re really pushing our PC and it needs 600W. Our PSU is rated at 80% efficiency. Here’s what it’s really drawing from the grid:

600W / 0.80 = 750W

Ideally, our PC will draw about 750W from the wall under load. The remaining 150W are, quite simply, wasted and usually dissipated by the PSU as heat.

Nothing Is Constant, Not Even Loss

Our example above only holds true in an ideal world though, and since we don’t have super-efficient Star Trek technology, things usually don’t end up being that straightforward. A computer is used in various states, ranging from idle to full throttle, if you will, with every shade in between. Obviously, it will use the least power idling on the desktop, more in casual use and most under full load (3D graphics or intense calculations). Thus, we can’t expect to see constant power usage. Instead, we have to assume at least two states, namely idle and load. Now, let’s take a look at the efficiency of our hypothetical 600W power supply under various loads.

Uh-oh; what’s this? Our nice, straightforward explanation seems to get bent out of shape in that graph. Looking at the curve, we can see that the PSU reaches its peak efficiency at about 50% of its nominal capacity.

Now, a clever observer would suggest that simply making the PSU twice as powerful should solve the problem. While this is correct in principle, our helpful friend would be forgetting something: the idle state. And this is where modern switching power supplies run into trouble. If their load drops to below 10%, efficiency plummets to 50 or 60%, possibly even less. Ironically, this situation is only exacerbated by the power-saving mechanisms implemented in modern PC components. For example, a powerful system with a good graphics card can get by with as little as 65W when idling, but draw a good 500W under load. Thus, you have to ensure that the PSU is neither overtaxed nor under-challenged.


This time, let’s say our 600W PSU is supplying 65W to the system. What load does that correspond to?

(100% / 600W) * 65W = 10.83%

Now, take a look at our chart, and you’ll see things aren’t looking too good. Let’s repeat our calculation, this time assuming a 68% efficiency.

65W / 0.68 = 95.6W

Despite the fact that the system really only requires 65W, the PSU is drawing almost 100W from the wall and turning the remaining 30W into heat. And these are the numbers for the more efficient of two hypothetical power supplies, too! Not to get ahead of ourselves, but there were a pair of efficiency curves in that diagram, one for a cheap PSU and another for a more expensive one. And wouldn’t you know it, the supposedly cheap (and fictitious) DragonMegaHyperCombatUltra PSU for 30 bucks turns out to be a real power hog when the system is idling, driving up your power bill in the long run.

Again, this is only a hypothetical example. For our next trick, we’d like to show you what actually happens. As it turns out, we can easily allow for the impact of efficiency in our calculations. Oh, and it’s just as easy to prove that cheap PSUs will often turn out to be a lot more expensive than you might think in the long run.