On PSU Efficiency

[Phaedrus2129]

"Make sure your power supply is 80+ certified!"

It's become almost a mantra of PSU recommendations in the past few years. Make sure your power supply is 80+, 80+ Silver, 80+ Gold. But how many know what it really means, and how important it is? If you think you know, odds are you're wrong!




Efficiency is a measure of how much of the power a PSU draws from the wall actually reaches the computer. Thus if a PSU is 80% efficient, and is drawing 375W from the wall, only 300W will reach the computer, and the rest will be given off as heat.

One common misconception, however, is that this means that the computer is getting less power than it needs/wants(?). But this is false. The wording is tricky. The amount of the power the computer needs is the independent variable; that changes on its own and is not affected by the PSU at all. The power drawn from the wall is the dependent variable, and changes in response to the independent, the power the computer is drawing.

The amount of power one draws from the wall is given thusly:

P(wall) = P(comp) / efficiency

So if your computer is demanding 300W from the power supply, which is, say, 70% efficient:

P(wall) = 300W / 0.70 = 430W
(actually it's 428.something, but I'm rounding)

Having a more efficient power supply means that less power is wasted as heat, thus you draw less power from your wall socket. Let's say your PSU is 80% efficient:

P(wall) = 300W / 0.80 = 375W

Thus having a more efficient power supply means you draw less power from your wall socket, thus have slightly lower power bills.



But a power supply's efficiency isn't always the same. Efficiency changes with the load being put on the power supply. Efficiency is generally lowest when a PSU is delivering very little power, or when delivering power close to what it's rated for. It's highest in the middle, around 40-60%:

So you should get a PSU the size so that your peak draw is around 50% of the PSU's load, right? FALSE, at least in most cases. Think: 90% of the time your computer will be idling, usually drawing only 75W-200W, depending on the computer. So you want your computer to be very efficient in that power range. Generally you want your idle power draw to fall into ~30%-40% of the PSU's rated wattage, and your peak power draw (think LinX + Furmark + HDTune all at once) to fall into ~70%-80% of the PSU's rated wattage.

The exception is with computers that will be under heavy load all the time, such as folding rigs. These should have the extra headroom anyway; a power supply should be chosen so that the computer is drawing ~60-70% of the PSU's rated wattage under folding power consumption levels. But this does not apply to computers that will be idle most of the time, like 99% are.


So what's this 80+ stuff? 80+ is an organization that tests power supplies for efficiency and a handful of other things (such as power factor correction, which we can ignore for now). 80+ has five certification levels: 80+, 80+ Bronze, 80+ Silver, 80+ Gold, and 80+ Platinum (the latter not yet being attained in modern commercial power supplies). The higher the PSU's efficiency at certain load levels, the higher the certification level.

However, 80+ is flawed, for a couple of reasons. First is that it only tests at three load levels: 20%, 50%, and 100%. This leads to power supplies being optimized so that their efficiency is highest in that wattage range. Theoretically, one could make a power supply that was efficient only at those load levels:

Though this is unlikely and I doubt that there are any PSUs like that on the market.


The second, and more important, problem is that 80+ only tests power supplies at room temperature (23*C, though engineering room temp is 25*C, so why the difference I don't know). A power supply's efficiency goes down as heat goes up, so a power supply that might meet 80+ certification standards at their temperature testing levels, might not achieve those efficiency numbers in real world conditions of 30*C-40*C, or worst case condition levels of 50*C.

So 80+ doesn't really mean much, even when it comes to efficiency, since it only applies if your power supply is in an unrealistically cool environment.





80+ means even less when it comes to a power supply's overall quality. 80+ does not make a PSU good. Efficiency is just one aspect of PSU quality. While high-end power supplies tend to be fairly efficient, not all efficient power supplies are high-end. Many shoddy manufacturers will sacrifice important things such as ripple suppression, voltage regulation, or build quality in order to make a unit that meets 80+ or higher efficiency levels, so they can slap that big 80+ label on there, because it's good marketing. Efficiency is just icing on the cake; and it doesn't matter how much icing is on there if the cake is made of rusty nails and cyanide.


Take home message? Efficiency and 80+ certification is just one aspect of a power supply's quality, not the be-all, end-all. It's worth taking note of, but you should check reviews to make sure of DC output quality (ripple/noise suppression and voltage regulation) and build quality (capacitor choice and soldering quality), as well as protections sets and aesthetics/functionality.



Thanks for reading.

 

[plugwash]

First is that it only tests at three load levels: 20%, 50%, and 100%. This leads to power supplies being optimized so that their efficiency is highest in that wattage range. Theoretically, one could make a power supply that was efficient only at those load levels:

As someone who knows about electronics it seems a PSU vendor would REALLY have to screw something up in a very weird way to have it efficiant at 25% 50% and 100% but crap between those figures.

The real problem is that their tests don't go low enough. Many systems with modern components are going to have a high ratio of idle power to peak power. PSUs need to be specced to handle peak power plus a safety margin plus room for expansion and this (unfortunately) means that most of the time they are running well below the peak in the power curve.

IMO 80 plus has achived it's goal of driving up the overall efficiancy of PC power supplies by attatching an obvious brand to efficiancy and setting standards across the industry. While 80 plus is no guarantee of a good PSU all the good PC PSU vendors seem to be using it. So a PC PSU (specialist PSUs excepted) without it most likely means a very old design that is inefficiant and likely crap in other ways.

 

[Phaedrus2129]

Yeah, I wrote that bit a while ago and haven't updated it to make the point about efficiency at 10% load and standby. Making graphs is frustrating.

 

[limitedaccess]

How does PSU aging affect efficiency graphs? For instance if your PSU peak power goes down over time, does the entire efficiency graph shift correspondingly so that if you were at 80% efficiency brand new you are still at 80% 2 years later? Does it not change, so the load % in comparison to your new lower peak is used to calculate efficiency? Or does efficiency itself also affected by aging? Maybe a combination of these?

 

[Phaedrus2129]

I don't know if I understand your question.

A power supply's efficiency will degrade over time to an extent, as traces and leads corrode, dust builds up, capacitor capacitance decreases and ESR increases, and various other processes affect the various components.

The efficiency curve will generally keep the same shape however, just shift a percent or two lower. At most you might see the high-load part of the curve dip lower due to component derating.


A power supply is inefficient at low load because the circuitry requires a minimum amount of charge running through it to be active; for instance a PSU might need about 20W to keep the circuitry energized, even if you're only pulling 50W from it.

Conversely a power supply is inefficient at high loads because higher current means more heat, means higher resistance, means more power wasted as heat, thus lower efficiency.

Between 40% and 60% load these two factors reach their intersection point, and you get the unit's highest efficiency.


So while the overall efficiency of the PSU will drop with time, most of this is due to static increases in resistance. The energy needed to charge the main circuits will not have greatly changed, so the low-load part of the efficiency curve will look the same. Corrosion and overall derating will have more affect at high load so that part of the curve will gradually become steeper, but not hugely so until the PSU is very old. For the most part the curve should look almost the same as the PSU's efficiency degrades, just lower.


And efficiency degradation takes a long time; a PSU's average efficiency may drop only one percentage point in the first three years, less if it's a top-of-the-line unit.

 

[Zero82z]

Sigh.

 

[Tudz]

The 80 plus certification ratings seem as good as anything you're going to find as far as efficiency standards go. Most efficiency standards are BS, at least 80 plus is useful.

Yeah, having below 20% would be nice, though I dont think we're at the point where it matters all that much. If you're using an 900W PSU, 20% is 180W. That's similar to the idle consumption of a PC running a GTX 460 (source1 source2). If you're using an 900W PSU to drive a single GTX460, you're doing something wrong. If you're running a HTPC which might idle around 100W or lower, you're probably using a PSU that's around 400W or less anyway. So really, the 80 plus standard matches the range in which you'd expect to use a PSU unless you bought something grossly overpowered for what you need. I find it unlikely that PSUs would ever be tuned to operate at 20, 50 and 100% load with downward spikes inbetween.

Testing at a standard temperature (and pressure and density) seems practical to me. It'd be nice to see some charts of efficiency vs temperature for different PSUs, but I dont imagine one PSU could be tuned significantly better than another for a standard room temperature but perform worse at an elevated temperature, thus 80 plus certification is fine for a comparitive tool. Also 23 degrees is a pretty standard room temperature. Through winter my home usually sits below 19C and in summer I try to keep it below 30C and I live in Australia. If people have aircon I doubt most they'd let their houses get significantly above 25-30C in the summer.

So yeah, I dont see a problem with 80 plus certification. Better standards are always good and if regulatary bodies wanted to improve on 80 plus I'd be all for it, but 80 plus is pretty good as far as efficiency standards go. Sure, it'd be nice if manufacturers were required to post graphs of efficiency vs load vs temperature on their boxes, but I doubt most people know what their computer idles at anyway so standards like 80 plus are good for those who just dont give a shit or aren't electrical engineers. And manufacturers would probably find another way to bullshit the data anyway.

P.S. I like your MS Paint graphs

 

[Bad ConNecTioN]

When you consider how much a driving factor 80+ has been in pushing the efficiency envelope, its a pretty solid standard. And as the previous poster alluded to, the 20/50/100 ratings are usually right on IF you buy a psu in your PC category.
The problem is, people still have a habit of buying grossly overpowered PSU's compared to their rigs. Lots of psychology at work. I think that's the next issue that should be tackled through PSA style marketing, by either a testing firm like the people behind 80+, or the various OEMs. That said, I'm not sure they want people buying smaller units...I think their profit margins are higher on the more powerful PSUs.

I'd like to see something like the new calorie decals you find on various candies, but for power consumption. This way, a customer can see the various consumption ratings on their parts and make an appropriate PSU choice. Not fool proof, but a place to start I think. Give the ratings a blue, yellow, or red color to distinguish their power use within their class (tier might be a better word). How do you define a class? No clue, but there needs to be a way to highlight how one card might draw twice as much power, even though it performs the same.

Original picture not mine...

 

[Tudz]

I agree Bad Connection, power ratings on actual components would be a big and useful step forward. It is dependant on things like firmware and driver versions, but it'd still be nice to see as a standard for labelling how much each component in your system is adding to the total power consumption.

For me, just an idle, load and average power consumption would be fine, though I'm able to figure things out for myself... the great unwashed masses may prefer some sort of "rating" system

 

[limitedaccess]

I don't know if I understand your question.

Sorry Just to clarify. Mainly trying to learn something new here.

Take your 4th image (2nd graph), using the 80+ gold as an example. In this case with a 80+ gold 500W PSU brand new.

So at now new -

If 500W of load is placed on the PSU, we are at 87% efficiency, at 450W we are at 90%. And we will say our hypothetical system does require 450W exactly at full load.

Now say the PSU has aged and now can only provide 450W max at 100% load. Is that 450W now 87% or 90%? Basically is the curve % based on what the PSU could do new or what it actually does currently?

Or does the entire curve degrade somewhat? So it is neither 87% or 90%.

 

[Phaedrus2129]

It's based on what the PSU could do when new, or in other words the entire curve drops.