This was originally posted by me over at Anandtech almost a year ago and was a sticky for about a month. Here it is in it's near entirety (I removed the part about calculating how much wattage you need until I get a chance to sit down and refine it for a modern PC)....
 
The power supply is the single most overlooked component in a computer system.
 
 
Common sense is going to tell you that the best way to determine the quality of a power supply is to stick with a brand that%u2019s known for quality products. Also, COMPETENT reviews help as well.
 
One indicator I%u2019ve used is the UL number of a power supply. UL actually certifies that a product is "safe" to use within a determined operating range and environment. Although I don't believe they actually load test every power supply, cross referencing a UL number can be handy for many reasons.
 
You can use a UL number to find out who actually makes a particular power supply. Also, if the UL logo is on the same label as the specs, you can be pretty sure that the UL listing pertains to the specs on that very label. Can you believe that there are a few companies that claim UL listing, but don%u2019t put the logo on their label? That%u2019s because the specs on that label are not the specs given to UL. And it%u2019s a regular Easter egg hunt trying to figure out what the actual specs are using the UL number. So why bother? If the company can%u2019t put a UL logo on the same label as the specs, they have something to hide and not worth doing business with.
 
Another way to classify the quality of a power supply is just from its weight. Simply put, a good power supply has good components in it. A better power supply has larger caps, transformers, heatsinks and overall more components than a cheaper unit. All of this adds up to more weight.
 
Also keep in mind the %u201Cget what you pay for%u201D adage. It doesn%u2019t always apply from brand to brand, especially if people take into consideration USEFUL improvements like modular cables, active PFC, etc. all add to the cost of a power supply. But when you see a power supply with a bunch of lights and other pretty things, you have to take into consideration that this added bling isn%u2019t free.
 
Picking the right power supply
 
Watts don%u2019t mean squat!! Know how to read the label!
 
I hope by now that you all know that Amps multiplied by Voltage equals Watts. And since a power supply for a computer puts out multiple voltages, the statement of how many total watts this unit puts out is really subjective.
 
The wattage rating is on the box of a power supply (like "Billtronic's 500W Mega-kleen-power" ) is the total capability of ALL of a power supply's rails COMBINED. The 5V, 12V, 3.3V, -12V, -5V and 5VSB capability are all added up to calculate a power supply%u2019s %u201Ctotal wattage rating.%u201D That total number really tells you nothing about the power supply's actual capability as it pertains to your particular PC.
 
First you have to ask, "is that wattage continuous power or maximum peak power?" Some power supplies will give you output ratings based on what the power supply can continuously output, while others give you peak power. For you audiophiles, this is similar to the difference between RMS and Peak. Some companies will actually rate a power supply at what it can continuously put out, but with a tolerance of 10% +/- from actual spec (12V, 5V, etc.)! Intel%u2019s ATX specification for power supplies only actually allows for a 5% tolerance!
 
There are also variables that come into play like %u201Cwhat was the temperature at which the testing was performed?%u201D %u201CFor what period of time was the testing performed at the specified wattage?%u201D Basically, you should look at the amperage each rail is capable of and then just consider that the power supply's BEST CASE SCENARIO capability.
 
The first thing you can do is to try to figure out your computer's WORST CASE SCENARIO load. There are several calculators on-line that allow you to "add up" your computer's power. Unfortunately, the bulk of these give you a final calculation in wattage. But if you can figure out what your +12V load needs to be, which is going to be the bulk of the wattage of a modern PC, by adding up your drive motors, fan motors, lights, pumps, video cards and CPU's, and add them all up, you'll be pretty close to figuring out what your PC needs.
 
You%u2019ll soon enough figure out how important the way the manufacturer distributes power across the rails really is. If you have a 500W power supply with 40A available on the 5V line and you're using a Prescott with SLI video cards, you might be in trouble because the 5V line alone is using up 200W of that power supply's total power not leaving much else for other rails! Given that most power supplies give you 20 to 30A on the 3.3V (which is way high by today's standards, but even 30A on the 3.3V is only 100W) and split up about 20W for negative voltage and stand by, you're only left with 180W for the 12V rail. That's only 15A! Mind you, we're talking maximum combined peak power, but better safe than sorry, right?
 
If you don't have the time or resources to do this, then just do this instead: Try to figure out if your PC is going to be 5V heavy or 12V heavy, and then buy the biggest, best quality power supply you can afford with the load balanced most appropriately for your PC. For example: If you have a Pentium III or an Athlon XP board without an ATX12V connector (like Biostar Socket A motherboards never have the 2x2 connector) then something with a relatively high 5V is most suitable for you. If you have a Prescott or an AMD64, consider something with a high 12V rail or rails (combined wattage) like a Silverstone ST56ZF or an OCZ 520ADJSLI. If you have PCI Express video card or cards, consider something with a really, really high 12V rail or rails (combined wattage,) like one of the SLi approved power supplies on nVidia's website.
 
So now back to helping you guys and gals read a PSU label. Use this as a reference:
 
 
 
This one is very simple. This power supply gives us 30A on the +3.3V rail and 30A on the +5V rail. Underneath these two, you%u2019ll see where the maximum combined capability of these two rails is 150W. That means, you can load up the +3.3V to 30A by itself, and you can load the +5V rail up to 30A by itself, but you can%u2019t load them both up to their maximum simultaneously. The rails are not additive. You%u2019ll also see 0.5A on the %u201312V and 2A on the +5VSB. I%u2019ll get back to the 12V rails in a minute.
 
Note that the total power of this particular power supply is 460W.
 
Ok%u2026 Now look at the 12V rails. There%u2019s two 12V rails (I%u2019ll explain why later) and they are rated at 18A each. The maximum combined wattage of those two rails is 32A.
 
What?!? But 18A plus 18A is 36A? Like the +3.3V and +5V rails I just mentioned, +12V rails are not additive. You can load each one up to 18A, but you can%u2019t load them both up to 18A.
 
Now let%u2019s go back to the %u201Cwatts don%u2019t mean squat%u201D phrase. I don%u2019t want to slam any brands, but take a look around at some 500W and 600W units. You%u2019ll actually find that even though they may have more %u201Ctotal wattage%u201D than this particular unit, this unit actually has more USABLE power on the 12V rails. Pretty interesting, right?
 
Picking the right power supply
 
So why do they split up 12V rails?
 
With the demand on +12V becoming greater and greater, Intel decided it would be "safer" to split the duty of supplying +12V across two rails. It's "safer" because less amperage can get to the end of a connector, therefore being less of a burn or a shock hazard. Furthermore, with the isolation of rails, shorts on one rail can be prevented from damaging components on another rail.
 
To split the duty up between two (or more) +12V rails, one can use cooler running, cheaper transistors and transformers to supply the power to multiple +12V rails. Though typically, a rail is split by simply taking a single +12V source and breaking up the capability with a series of over current protection; essentially a "logic" that prevents a certain amount of current from going to a wire, or group of wires.
 
Some people have questioned the principle of multiple 12V rails. And for good reason! But I don%u2019t think multiple 12V rails in general should be shunned. But it%u2019s best to know what rails go where when considering using a multi-12V rail power supply with a high end system.
 
ATX specifications only say that the CPU (the 2x2 4-pin connector) is put on a separate rail from the ATX connector (the 20 or 24-pin) and the drive (also used for fans, lights, etc.) power connectors. They also specify that no one rail should have more than 20A available on it (that%u2019s their %u201Csafe%u201D limit, so to speak.)
 
So if you breeze through reading that, you would say %u201COk. The CPU gets it%u2019s power from the 12V2 and everything else gets it%u2019s power from the 12V1.%u201D But then you realize there%u2019s a problem with that. 20A for just a CPU, even a dual core or even a dual CPU, is overkill. And 20A may be enough for some drives, lights, fans, etc. But what about PCI express video cards that regulate their voltage from the 12V rail via an auxiliary 6-pin connector? High-end video cards can easily tax 7A or more EACH off of the 12V rail. 20A leaves zero overhead.
 
Unfortunately, some power supplies adhere to the %u201Cquick read%u201D version of the ATX standard and put everything but the CPU on one rail. This is where everyone seems to be running into problems. Fortunately, some other power supply companies have gotten creative with rail distribution. I%u2019ve seen power supplies with the PCI express connectors on 12V2 and even some with one PCI-e connector on each of the two 12V rails. THESE are the kind of dual rail power supplies you need to look for.
 
Some power supplies have more than two rails. The Antec NeoHE, for example, has three. Two modular connectors are labeled for 12V3 use. These are the two ports one should plug their PCI-e connectors into. Other power supplies have four 12V rails. These typically adhere to a standard other than ATX called %u201CSSI%u201D but PCI-e is taken into consideration by keeping the PCI-e off of the same rail as all of the drives. Even if a PCI-e is plugged in using a typical drive Molex, that rail is still separate from the ATX connector, and the 2x2 4-pin connector.
 
Picking the right power supply
 
Efficiency:
 
The calculation for efficiency is DC Output divided by AC Input.
 
When a power supply is more efficient, it will use less power from the wall than one that is less efficient even if it produces the same amount of DC power.
 
The obvious upshot of this is a lower power bill. But also, the difference in wattage is dissipated in heat. So a more efficient power supply runs cooler. This can be used to an advantage one of two ways. One: A power supply will last longer if it%u2019s not exposed to prolonged temperatures. Two: A quiet fan can be installed because not as much air flow is required to cool a more efficient power supply.
 
Efficiency ratings are very subjective, though. First off, no power supply is going to have the same percentage of efficieny all across the board. One may be 75% efficient at 200W, but drop down to 70% under a 300W load. Some power supply companies only tell you what the best efficiency is, but not at what load that power supply obtains that efficiency. Other power supply companies only tell you worst-case scenario, like "70% nominal at full load." Some power supply companies may be rating their efficiency at 230V instead of 115V. A PSU that runs at 230V is often more efficient because higher voltage means lower amperage and lower amperage means less resistance and less resistance means less heat. Q.E.D. Furthermore, a power supply may be more efficient at a lower ambient operating temperature than another. If a power supply is 80% efficient at 20C, that doesn't mean it's 80% efficient at 50C. This will lead me to explaining "de-rating curves" in my next post.
 
Picking the right power supply
 
De-rating Curves:
 
A de-rating curve is something every power supply is subject to. As a power supply gets hotter, it%u2019s ability to output power is reduced. This relationship is called a de-rating curve.
 
For example; most decent power supplies are rated at 20C and have a de-rating curve of 2W per degree C. That means for every degree over 20C, your maximum sustained output is reduced by 2W. So in a more typical ambient temperature of 50C, a 500W power supply may only be able to output 440W. It%u2019s not a substantial loss of power, but not all power supplies have this de-rating curve and not all environments are at or below 50C and typically even a power supply with a good de-rating curve can drop exponentially when temperatures exceed 50C.
 
So how can you %u201Ctame the curve?%u201D There%u2019s lots of ways. One is to simply buy a lot bigger power supply then you actually need. That%u2019s a no-brainer. The other thing you can do is make sure that your power supply is not solely responsible for evacuating heat out of your chassis. Make sure you have a fan in the back below the power supply to exhaust heat, but also make sure you have an intake fan for positive pressure, because it the pressure inside the chassis is less than the pressure inside the power supply, you can actually DEFEAT the airflow of the power supply! I%u2019ve seen a few instances where a user had a 120MM in the back and no intake. The PSU was trying to also suck air into it%u2019s housing, but the vacuum caused by the rear exhaust fan actually reversed the airflow going through the power supply! Very not good.
 
UPDATE: In PC Power and Cooling's "Power Supply Myths exposed" they show a "500W" with a de-rating curve of 4W/1C. This isn't completely fictitious, but I don't think any of you guys are using the kinds of power supplies that have a 4W/1C de-rating curve.
 
Picking the right power supply
 
Resistance: Modular connectors, adapters and splitters.
 
Years ago, there was this cat named Ohm and he explained to us that resistance sucks.
 
Ohm%u2019s law as it pertains to resistance in electrical current is R (resistance) X I (current) = V (voltage.) So you can see, the greater the resistance, caused by either length of wire, gage of wire or having to go through connectors and/or the greater the current, the less voltage you get.
 
In simple terms, having a modular power supply may drop your voltage a little because of the resistance between the modular interface and the cable. And using a 20-to-24 pin adapter or any kind of splitter can cause a slight drop in voltage because of the resistance caused by any imperfect contact between the pins of such an adapter or splitter. But on that same note, every single connection you make (PSU to drive, or motherboard, or video card) is another connector that is going to create a little more resistance.
 
There%u2019s been a lot of scare tactics used to convince people to not go with a modular power supply. But the reality is, even at high loads the resistance is quite minimal if the correct measures are taken. For example: A PCI-e cable is going to have less resistance if there%u2019s 3 12V leads on each side of the cable and 3 grounds on each side of the cable. Unfortunately, some modular power supplies may only have one or two wires split into three for each row for a PCI-e connector. Some homework needs to be done on how the cables are constructed when considering a modular power supply.
 
And when using a modular power supply, adapters or splitter, make very certain that the connection between both interfaces is secure, firm and flush. Make sure all of your connectors are fully seated. This goes for standard power supplies and the connections you make to the motherboard, your drives, etc. as well. Because if you have a connector that is not fully seated, you create resistance. That resistance not only can cause a drop in voltage at the end of that particular wire, but also create heat. I%u2019ve actually seen BURNT connectors from cables not being plugged all of the way into their sockets.
 
One last thing; Gold plated contacts. They don't do any good unless they're interfaced with gold plated connectors. In fact, the mating of dissimilar metals is actually more prone to corrosion than if both connectors were tin. So if you get a modular power supply with gold connectors, keep in mind that it may be better to have gold only on the power supply side where the modular interfaces are also gold plated, but not on the component side. I haven't seen hard drives and motherboards with gold plated power connectors.
 
UPDATE: In PC Power and Cooling's "Power Supply Myths exposed" they state that "the voltage drop can be as much as would occur in 2 feet of standard wire." Actually, two feet of wire don't present much resistance. But they do make the point that they may "can easily loosen, corrode, and burn." That should read, "corrode or loosen and burn." Fears of corrosion are rather unrealistic. A power supply connector has as much chance of corrosion as any other contact point in your PC. Your video card? Your RAM? Even the connectors to your drives, motherboard, etc. Obviously, when you double the number of connectors you double the chance of corrosion, but unless you live on a House board, corrosion is rare. The loosen and burn I explain. Solution: There's no reason to keep unplugging and re-plugging your power connectors. Make sure they're in tight and leave 'em alone.
Picking the right power supply
 
Power Factor Correction:
 
The Power Factor of an AC electric power system is the ratio of the %u201Creal power%u201D to the "apparent power."
 
(Paragraph from Dan's Data) Power factor correction (PFC) is, essentially, what you do to complex AC loads (such as PC switchmode power supplies) to make them act more like simple loads (such as toasters).
 
There are two types of PFC, Active PFC and Passive PFC. This PSU has active PFC. Active PFC uses a circuit to correct power factor, Active PFC is able to generate a theoretical power factor of over 95%. Active Power Factor Correction also markedly diminishes total harmonics, automatically corrects for AC input voltage, and is capable of a full range of input voltage. Since Active PFC is the more complex method of Power Factor Correction, it is definitely more expensive to produce an active PFC power supply.
 
Passive PFC uses a capacitive filter at the AC input to correct poor power factor. Passive PFC may be affected when environmental vibration occurs. Passive PFC requires that the AC input voltage be set manually. Passive PFC also does not use the full energy potential of the AC line.
 
In some parts of the world, customers of the utility companies are actually charged more for poor power factor. In the EU, you are simply not allowed to use an electronic device with a complex AC load without any kind of correction! So certain inexpensive power supplies are simply not available over in Europe.
 
Despite being more efficient for your electric company, power factor may be less efficient to your power supply! The components used to correct power factor generate heat. Naturally, this heat didn%u2019t come from nowhere. It%u2019s using, and wasting, electricity. Furthermore, the heat being introduced to the other components of the power supply causing them to run hotter and therefore less efficient.
 
Bummer.
 
This is why you%u2019ll sometimes see certain models of power supplies available in the US with no PFC and available in the EU with PFC, but only capable of accepting a 230V input. Remember what I said about power supplies running more efficiently at 230V than they do at 115V? Same rule still applies here. The power factor correction circuitry isn%u2019t going to get as how with 230V coursing through it as it would with 115V because the amperage is going to be lower.
__________________
Forum rules: http://www.jonnyguru.com/forums/showthread.php?t=359