PSU 101: PC Power Supply Basics

[Jon Gerow]

We hope this helps some of you out there. If anyone has any input, please feel free to comment. We are not necessarily experts, but simple hobbyist with a thirst for knowledge and a desire to teach others. -- jon


• PC Power Supply Basics

a. Overview:

Computer components typically run off of DC, or direct current. Not only this, but different components actually require different DC voltages. Our cities power grids provide AC, or alternating current. So, for a computer to receive the electricity it needs to function, we not only need to convert AC to DC power, but we also need to convert the AC into several different DC voltages.

b. PSU:

PSU is an acronym for “Power Supply Unit” and is often used as a short form to describe a computer’s power supply.

c. SMPS:

SMPS is an acronym for “Switch Mode Power Supply.” A computer power supply is of the SMPS type, as opposed to a linear power supply (wall-wart type power supplies are typically linear.) SMPS power supplies are generally more complex than linear power supplies and can generate high-frequency electrical noise; however, they are smaller, more efficient and generate less heat than any linear design of equivalent output.

Switch Mode Power Supplies get their name from the MOSFET (metal oxide semi-conductor field-effect transistors) that switch between full saturation and full cut off at a high rate during the inverter stage of power conversion. Take notes, there will be a test.

d. Wattage Ratings on Power Supplies:

Simply knowing the wattage of a power supply is only part of knowing a power supply’s capabilities and quality. When you see the wattage rating of a power supply, you’re seeing the total maximum output capability of that particular power supply, but a computer has multiple voltage needs, and newer computers require more of the power supply’s capability to be on the +12V DC output rail. CPU’s and GPU’s regulate their power off of the +12V DC rail. Also, all of the computer’s motors run off of +12V DC: hard drive and optical drive motors, fan motors, pumps for water-cooling, etc. Just because a power supply can produce up to 500W of power does not mean it can put out 90% of that wattage as +12V, or even 40%.

e. Maximum Wattage… A Relative Term:

The wattage number you see on a power supply’s box is referred to as the total maximum output wattage. But maximum is actually a relative term depending on the power supply unit. Is the number the maximum continuous output or peak? If it’s continuous, under what conditions is this capability? What input voltage? What operating temperature? If peak, for what duration of time?

f. Rating Temperature:

Another variable that comes into play when considering maximum wattage is the operating temperature at which this maximum wattage was determined. It’s not uncommon for a power supply to perform differently under different thermal conditions. Often, as a power supply’s operating temperature increases, it’s capability to put out power decreases. This is referred to as the de-rating curve. Some power supplies are rated at 40°C, which is ideal since this is typically the maximum ambient temperature for air entering a power supply installed within a PC, while others are rated at 25°C, or “room temperature.” A handful of units is actually rated at 50°C, which is actually the typically the highest operating temperature possible for a computer power supply.

One should not mistake a power supply’s maximum operating temperature (typically 50°C) for the temperature a power supply’s maximum output rating is determined. A power supply may be capable of operating at a very high temperature, like 50°C, but may not be able to do its advertised maximum output at this temperature.

g. De-Rating Curve:

A power supply has a maximum power output and a maximum operating temperature, but rarely can it do both at the same time. As a power supply’s operating temperature increases, it is not uncommon for it’s capability to put out continuous DC power to decrease. The ratio between the rise of temperature and decrease in maximum DC output capability is called a de-rating curve because, when graphed, a gradually decreasing curve is formed. For example: an 800W power supply, rated at 40°C, with a de-rating curve of -10W/+1°C is only capable of sustaining a 700W load at 50°C. An 800W power supply with a de-rating curve of -10%/+10°C is only capable of sustaining a 720W load at 50°C.

This is actually quite important information to know, especially when implementing a power supply because many power supplies are rated at 25°C, yet typical operating temperatures are in the neighborhood of 40°C. Therefore, it is not unusual to expect less than 75% maximum capability from a deployed unit.

h. DC Output “Rails”:

Computer power supplies put out multiple voltages. Each of these separate voltages is called a “rail.” A computer power supply typically has a +3.3V rail, a +5V rail, a +5VSB rail (SB stands for “Stand By” as it is live as long as the unit is receiving any AC current) a -5V rail, a -12V rail and one or more +12V rail. Typically, all leads of the same voltage draw power from the same rail, so it is erroneous to state that each set of wires coming from a power supply represent a rail. Even a power supply with multiple +12V rails tend to distribute the power of a particular +12V rail across multiple wires to multiple connectors. Furthermore, multiple +12V rails are often split off of the same +12V source.

i. Multiple +12V Rails:

Many power supplies on the market have multiple +12V rails. This may be accomplished by having more than one +12V transformer or, more typically, taking the typical +12V output of a power supply and splitting it up into multiple, what are often called "virtual" +12V rails. Note that even units with multiple transformers may have these two individual outputs split further with "virtual" rails, or in some cases these two outputs may be combined to create one larger +12V rail.

The division of the +12V rail is done because a 240VA output (equivalent to 240W DC) may be potentially harmful or fatal or may cause fires due to the overheating of leads burning wire insulation caused by too much current being delivered to a connector, typically due to a short (as per UL and EM 60950 safety requirements.) Because modern PC’s regulate CPU and GPU V-core from a +12V source, a PC can require much power from the +12V rail of a power supply than PC’s of the past that only tend to use the +12V rail for drive and fan motors. It is not unusual for the power demands of the +12V rail to exceed 240VA, thus the need for multiple +12V rails. The assumption that each individual connector with a +12V lead is on its own rail is not uncommon. But the fact of the matter is that there may be several connectors all using the same +12V rail.

j. Computer Power Supplies with Power Factor Correction:

Some computer power supplies have power factor correction or “PFC” for short. Power factor correction comes in two flavors: Passive and Active. Passive PFC can improve a computer’s power factor to .75. Active PFC can improve a computer’s power factor to .99.

Although power factor does not typically affect a power supply’s ability to convert as much AC power into DC as possible, it does put a strain on the mains, and some utility companies have decided to charge customers based on their VA usage, as opposed to their actual Wattage usage. In some parts of the world, the sale of any type of switch mode power supply other than one with power factor is against the law.

Power supplies with active power factor correction are often capable of world-wide usage because the power factor correction circuit is capable of accepting a very wide range of input voltages. For this reason, many power supplies only come with active PFC. This allows the manufacturer to sell the product anywhere around the world, requiring nothing more than a different power cord to accommodate different wall outlet sockets. Be aware that some power supplies may have active power factor correction, but may only accept higher voltage input (like 230V). This is because higher voltages are delivered at lower current, therefore the components in the power supply need only be capable of handling half the current of a power supply built to work on lower voltage mains (like 115V.) The use of these smaller scale components can significantly reduce the cost of a power supply.

k. Red 110/220 (or 115/230) Switch on the Back of the Power Supply:

Power supplies that do not have active power factor correction often will have a red switch on the back of the unit. This switch should be set to the setting closest to the voltage coming from the wall. When set to the lower voltage on the switch (the 110V or 115V depending on the unit) the power supply’s input rectifier stage acts as a voltage doubler. If the higher voltage is selected, the rectifier stage works as a regular AC to DC conversion circuit.

[cypherpunks]

The above is quite right, but perhaps a little convoluted.

Temperature derating A power supply has a maximum power output and a maximum operating temperature. But it can't do both at the same time! To achieve maximum power output, it requires good cooling, which means cool input air. If you want to operate it in warmer air, you have to run it at less than full power.

Cheesy power supplies trumpet their power rating with 25ºC cooling air. This is a common testing temperature in the electronic industry, but a typical computer is a lot warmer than this. Better manufacturers will quote the power rating with more realistic 40ºC or even 50ºC cooling air.

(Note: I see Corsair's HX series is specified at 50ºC. But I do not see a temperature spec anywhere in the VX or TX spec sheets or manuals.)

Multiple +12V rails Actually, the 240W limit isn't an electrocution limit but a can-cause-fires limit. You're not allowed to send more than 240W (i.e. 20A at 12V) over any one wire. (Coming close would be horribly inefficient, and the wire would get warm, but wouldn't catch fire.) But current limit circuits cost money, so generally manufacturers share them across multiple wires. Obviously if the total power through these 10 wires is less than 240W, then so is the individual power through any one wire.

But then they have to decide which wires to share, and sometimes people had legitimate reasons to place a total load of more than 240W on the wires the manufacturer picked.

One way to fix this is to add more current limit circuits. But a much cheaper way is to quietly ignore that part of the ATX spec. Some actually advertise a large single rail, while others just omit the current limit circuitry. A lot of people like this, and Intel politely avoids testing it as part of ATX certification.

But this does not mean that the electricity is in any way shape or form separately generated. The separate rails are just like separate circuits in your house power. All the power comes from the same place; they just have separate circuit breakers.

(There are a very few power supplies which truly have multiple independent
+12V outputs. Some people think this is a good thing, and I think they're idiots; they don't understand the advantages of polyphase switching power supplies. A two-phase converter running at 50% duty cycle (where a good PSU designer tries to operate anyway) has, to a first order, zero ripple. Even if the duty cycle isn't exact, the ripple is still less than half. But that's a rant for another day.)

[Jon Gerow]

The above is quite right, but perhaps a little convoluted.

Temperature derating A power supply has a maximum power output and a maximum operating temperature. But it can't do both at the same time! To achieve maximum power output, it requires good cooling, which means cool input air. If you want to operate it in warmer air, you have to run it at less than full power.

Cheesy power supplies trumpet their power rating with 25ºC cooling air. This is a common testing temperature in the electronic industry, but a typical computer is a lot warmer than this. Better manufacturers will quote the power rating with more realistic 40ºC or even 50ºC cooling air.

(Note: I see Corsair's HX series is specified at 50ºC. But I do not see a temperature spec anywhere in the VX or TX spec sheets or manuals.)

I think I'll rephrase it a little.

Multiple +12V rails Actually, the 240W limit isn't an electrocution limit but a can-cause-fires limit. You're not allowed to send more than 240W (i.e. 20A at 12V) over any one wire. (Coming close would be horribly inefficient, and the wire would get warm, but wouldn't catch fire.) But current limit circuits cost money, so generally manufacturers share them across multiple wires. Obviously if the total power through these 10 wires is less than 240W, then so is the individual power through any one wire.

But then they have to decide which wires to share, and sometimes people had legitimate reasons to place a total load of more than 240W on the wires the manufacturer picked.

One way to fix this is to add more current limit circuits. But a much cheaper way is to quietly ignore that part of the ATX spec. Some actually advertise a large single rail, while others just omit the current limit circuitry. A lot of people like this, and Intel politely avoids testing it as part of ATX certification.

But this does not mean that the electricity is in any way shape or form separately generated. The separate rails are just like separate circuits in your house power. All the power comes from the same place; they just have separate circuit breakers.

I'll rephrase to reflect. Thanks!

(There are a very few power supplies which truly have multiple independent
+12V outputs. Some people think this is a good thing, and I think they're idiots; they don't understand the advantages of polyphase switching power supplies. A two-phase converter running at 50% duty cycle (where a good PSU designer tries to operate anyway) has, to a first order, zero ripple. Even if the duty cycle isn't exact, the ripple is still less than half. But that's a rant for another day.)

Well.. we'll try not to go off into this tangent. There's a lot of opinions on this based on a costerformance stand point that makes a good argument.

What I will say is dumb is when you DO have a truly multiple +12V source and then they combine it before splitting it and this virtually doubles ripple and noise. But again... off topic.

[cypherpunks]

What I will say is dumb is when you DO have a truly multiple +12V source and then they combine it before splitting it and this virtually doubles ripple and noise. But again... off topic.

Sorry to prolong an off-topic discussion, but no. That's exactly what I was complaining about. Just combining the two doesn't increase voltage ripple at all (the ratio of ripple current to output capacitance doesn't change), but that only applies if they're in phase.

Anyone with the slightest shred of sense arranges for the different supplies to operate out of phase. This halves ripple, at the very least, and can in the best case reduce it to zero.

The only time when that makes any sense is when you have an extremely "dirty" load that feeds back so much noise to the power supply that it bothers another "quiet" load more than the extra PSU ripple would.

Voltage ripple equals current ripple divided by filter capacitance. Current ripple is proportional to current. None of this is related to operating at low voltage, but low voltage supplies (like CPU supplies) need less voltage ripple. Which is why three- and four-phase CPU power supplies are universal these days.

Polyphase power supplies rock. Only an idiot would give that up.

[Jon Gerow]

Well... what I am talking about are the units I have seen that use two seperate transformers and then merges them into one rail have pretty much ALWAYS had nearly double the ripple. I'm not going to name names, but we've seen the same platform used for both single +12V and split +12V and the split +12V always has considerably less ripple and noise than the "merged" single rail version.

I'm not familiar with nor have I see a unit that uses what you are talking about. Not in the personal computer arena, at least.

[cypherpunks]

the units I have seen that use two separate transformers and then merges them into one rail have pretty much ALWAYS had nearly double the ripple. I'm not going to name names, but we've seen the same platform used for both single +12V and split +12V and the split +12V always has considerably less ripple and noise than the "merged" single rail version.

I'm not familiar with nor have I see a unit that uses what you are talking about. Not in the personal computer arena, at least.

Bizarre. No insult intended, but if I measured that, I'd go back and measure again to be sure I hadn't made a mistake. Out-of-phase operation is so obviously superior that I don't know of a multiple-output SMPS controller chip that even allows in-phase operation. It reduces input ripple (making RF interference testing easier to pass) as well. If you get the chance with such a supply sometime, could you 'scope the ripple on the two outputs at the same time and see if they're in phase or not?

Obviously, a manufacturer might take some of the improvement in the form of smaller and cheaper ripple-reduction capacitors, but I'd be shocked if the ripple actually got worse. I can imagine ways in which it could be accomplished—like building a "ripple filter" with such high ESL that it actually rings at the higher effective switching frequency—but they involve some pretty grotesque incompetence.

[Jon Gerow]

Bizarre. No insult intended, but if I measured that, I'd go back and measure again to be sure I hadn't made a mistake.

No offense taken. And note I said "units" not "unit." I've had more than a couple chances to measure and re-measure these units with multiple +12V transformers. I wish they were using an out-of-phase topology, but they're merely using a multiple transformer design, essentially two PSU's in one housing, and bridging the outputs together. Nothing more.

[cypherpunks]

I wish they were using an out-of-phase topology, but they're merely using a multiple transformer design, essentially two PSU's in one housing, and bridging the outputs together. Nothing more.

Er, that's all you have to do to the outputs. There's nothing out-of-phase about the topology, it's just the timing of the primary-side switches. See, for example, the ISL6322 polyphase PWM controller. The data sheet shows a schematic on page 5. Indeed, the outputs are simply connected together.

The interesting part is the timing diagram, figure 1 on page 11. Notice how the ripple current is not only at three times the frequency that the ripple from any one switch (and thus will produce only 1/3 the ripple voltage given the same output capacitors), it's also actually smaller in amplitude, because the rapid ramp-up of one phase is partially canceled by the ramp-down in the other two.

In fact, if the duty cycle were to be exactly 1/3 on, the ripple currents would cancel exactly and you'd have zero ripple.

That's why I'm baffled. If you simply connect two identical in-phase outputs together, nothing should happen to the voltage waveform. After all, if they're actually identical, then no current will flow through the wire connecting them.

And if you connect out-of-phase outputs together, the ripple actually cancels. That's just Kirchoff's current law.

Now, since real power supplies involve feedback, you need to be careful or you'll screw things up. In particular, you need to ensure that they share the load evenly, rather than having the one with a fractionally higher output voltage setting do all the work. But those are all well-understood problems; it would seem to take real skill to screw it up.

[Oklahoma Wolf]

See, for example, the ISL6322 polyphase PWM controller. The data sheet shows a schematic on page 5.

That looks rather more complicated than the topology of any PSU I've seen. I would surmise that most of these units with more than one transformer are doing so not because they wish to employ polyphase operation, but because they can't figure out how to keep a single transformer small enough, or something of that ilk

This Topower build seems to be one such unit: http://www.hardwaresecrets.com/article/403/6

[cypherpunks]

That looks rather more complicated than the topology of any PSU I've seen.

It wasn't in a "PSU", but I'm sure you've seen it. That's a common motherboard processor power supply chip. When you need to supply a lot of current with very tight regulation in the face of wicked load steps, polyphase designs are mandatory.

I would surmise that most of these units with more than one transformer are doing so not because they wish to employ polyphase operation, but because they can't figure out how to keep a single transformer small enough, or something of that ilk

Oh, no doubt, but the question is, once you've decided you need multiple transformers, why not take advantage of them?

Now, the Tagan TG-1110 you mention is an example of just running the two transformers in parallel because they didn't want to redesign the control circuitry. As the review notes, the power components cost the same, but I assume there is some time-to-market advantage in a "brute force" design approach.