How Power Supplies Work - Turbo Nerd Edition

[CPotter]

 

[Rex Hite]

That's a nicely done video!

[WhitetailAni]

That poor 990FX board...

[WolframaticAlpha]

You should've put the burning motherboard of the boring power supply with linus. It would've drawn more views.

[Nystemy]

As an electrical engineer myself, videos like these sometimes makes me cringe. Or most times makes me cringe to be fair...

There is the small niggles like "Switch-ing mode power supply" being technically incorrect, since the technical term is just "Switch mode power supply" or SMPS for short, and Varistor aren't called "VerY-resistors". And the primary side mosfets don't actually make a square wave, it is more a saw tooth to be fair. _{(Not that the mosfets even generates it, it is more a side effect of flyback converters.)}

 

But I will be honest and say that this video is honestly well made.

And isn't full of actually incorrect and misleading information. A rare feat outside of electrical engineering channels. _{(considering how even such channels at times gets important details wrong as well.)}

 

Though, could have explained flyback converters to be fair, they are actually rather trivial in practice.

 

Since all they do is let current flow through the primary side, this induces magnetic buildup in the core, that induces a reverse voltage on the primary side, trying to prevent an increase in current flow. On the secondary side it also induces a reverse voltage that is blocked by our rectifier.

 

When our mains side transistor turns off. Current suddenly stops flowing on the primary side, the magnetic field in the core collapses, inducing a forward voltage in both the primary and secondary side. Our primary side won't create any noticeable current since out main switching transistor is off and won't let it pass. Our secondary side however happily lets current flow through the rectifier in the correct direction.

 

Then we just regulate how much power we send over by how long we keep our main side transistor on for each cycle. We can also reduce the frequency, but that limits our ability to recapture the energy in the transistor gate itself, ie our switching losses gets higher.

 

And a fun side effect of flyback conversion is its ability to generate any arbitrary output voltage regardless of input voltage. The transformer can on paper have a 1:1 turn ratio and still do 400 V to 12 V conversion just fine. But in practice, this would require us to have a rectifier able to handle a 412 V reverse bias at the minimum, not accounting for switching noise. _{(something that is really common in these supplies.)} 412 volts isn't much, there is 1000 V rated diodes on the market, but typically the secondary side will use synchronous rectification using FETs, since this results in less power loss, and high current transistors typically don't also do high voltage, unless it is physically huge and typically water cooled.

 

Also a quick note on capacitors.

It is actually very hard to just make a list of capacitors and say, "These brands are good." Because each brand have different types of capacitors with different ratings, aiming at different applications. Where the ratings are more than just capacitance, voltage, ESR, MTBF @ temp. There is additional specifications that are harder to see at the surface. Like recommended peak and ripple currents at a given temp, and the ESR over frequency since it varies, not to mention the obvious package/formfactor/footprint it is built to. Then there is safety features like how it vents, if it has internal inrush limiting, among other details.

Just looking at the brand is a bit short sighted to be fair. But a Nippon Chemicon is generally of better specs than a Rubycon, but they are both good capacitors. The important part is that they actually live up to their specs, and some of the really "scummy" "brands" will actually rip off name brands and not deliver even close to what is stated on the tin.... And this is the kind of crap one should avoid.
 

Though, I don't know how many times I have talked to people/customers and heard the statement, "why is there so many different but similar parts, can't the industry just make 1 component of the type to fit all applications?!", and the answer is that it isn't that simple. It might though seem silly how there is about a 1000 different 1µF electrolytic capacitors on the market, but they have different voltage ratings, ESR, packaging, and MTBF values, and construction techniques, so they aren't even close to interchangeable. And that is just 1µF electrolytic caps, then there is different values and dielectrics, and then all the other component categories like resistors, inductors, transistors, diodes, LEDs, IC, terminals/contacts, buttons, etc...

[akio123008]

I like that video. You guys took a topic that's easy to mess up and.. didn't mess it up!

 

2 hours ago, Nystemy said:

the primary side mosfets don't actually make a square wave, it is more a saw tooth to be fair.

You're probably talking about the current through the primary (on a flyback) which is indeed sort of a sawtooth, while the video is probably talking about the voltage, which is pretty much square.

 

 

2 hours ago, Nystemy said:

could have explained flyback converters to be fair, they are actually rather trivial in practice.

Haha well perhaps to you or other engineers that is the case, but I suspect it's less trivial to the average LTT viewer. (who is younger than you might expect) I think the amount of detail they included was just right.

 

Computer power supplies don't use flyback converters anyway, they usually use some form of (resonant) forward converter. (the resonant variant is mentioned in the video).

[Elisis]

1 hour ago, akio123008 said:

they usually use some form of (resonant) forward converter.

 Either resonant LLC, two switch forward, or active clamp, in fact. Though the latter two are generally only present in cheaper models, of course. 

[Tech User]

 

2:08 The magnetic field from the first coil does not move the electron in the second coil. It, in fact, induce the voltage first (V=-dB/dt), then the load circuit determines if electrons flow based on ohms law. (Even if there is voltage potential, electrons won't flow in an open circuit)

 

3:33 Without a linear regulator that shaves off the ripples, a linear power supply will have that annoying 50/60 Hz hum. It is also common for audio devices to use a switching regulator -> filtering -> linear regulator system to deliver clean yet efficient power.

 

3:54 The transformer/ full-bridge rectifier system is not really sensitive to the wall power frequencies. The voltage part is correct though, that's why in old electronics there is often a 120/230 volts switch. 

 

11:50 Ohmic heating is not really the concern here. Although it is correct, copper traces on PCB have very low resistance (<<1ohm, can be even much less if power plane is used) so it doesn't make that big of a difference. As the article shown in the video, the primary benefit of a 48v power supply is the potential (pun intended) to make VRMs more efficient. On a side note, Linus sometimes talks about 12vo like its some sort of magic that switching regulators on the motherboard can just somehow be more efficient than the current power supply scheme. It is in fact that switching regulators are most efficient at around the mid-range of their spec, so having multiple switching regulators that can be turned off at idle would improve the efficiency at idle since the ones active are still running at their mid-range.  

[Tech User]

4 hours ago, Nystemy said:

 

"Switch-ing mode power supply" being technically incorrect...

I mean wiki uses "switched-mode power supply" while TI have used all of them: "Switch Mode Power Supplies", "switched-mode power supplies", and "switching power supply" in their publications so ¯\_(ツ)_/¯

[Elisis]

2 minutes ago, Tech User said:

I mean wiki uses "switched-mode power supply" while TI have used all of them: "Switch Mode Power Supplies", "switched-mode power supplies", and "switching power supply" in their publications so ¯\_(ツ)_/¯

The 'real term' entirely depends on which EE book you read first lmao

[wONKEyeYEs]

9 hours ago, CPotter said:

 

Judging from the forums, The most pertinent point would be "DON'T SWITCH YOUR CABLES!" @10:40.

[Orian Pax]

12 hours ago, CPotter said:

 

I needed this video!!!

Waiting for gamers nexus video on gigabyte PSU

[Nystemy]

16 hours ago, akio123008 said:

Computer power supplies don't use flyback converters anyway, they usually use some form of (resonant) forward converter. (the resonant variant is mentioned in the video).

It depends to be fair.

A flyback converter has major advantages in regards to voltage conversion. Ie, choosing an arbitrary input to output ratio is simple, since it only moves over clumps of energy.

A resonant mode supply on the other hand is little more than a high frequency inverter running a regular linear transformer, but at far higher frequencies and with materials with less losses, but there is methods to provide some output voltage regulation as well with some downsides in efficiency. The saving grace for resonant mode supplies is the PFC_{/(boost converter)} circuit allowing for a nice known input voltage to the converter almost regardless of mains voltage.

Resonance ones also needs to keep themselves in tune, and the synchronous rectification on the secondary side is where the real problem comes in for DC output voltages. _{(This is however not a problem for induction heaters, induction stoves, AC welders, etc...)} Since the secondary side transistors will typically have a larger gate charge associated with them, due to having relatively low ESR. And keeping the secondary side resonating along is going to require some tuning. _{(This isn't hard to do in practice, all one needs is a variable capacitance or inductance to trim the oscillator and keep it in phase with the main converter. A mostly single chip solution. (or one can use an analog mux, some passives and a micro. I haven't done this before....))}

 

The resonance type supplies are more efficient and less electrically noisy, but are more fiddly to build, and even winding the transformer itself needs to have some additional care. Since parasitic capacitance between the windings will skew the resonance point, sometimes by a lot. And when one builds a PSU to a budget, one don't want to hand trim each unit to make it work... So the transformers aren't going to be as cheap.

 

Flyback converters can reach similar efficiencies, typically a percentile or two behind _{(if both are well designed and built)}, but they are far simpler to just churn out by the millions. And the issues surrounding higher currents can typically be circumnavigated with a multi phase design, something I have only seen in some server supplies so far... _{(Flyback based PSUs that fits on the palm of one's hand and still deliver 2.5+ kW and is just casually air cooled makes one wonder how warm it runs...)} And recapturing the gate charge in the secondary side transistors is something that as well can be done, all though, self resonance does have advantages over non resonance based energy capture, especially when one wants to switch the transistors fast, an inherent disadvantage of flyback converters.

The biggest downside with resonance converters is when the ratio between the input and output voltage vary over a large span. Ie, 127-355 V _{(rectified mains)} in and "constant" 12 V out. Then a flyback one quickly takes the upper hand in efficiency. But add a high efficiency boost converter (typically in the form of PFC) on the mains side, and the resonance one will have move towards having an advantage, especially if the flyback converter also needs PFC as well. _{(Depends on the efficiency of the PFC circuit though, resonance + PCF might be overall less efficient than just flyback.  But there is other reasons why one has PFC after all. (There is reasons why some data centers use DC power, saves a lot of losses in the PFC circuits, but then one also has a very stable 300-400 V DC to work with, so resonance mode is still a nice PSU. And data centers partly use DC since they then don't need inverters for their battery backup, saves a good few % of peak power needed from the batteries, and also gives longer run times and there is less to break and maintain.))}

 

They both have pros and cons, and can both achieve 90+% efficiency across a fair portion of their operational range.

[akio123008]

4 minutes ago, Nystemy said:

They both have pros and cons, and can both achieve 90+% efficiency across a fair portion of their operational range.

Just to clarify, I wasn't saying I don't like flybacks. In fact, if I were to build a psu it would probably work that way because it's the easiest to build.

 

What I was pointing out though, is that I don't know of a single psu on the market that uses a flyback converter, so I'm not sure if explaining its workings in this video would make the most sense.

 

(while it does perhaps deserve its own video)

[Nystemy]

47 minutes ago, akio123008 said:

Just to clarify, I wasn't saying I don't like flybacks. In fact, if I were to build a psu it would probably work that way because it's the easiest to build.

 

What I was pointing out though, is that I don't know of a single psu on the market that uses a flyback converter, so I'm not sure if explaining its workings in this video would make the most sense.

 

(while it does perhaps deserve its own video)

I were mostly providing additional content on the topic.

Stating the pros and cons of both.

 

I have built both types myself.

Resonance is nice if one has a mostly fixed input to output ratio. Like a PFCed power supply supplying 12 volts.

And flyback is wonderful if one wants still really good efficiency, but with less hassle. Or where one has a varied input to output ratio. Like battery charging cars. _{Though, the industry is set on shoehorning in resonance mode supplies here regardless, preferably with one huge transformer and banks of water cooled IGBTs switching in unison, else they break and release the magic smoke.... The car industry seems inept at the advantages of a multi phase design.....}

 

Though, even the PC industry is inept at multi phase PSUs, VRMs exist up to and beyond 16 phases, but power supplies are stuck at 1 _{(unless one goes to a few specific server PSUs)}... Going to 2 would halve the peak currents, and quarter the peak power dissipation. For every part involved. (Though, just in case an electrician comes in, I do Not mean phases in regards to the mains wiring, but phases as in power stages working out of phase with each other.)

 

Though, here is a quad phase, air cooled, 12V 200A server PSU, my experience would indicate that it is a flyback one. Could be a full bridge one though..
But main take away is, multi phase designs aren't bad, this is actually tiny for the amount of power it delivers. Together with the caps, this PSU is still having a smaller volume than a standard ATX supply.

This is however "only" 92% efficient when close to its max load of 2.4 kW, but I think the main reason for this low efficiency is due to the fact it needs to supply 200 amps while still being the size of two decks of cards. I am surprised that it doesn't melt its buss bars to be fair... In short, I think most of the efficiency is gulped up by conductive losses. It also delivers an impressive 5V 1A standby.

[dogwitch]

54 minutes ago, Nystemy said:

I were mostly providing additional content on the topic.

Stating the pros and cons of both.

 

I have built both types myself.

Resonance is nice if one has a mostly fixed input to output ratio. Like a PFCed power supply supplying 12 volts.

And flyback is wonderful if one wants still really good efficiency, but with less hassle. Or where one has a varied input to output ratio. Like battery charging cars. _{Though, the industry is set on shoehorning in resonance mode supplies here regardless, preferably with one huge transformer and banks of water cooled IGBTs switching in unison, else they break and release the magic smoke.... The car industry seems inept at the advantages of a multi phase design.....}

 

Though, even the PC industry is inept at multi phase PSUs, VRMs exist up to and beyond 16 phases, but power supplies are stuck at 1 _{(unless one goes to a few specific server PSUs)}... Going to 2 would halve the peak currents, and quarter the peak power dissipation. For every part involved. (Though, just in case an electrician comes in, I do Not mean phases in regards to the mains wiring, but phases as in power stages working out of phase with each other.)

 

Though, here is a quad phase, air cooled, 12V 200A server PSU, my experience would indicate that it is a flyback one. Could be a full bridge one though..
But main take away is, multi phase designs aren't bad, this is actually tiny for the amount of power it delivers. Together with the caps, this PSU is still having a smaller volume than a standard ATX supply.

This is however "only" 92% efficient when close to its max load of 2.4 kW, but I think the main reason for this low efficiency is due to the fact it needs to supply 200 amps while still being the size of two decks of cards. I am surprised that it doesn't melt its buss bars to be fair... In short, I think most of the efficiency is gulped up by conductive losses. It also delivers an impressive 5V 1A standby.

crazy what some of the server psu ratings are both by out put and why is it not melting for being so small.....

[Nystemy]

8 minutes ago, dogwitch said:

crazy what some of the server psu ratings are both by out put and why is it not melting for being so small.....

Even among servers, supplies like this are rare to be honest.

Most power supply manufacturers seems happy with just overbuilding one big power stage instead of going with a handful of smaller ones.
Though, a lot of things in society follows that logic, even when some redundancy exists.

Might some day make an 12VO PSU just for fun, but build it with 4-8 power stages and see how overkill it can get. The advantage with more power stages in parallel is that it reduces current density, and also allows for lower resistance, and thereby greatly reducing power losses from both sides, all while improving transient response. Though, also increasing costs.

[Egg-Roll]

On 6/13/2021 at 3:19 PM, Nystemy said:

And isn't full of actually incorrect and misleading information. A rare feat for LTT.

Fixed it for you, all jokes aside the video was done by Alex so I would expect it to be accurate, and I doubt his ego is as bad as Linus's and would actually ask (or research into it) fellow co-workers if he didn't know about something...

Should have had 0100110001101001011011100111010101110011% less Linus and more Alex, guess Linus was bored that day. Either that or Alex threw him in as a sick gag, at which time I'm ok with then.

 

Overall an actually decent video for LTT.

[dogwitch]

22 hours ago, Nystemy said:

Even among servers, supplies like this are rare to be honest.

Most power supply manufacturers seems happy with just overbuilding one big power stage instead of going with a handful of smaller ones.
Though, a lot of things in society follows that logic, even when some redundancy exists.

Might some day make an 12VO PSU just for fun, but build it with 4-8 power stages and see how overkill it can get. The advantage with more power stages in parallel is that it reduces current density, and also allows for lower resistance, and thereby greatly reducing power losses from both sides, all while improving transient response. Though, also increasing costs.

oh yeah. but you got to admit its dam impressive. none the less where we come in 20 years of psu

[vsteel]

I think the greatest feat of this video is breaking it down for the general public.  While the electrical engineers can argue over the pros and cons of various components or conversion circuits, most people don't care.  The average person has only had their science class where they hook a light bulb to a battery and been told the difference between AC and DC circuits. 

 

We have to remember sometimes it is easier to get a concept across with a non perfect description than trying to get into the intricacies.  For the general public, once the phrase "well technically" is uttered, people tune out.   They got the general idea so they are moving on.

[derpkotacasper]

If no one else is gonna ask it, I'm going to: Are you guys beefing with Gamer's Nexus...? This sure feels like a direct response to Steve's current excitement over his power supply testing equipment and there were a few other moments this month it felt like the two channels were being either passively or directly aggressive towards one another. I'm about to make y'all touch noses until you learn how to be friends.

[Spotty]

14 minutes ago, derpkotacasper said:

If no one else is gonna ask it, I'm going to: Are you guys beefing with Gamer's Nexus...? This sure feels like a direct response to Steve's current excitement over his power supply testing equipment and there were a few other moments this month it felt like the two channels were being either passively or directly aggressive towards one another. I'm about to make y'all touch noses until you learn how to be friends.

No, of course not. This video was uploaded to Floatplane almost 2 weeks ago, a week before Gamers Nexus published their review of the Dell PSU. It's just a coincidence that both channels published videos to youtube around the same time that included explanations on how PSUs work.

LTT was sponsored by Seasonic to do a video on power supplies. The Turbo Nerd series is something they did before with motherboards.

[derpkotacasper]

3 minutes ago, Spotty said:

No, of course not. This video was uploaded to Floatplane almost 2 weeks ago, a week before Gamers Nexus published their review of the Dell PSU. It's just a coincidence that both channels published videos to youtube around the same time that included explanations on how PSUs work.

LTT was sponsored by Seasonic to do a video on power supplies. The Turbo Nerd series is something they did before with motherboards.

Ah, at least you acknowledged what raised my eyebrow. Very well then, I really didn't want for there to be beef between the two channels unless it would result in fun and friendly competition lol