How much does DC to DC conversion increase efficiency?

[Pentium4]

I'm curious how much it can increase efficiency compared to linear regulation. Is it easier on the main transformer as well?

[tom66]

Re: How much does DC to DC conversion increase efficiency?

Compared to linear regulation? Very few computer power supplies use linear regulation, except maybe for the -12V regulator.

[Pentium4]

Re: How much does DC to DC conversion increase efficiency?

Okay I probably used the wrong terminology compared to non DC to DC conversion

[eccerr0r]

Re: How much does DC to DC conversion increase efficiency?

Still don't understand the question. It all depends on the conversion circuit too. Linear regulation is notoriously inefficient but it's very quiet in terms of power noise - best for audio. About all the noise you get is 50/60 cycle hum if not filtered properly. But it needs to dissipate the extra voltage as heat to get the clean power.

Now DC-DC conversion specifically means converting one DC voltage to another. The problem is that power comes in the house as AC. So really what can you compare to? That's why this question is not clear.

If you meant you want to convert 120VAC -> intermediate DC and then that DC -> your computer via DC-DC converter, this probably is an overall loss for individual users. You have to go through one more stage for no benefit.

For people with racks and racks of machines, this may be beneficial. The first stage from 120V to say 48VDC can be done efficiently in one phase and that 48VDC can be passed along to your machines via a large bus bar. The smaller differential and the pre-regulated DC makes the machines' DC-DC converter smaller and more efficient. However there is loss in that first converter from 120V to the intermediate, so the gain isn't as huge as it seems. It's a huge space savings however, and less power dissipated at each individual machine.

This is IMHO though...

[Wester547]

Re: How much does DC to DC conversion increase efficiency?

If you mean DC-DC conversion as in the way modern power supplies regulate the +12V rail down to lower voltages/outputs like +5V and +3.3V, then I would say more efficient. It helps with cross loading, especially with group regulated power supplies and, to put it simply, makes it 'easier' for the PSU to generate and sustain certain voltages, perhaps even with less noise and ripple (it may be easier for the transformer, though I think that deals with more with the primary side assuming that the +5V and +3.3V rails aren't literally connected to the same transformer as cheaper PSUs may have them share just that), especially when the power draw is high (assuming that, of course, the MOSFETs or schottky rectifiers that are being used in DC-DC conversion are overspec'd by a significant enough margin, and that the output filtering is good enough). This is going to sound hokey, but wouldn't your job be more efficient for you if you could keep all the necessary assets/your work in one location instead of three? Same thing for a power supply - it helps with latency. It may also improve upon the combined rating. Minimum load resistors, if I'm not mistaken, are also theoretically there to ease crossloading in group regulated PSUs, though it seems they are there for the wrong reasons in cheapo power supplies...

As far as linear regulators go I know +5VSB falls into that for a PSU (having its own linear regulator). But back to DC-DC conversion, what's also good about it is that it gives allowance for MOSFETs to be completely used in the secondary circuit of a PSU (I believe), and they are a large step up from schottky barrier rectifiers because MOSFETS, or essentially transistors, can switch much faster (especially if the RDS(on) specification is low enough) as they are capable of an 'on' and 'off' state rather than the one direction the current flows in with diode(s) (schottky barriers being packages that contain two diodes in parallel). I like how even many older Hipro PSUs always use DC-DC conversion to generate the +3.3V rail from the +5V rail because of this, being group regulated. When done right it can be a very effective method. Some consider it cheap, considering it was the old way, now and then, that very old PSUs would generate +3.3V, but so long as isn't just done to save cents...

This may go into arithmetic that's well above my head, so if I have any of this wrong in my attempted explanation feel free to point that out.

[Pentium4]

Re: How much does DC to DC conversion increase efficiency?

If you mean DC-DC conversion as in the way modern power supplies regulate the +12V rail down to lower voltages/outputs like +5V and +3.3V, then I would say more efficient. It helps with cross loading, especially with group regulated power supplies and, to put it simply, makes it 'easier' for the PSU to generate and sustain certain voltages, perhaps even with less noise and ripple (it may be easier for the transformer, though I think that deals with more with the primary side assuming that the +5V and +3.3V rails aren't literally connected to the same transformer as cheaper PSUs may have them share just that), especially when the power draw is high (assuming that, of course, the MOSFETs or schottky rectifiers that are being used in DC-DC conversion are overspec'd by a significant enough margin, and that the output filtering is good enough). This is going to sound hokey, but wouldn't your job be more efficient for you if you could keep all the necessary assets/your work in one location instead of three? Same thing for a power supply - it helps with latency. It may also improve upon the combined rating. Minimum load resistors, if I'm not mistaken, are also theoretically there to ease crossloading in group regulated PSUs, though it seems they are there for the wrong reasons in cheapo power supplies...

As far as linear regulators go I know +5VSB falls into that for a PSU (having its own linear regulator). But back to DC-DC conversion, what's also good about it is that it gives allowance for MOSFETs to be completely used in the secondary circuit of a PSU (I believe), and they are a large step up from schottky barrier rectifiers because MOSFETS, or essentially transistors, can switch much faster (especially if the RDS(on) specification is low enough) as they are capable of an 'on' and 'off' state rather than the one direction the current flows in with diode(s) (schottky barriers being packages that contain two diodes in parallel). I like how even many older Hipro PSUs always use DC-DC conversion to generate the +3.3V rail from the +5V rail because of this, being group regulated. When done right it can be a very effective method. Some consider it cheap, considering it was the old way, now and then, that very old PSUs would generate +3.3V, but so long as isn't just done to save cents...

This may go into arithmetic that's well above my head, so if I have any of this wrong in my attempted explanation feel free to point that out.

Sorry I didn't explain better but this is the answer I was looking for. Thank you I didn't know the MOSFET's could be used like that on the secondary side, that's interesting!

[Behemot]

Re: How much does DC to DC conversion increase efficiency?

All of the modern PSUs with DC-DC have actually very low noise levels, usually arround the level you think it's more like background noise, that's what I got from reviews I have seen so far

Also MOSFETs have way lower losses, when you compare 0,5+ V on shottky and few miliohms on FET, I think you can calculate it. That also means much higher switching currents in the same power envelopes so most modern FETs go from 30 to 90 A per package.

Also due to the low losses you can cool them very well, Seasonic first brought FETs being cooled with the very PSU casing back with the X line, now it's quite common. Than you can spare heatsinks, cool it better, reduce noise or increase overall power.Or any combination of above said

[eccerr0r]

Re: How much does DC to DC conversion increase efficiency?

You can use MOSFET synchronous rectification (i.e. "secondary" ) in traditional AC-DC converters, to increase the secondary rectification efficiency in 120VAC PSUs too. So there's no distinction here why a DC-DC converter is advantageous compared to 120VAC conversion.

[Behemot]

Re: How much does DC to DC conversion increase efficiency?

Obviously, it is. Modern transistors can bring as much as 95 % efficiency in DC-DC converters, show me where have you seen that achieved by synchronous rectifier?

[eccerr0r]

Re: How much does DC to DC conversion increase efficiency?

Please factor in the conversion efficiency getting to the input of the DC-DC converter. Remember you have to consider the whole system.

Let's say you have 95% efficiency from 120V/220V to the DC-DC converter. Then another 95% on the DC-DC converter. Your system is now only 90% efficient...

The whole point is if you can remove a converter, THEN you can realize some savings.

[Behemot]

Re: How much does DC to DC conversion increase efficiency?

Obviously having one strong rail AND couple DC-DC IS more efficient than any other way. Or have you seen some other topologies with 90 % overall efficiency on consumer market? I haven't.

[eccerr0r]

Re: How much does DC to DC conversion increase efficiency?

There are some 120V PSUs that reach 90% because they were demanded by power centers. Most of them are in servers, in which there are even 94% efficient converters from 120V to motherboard. Consumers are just getting the 80+ devices and they're getting better.

Really there's no way to realize the so-called advantage of DC-DC conversion on off the shelf parts - they all need to be redesigned. There are already DC-DC converters on motherboards - they all need to be ripped out and changed to the common voltage to truly take advantage of this. If people were planning on changing their home machine from traditional to a 120V - to - 48V, then 48V to 12V/5V/3.3V, this is no where efficient not to mention additional cost. People have been quoting the 48V to to 12V/5V/3.3V numbers as the high 95% efficiency but the 12V to 1.5V converters, though probably also 95% efficient, adds more to the mix...

And nobody talks about the 120VAC to 48VDC (or whatever intermediate voltage, may it be 12V) conversion. How efficient can this be? Probably no more efficent than you can buy PSUs off the shelf right now. Larger PSUs (hundreds of amps) can be more efficient just because they can share the control logic together - not something the home can take advantage of.

Not to say there are other advantages to unifying things.

[Behemot]

Re: How much does DC to DC conversion increase efficiency?

Special PSUs like in big network switchers or blade towers operate at voltages in 24-48 V range. But I have been deliberatelly talking about consumer market.

I know there are very high efficient PSU on server market. If you will search, I bet you will find out that we are now getting on market what used to be mainstream in server area some 5 years ago. Just slightly improved and fit into standard ATX PSU chasis.

Synchronnous switching IS preset in todays high-end PSUs for the only rail it has (12 V) and DC-DC is ADDED to that for the smaller voltages. It has so many advantages, as mentioned for crossloading and so. Just have a look how many parts you will find in most half-bridge PSUs. Tens of resistors and stuff, it all adds to looses.

Now you don't need this at all. You have one rail, one feedback. watching it. Than couple small PWMs for single- or double-phase DC-DC and some transistors. This can operate at up to 1 MHz thanks to modern transistors and polymer caps.

I am not saying you can't achieve high efficiency the old way. It's just you had to built thinks like tanks, using many components in parallel to improve looses. You could achieve 80+ quite easily. Bronze if you made things better. Silver if you made it top.

Now, you use single rail with DC-DC and got silver just like that. Add synchronnosu switching and higher frequencies and you got gold. Add some other fancy stuff and you have 80+ platinum. Tweak it further and you are looking towards 90+. Thats where we are right now.

Why now and not sooner? Well, Seasonic and Enermax who are pushing market forward are not really making server PSUs. They sometimes have to invent what big OEMs have for years now, but they have no reason to sell it when they can sell worse stuff for same money (thats this capitalistic system you see). So they do. Than OEMs have reason to improve their stuff too, finally, so they do. And the market has just made technological progress. That's how I see it anyway

[Behemot]

Re: How much does DC to DC conversion increase efficiency?

To not start a new thread, I'll post here.

I am lately thinking about next step in improving PSUs efficiency. I still cannot loose the impression we will see IGBT. We have long enough experience with that in high-end UPS market, where some fanufacturers claim near-100 % efficency in coversion to 12 V under soem circuimstances. And 12 V is exactly what we are using in todays PSU's with DC-DC converters.

Also I am personally thinking about spreading of higher-power rated IGBTs with integrated drivers, what once again has obvious advantages of having everything in one package (much shorter paths thus lower capacitances and latencies, resulting in faster switching times). That all goes well with increasing switching frequencies up to 2 MHz, which is limit of todays aluminium polymers. Than maybe tantalum polymers will take their place? Whaddaya think?

[tom66]

Re: How much does DC to DC conversion increase efficiency?

IGBT is more lossy at low power than MOSFETs.

Also, IGBTs are only good up to around 30kHz, so transformers will be larger.

You do see them in high end server PSUs where they are more efficient at higher power.

In general you're unlikely to see them in a computer PSU <1000W.