Fanless PSU build-up, first phase

[gmat]

Originally posted by Axle

But these MOSFETS, they are LIVE, am I correct? I mean why else would most sinks be live then?

It's only a side effect.
There can be 2 reasons:
1 - the backplate of MOSFETs is also the middle pin. Depending on where you take your reference you may end up with mains going through.
2 - EMI. Those components emit strong EM pulses and heatsinks are like big antennas. Basically what you get is a voltage which is the picture of EMI inside your PSU.
It has no real use, but if you plan to remove the original HS you may shunt the holes in which they where connected, to 'emulate' a connection through the HS (just in case they were really used as a big 'wire').
Now make sure you use mica shims behind those MOSFETs to prevent any leakage current through your WB (and watercooling system...).

side note: on thoe 'real' fanless PSUs, the backplate is really at 0V since it's in contact with the PSU frame...

I basically want to "pull a bladerunner", with some minor differences. Even if anything else is getting hot, I'm going to cut a hole in the PSU box, and then the top of my case, seeing as heat rises. But I realize this may pose some electromegnetic issues, I'm just not sure what.

No real problem apart from those:
- your neighbours on the next floor will hate you if they're soccer (or generally TV) fans. EMI can be strong and scramble their TV reception
- make sure nothing falls in there.. No liquid no solid nothing. You know what means 'high voltage'...
- if you're near a military base they'll knock at your door with angry looking faces.

Oh (feel free to laugh-I'm really not an EE), what about replacing the MOSFETS with Active PFCs? Probably around the supidest thing you've ever heard, but I had to ask.

Thankyou sir!

No problem
Active PFC is actually extra circuitry that controls the switching components, to maximize their efficiency and 'maximize' PF. PFC aka Power Factor Correction modify the global impedance of switching circuitry, to bring the PF as near as 1 as possible. This has for result to 'smooth' current pikes drawn on the mains, and reduce your electrical bill. 'Active' PFC adds active components to control the switch so that it occurs at 0V difference, and other neat features, to eliminate harmonic noise and current spikes (and EMI as a side effect).
In short with active PFC a PSU will handle more smoothly current pikes and quick load variations (which happens a lot in our modern PCs), and produce a cleaner output under load, and be cleaner in your electrical system.
The downside is, it needs more componentry, and is way more expensive...
Mhhh i read my post and i'm not sure i'm clear. Some links:
http://www.national.com/appinfo/power/ -> good basic info
http://www.traceengineering.com/tech...otes/tn11.html -> what's PF
http://www.epanorama.net/links/psu.html -> great links on PSU infos

[MadDogMe]

MOSFETs have a 'mica shim' under them when the HS is live, I can't understand why they make the HS live either, maybe to act as a trace?, but BladeRunners PSU still worked after he got rid of the 'live' HS...

Cheers gmat, I was'nt to sure about the 5vmobo mod, it seemed alot was'nt explained properly. a new PSU is in order I think (Antec 550 T/P@£99). how come the 5vSBV 3.5, 1.5AGP and 2.55DDR V's are all fine, it's only the +5&12 that are diminishing?...
I'll use multicore mains wire for the ATX mod (I'll do it anyway even with new PSU cause ATX's are loose~crap!), the block connector is mains rated for 3AMPs as well, I'm pretty good with a soldering iron even if I do say so myself . flux is the key I think, and tinning the wire properly...
********************************************
O/T, did we ever pick your brains about a way to make a Eheim1250 start in the same direction every time?, the sine wave of the AC means it'll start either clockwise or anti~clockwise. if it could be controlled it'd be easy to mod the pump for more flow+head ...

[gmat]

Originally posted by bigben2k
What's a PFC?

I'm looking into building a PSU array, and I have to somehow fit a chip in the PSU that adjusts the output (it also has an output for the array controller). I think it works with the regulator, but I'm not sure...

Yes a regulator is what you want. Make sure you protect it with decoupling capacitors... small ones will do as you're not inflicting it a huge difference.

[gmat]

Cheers gmat, I was'nt to sure about the 5vmobo mod, it seemed alot was'nt explained properly. a new PSU is in order I think (Antec 550 T/P@?99). how come the 5vSBV 3.5, 1.5AGP and 2.55DDR V's are all fine, it's only the +5&12 that are diminishing?...

Because those are the lines where current is drawn. 5V powers electronics components, and 12V powers motors and high-current-draw components through mobo regulators. You can easily get a 10A draw on the 12V and a 5A draw on the 5V, or more. The -12,-5 and others are there only for reference points, a few Op Amps, and basic mobo functions (BIOS with CMOS save, power save, etc) so they get almost no current.

flux is the key I think, and tinning the wire properly...

You know what you're talking about. Good luck then

********************************************
O/T, did we ever pick your brains about a way to make a Eheim1250 start in the same direction every time?, the sine wave of the AC means it'll start either clockwise or anti~clockwise. if it could be controlled it'd be easy to mod the pump for more flow+head ...

If we could get to the motor coils, yes. A variable impedance controlled by a JFET at startup would 'dephase' slightly the sine going to one coil, making sure the first sequence followed by the rotor is 1->2 and not 2->1. But with no internal access to the pump electrical guts i dont see any easy solution...

[MadDogMe]

i dont see any easy solution

Nothing that could be mounted as a seperate curcuitboard?, no way of implementing it into the wires?, it HAS to be on the coils?. I wish I understood the wave form and frequency more, as it is I have no idea how it all works...

Don't know if it would help but, it does'nt matter which way it turns(clock+/counter), as long as it turns that way only...

This would be an 'Uber Mod'! if we could suss out the motor, without it we are stuck with valves which are inherantly restrictive ...

[Since87]

Originally posted by gmat
Whoops. That's a common misconception, BB2K.
Heat is produced by inefficiencies, aka losses internal to the PSU. Usual PC switching PSUs have about 70% efficiency. The 30% left are mostly heat.
Higher rater PSUs have simply fatter transistor and lines to sustain heavier current. But as power increases, losses do as well.
So in brief:
- fatter PSU + same setup = same heat
- fatter PSU + heavier setup = more heat
Also, higher wattage PSUs have often lower efficiency (to maintain decent costs)... = even more heat !

The efficiency of a powersupply is not fixed for all operating conditions. The manufacturer's efficiency specification will most likely be measured in a "maximally loaded" situation. (How much of the load is on +5V and how much is on +12V is unknown unless you can get the manufacturer to cough that data up.)

A major portion of the losses is going to be "I^2*R" losses due to currents flowing through resistances. Assume a single 12V output supply. If you gradually increase the current draw from zero to maximum:

The output power will go up directly proportionally to I. (P = 12V * I)

The resistive power supply losses will go up proportional to I^2. (P= I^2 * R)

Therefore the supply will most likely be more efficient at output powers below its maximum ratings. A "fatter" supply with an equal efficiency rating (i.e. One with lower resistance in the current paths.) should be beneficial. (There are other losses in addition to the resistive losses and the maximum efficiency will peak somewhere between zero output and max output.)

Whether this hypothetical equally efficient fatter supply can be found I don't know. I haven't done much looking, but most of the supplies seem to be rated around 70% efficient.

[gmat]

Originally posted by Since87
(...)
Whether this hypothetical equally efficient fatter supply can be found I don't know. I haven't done much looking, but most of the supplies seem to be rated around 70% efficient.

Valid point, though i've seen that higher wattage PSUs tend to have lower efficiency. That's more like 80% for 250W or lower PSUs, 70% for 400W and 65% for 550W. Higher efficiency ratings exist of course, but they come at a price...
Active PFC helps getting the efficiency up, again rising the price.

You say "major losses comes from RI^2" but that's not exact in a switched PSU. Huge losses come from the regulation. Mostly resistive components (resistors obviously, coils) don't produce as much heat as MOSFETs themselves...

All comes down to -> want a cooler PSU, buy a more expensive one (beware of overrated products though..)

[Since87]

Originally posted by gmat
Well i'm an EE so i think i understand them
Mosfets rated for a higher current simply have a 'broader path'. If you swap a mosfet with a higher wattage one - and drawing the same current of course -, you'll get *exactly* the same heat produced.

It's not quite that simple. One of the most important characteristics of power MOSFET's is Rds-on. This is the resistance of the current conducting channel when the MOSFET is turned on.

You can get MOSFET's with very low Rds-on, but they are expensive. PC PSU's have a very low profit margin, so the manufacturers aren't going to be putting expensive state of the art parts in them if they can avoid it. I'd almost guarantee that I could find a MOSFET that would consume less power than the ones used in a given PC PSU.

It's not an easy task though. There are a lot of characteristics to be taken into account. (Max drain to source voltage, Peak current rating, Continuous current rating. Gate capacitances, Gate threshold voltages, Miller plateau voltage, Temperature deratings, to name a few.)

Any decent switching powersupply is going to be somewhat tuned to operate with the power MOSFET's designed in. Choosing a replacement would require a lot of knowledge of the switching control circuit to avoid an unstable system. That means you'd have to do a fairly thorough job reverse engineering the power supply, before you could even start. Much too big a PITA for me to be interested in doing it.

Not meaning to pick on you GMAT. It's just a matter of where one's experience lies. Most of the EE's I went to school with are full time programmers these days. I'd be surprised if many of them knew this stuff as well as you.

[gmat]

Hehe. No problem This is very interesting.

Actually with some friends (and unlimited access to Alcatel's HW labs - oscilloscopes, probes, HW experts and so on) we took some time to dismantle and reverse engineer a PC PSU.
This was frightening.
Imagine that, maybe 95% of the price is made of.. MOSFETs. They build the rest of the PSU around them. The switching circuitry feeds them an incredibly dirty, spiky and noisy signal, not to mention it's way out of level. Of the remaining 5%, maybe 4% is made of heatsinks and fan. The rest is the crappiest possible condensers, diodes, resistors and coils they can find. No wonder those regulators get damn hot.
So yes you're totally right when you say one cannot change PSU MOSFETs easily...

Comparing with an indutrial-rated PSU (such as those from RSG, or others) brings no similarity.. Not even close...

[Since87]

Yeah, it'd be pretty easy to design a much better power supply than the best PC PSU sold. I don't think you could make any money though, even specifically targeting the enthusiast market.

Most enthusiasts are extremely cheap when it comes to PSU's.

[MadDogMe]

How difficult would it be to make your own 500~600watt PSU?, I'd imagine the auto shutoff, load monitoring gizmos would be the hardest?. ]JR[ over at CoolHardWare.co.uk made a basic one for a TEC. that was 12(or24)Volt only, how much harder to make one with all the bells and whistles?. could you make one with a dedicated ampage to each V~line(non switching?) for around the same cost as buying a good switching PSU (550w @ £100?), or am I daydreaming/wishful thinking? ...

You could use an old PSU case and maybe the PCB?, and reverse engineer it from there with GOOD componants?(non~non~switching of course! ) ...

PS, you could make the 12v separate no?, cause it does'nt have to be as 'clean' as the +5v, +3.5v and 1.5v does it?...

[gmat]

Originally posted by MadDogMe
How difficult would it be to make your own 500~600watt PSU?, I'd imagine the auto shutoff, load monitoring gizmos would be the hardest?.

Since you've already got the soldering skills... not so hard, provided you have the appropriate testing/tuning hardware (oscilloscope, good multimeter, probes etc)
*Building* it (and getting components) is easy enough, the PCB can be kept simple and components are big.
*Tuning* it so it actually works with a good efficiency is another story...
There are EE that are *specialized* in power supplies, they're even called PSE. I know a bit because i've already built a few ones, but i'm nowhere near the skill level necessary to build a "bells & whistles" 550W ATX PSU. Even a full-time EE (which i'm not.. i'm a programmer now) wouldnt. This is really a specialty. I used to know one PS Engineer who i worked with. His work consisted mainly in designing power supplies of our products. He was also the PCB designer and EMI specialist (one excellent dude !). Knowing him, he would say "buy it, don't even bother trying building one."

]JR[ over at CoolHardWare.co.uk made a basic one for a TEC. that was 12(or24)Volt only, how much harder to make one with all the bells and whistles?. could you make one with a dedicated ampage to each V~line(non switching?) for around the same cost as buying a good switching PSU (550w @ ?100?), or am I daydreaming/wishful thinking? ...

Yes. For $150 you won't get a multi-stage, non-switched, 550W PSU. Even buying spare components will land you in the $400 zone, or more. If you want dedicated lines etc, you'll have quite a bulky unit - think of a 2U 19" rack just for your PSU.

You could use an old PSU case and maybe the PCB?, and reverse engineer it from there with GOOD componants?(non~non~switching of course! ) ...

Not easy, see my first remark. Building it is not the problem. Tuning the switcher, the active PFC and other timing sensitive / impedance sensitive components requires time and skill. (and money, or easy access to a HW lab..) And yes you'll HAVE to re-tune it after swapping components.

PS, you could make the 12v separate no?, cause it does'nt have to be as 'clean' as the +5v, +3.5v and 1.5v does it?...

From what i've deduced, 12V doesnt have to be as clean as 5V. But given the heavy load it usually takes, it'll require at least a PFC stage.

I don't want to discourage you. Building a power supply is a very good experience indeed. I've learned a lot when i did it. But try to build a dedicated 1-line PS first, and see how it turns out. A 12V (or 24V) @ 150W sounds good. Give it overtemp and overcurrent protection. Then and only then move to a multi-line w/ active PFC and overtemp / overcurrent protection...
Good schematics are available just anywhere on the Net. But that's a lot of work...

If you don't have the time (or hardware) to build one yourself... check the catalogs of good PSU mfgers. I'm getting RSG because they're in Europe, but i'm sure you can find some good ones in the USA.
Just remember, you usually get what you pay for...

[bigben2k]

I concur.

To build your own PSU, you will need an oscilloscope, to see where the line noise comes from, and how to remove it. You also have to have a good power resistor, for some power switch tests.

I've always wondered if the line noise comes from the AC supply, or if it's more from the switching of higher/lower power draw on the supply lines.

[Since87]

Originally posted by MadDogMe
How difficult would it be to make your own 500~600watt PSU

I did say it would be "pretty easy" didn't I?

Let me explain what "pretty easy" means to me.

Working full time (about 50 hours a week) it would take me about six months to design and produce a 500 Watt PC PSU with substantially better regulation and efficiency than the "high end" stuff sold these days, provided:

1. I had access to all the test and production equipment I have at work.

2. I get a substantial amount of support (Maybe a month each) from a PCB layout engineer and a mechanical engineer.

3. I get about three months of technician time for prototyping and testing.

If I could sell 1,000 of them, the price might be as low as $300 each, unless I actually wanted to make a profit.

Throw in the fact, that doing this involves working with high voltages, and quite likely, a lot of smoked components along the way and you've got fairly good idea of what I meant by "pretty easy". It should be "pretty easy" for the PSU manufacturer's to do this. For even an extremely talented enthusiast working in his garage, this would be a monumental task.

There's no way you could pack a non-switching supply into the space available for a PC PSU.

Providing over-current protection isn't difficult. It's just more components. Doing it without sacrificing substantial efficiency is a bit tricky.

[BladeRunner]

The best I can measure the internal psu temp, with an infra red laser temp device, the hottest thing in my PSU is a coil at 42C when the system is under load.

Imo all current PC PSU's whatever make, are generally pretty poor by design and in general build. I have a PC power & cooling 600 watt PSU here and even that looks shoddy in the component layout and build. Until there is a real demand for a fanless PSU of quality construction with high power output ratings without the heat issues, they will keep making the current cheep & inefficient designs. The fanless ones that are availble now are usually low watt rated, and or will get too hot anyway with a hi performance PC.

There are things in a PSU that get hot once the main items are cooled effectively, especially when the forced airflow is totally removed. This is why I made my design partly cool the transformer in the block mounting. The other thing I had to do was thin out the wiring inside and make a bracket to the block to help cool a coil that was getting far too hot under load 70+C

Water-cooling a PSU is a much more complex thing than any other part of the system mainly because they are not designed for zero airflow. zero airflow is a world apart from just a fan near it at 5v. Mines almost a year old but it could still die tomorrow for all I know. I've done the best I can and as far as I can tell nothing is getting overly hot now, but some small part might be..... leading to a reduced service life.

At the end of the day I'd much rather have gone out and bought a much higher quality PC PSU that was designed from the outset to be fanless, powerful and cool, even if it were 4x or 5x as expensive, but as there isn't anything available, the only solution left was to work on what I had towards my zero fan silence with performance goal.

[gmat]

The master has spoken

Now you see why i have chosen the "real fanless" route...
Putting up a combo of small-profile fanless PSUs should provide plenty of power with little trouble.. or i hope so
More news on this in 2003 - RSG gave me 4 to 5 weeks of delay before shipping.

PS Bladerunner: do you still run it on top of your case ? Or did U finally put it in its original place ?

[MadDogMe]

That's the first I've seen of the coil being cooled, it was'nt on the 'fanless' site. when are you going to get a forum up and running?, or what's the best forum for keeping track of your mayhems?...

[Brians256]

This has got to be one of the best currently active threads, so keep up with such great informative posts!

I will add a bit of info to the MOSFET question, however. The heat generated by the MOSFET's is related mainly to two characteristics: on-resistance and switching speed.

The on resististance is shown as RDS(on) in the spec sheets, because it is the resistance between Drain and Source when the MOSFET is fully on. The lower this number is, the less power is wasted during the time when the MOSFET is supplying energy to the PSU output filters.

The switching time is usually specified as maximum frequency, and it is the time that it takes for the MOSFET to transition from fully off to fully on. A higher frequency means that the MOSFET can transition faster, and it spends less time in between the ON and OFF states. Why is this important? Because when the MOSFET is neither fully on or fully off, it is passing current at non-low resistance. At higher frequencies, the MOSFET could spend significant percentages of it's time in the transistion state, where it is wasting energy as heat. This speed/frequency is also related to the gate capacitance and the energy losses can be even worse if the driving circuitry cannot quickly overcome the gate capacitance.

A switching power supply attempts to keep the MOSFET either fully on (with low resistance) or fully off (with near infinite resistance). In either case, the power loss is low.

Another side benefit of faster MOSFET transitions is that the inductor becomes more efficient. Inductors store energy when the voltage changes. The faster the change, the more energy is quickly stored by the inductor.

There is indeed a whole science and art to this subject. If you want a very simple low power SMPS circuit to examine, look here for a cheap example. That is a single voltage low power circuit and I still don't fully understand it. I've been looking at it for the last two weeks and I still don't understand it completely. I don't consider myself slow, and so I think I am beginning to understand why these power supply engineers pull in a good salary. It's complicated.

[bigben2k]

Interesting circuit.

The Zener sets the output at 5.6 v minus a 0.7 V drop at Q2 (or Q1). The coil compensates for the negative AC wave drop near zero, and C2 for the positive (or vice versa), While C1 actually provides the power to them (I think?).

1N5819 is the back current prevention, and the little 47 uF filters out the remaining ripple.

I'm not convinced that the output is better than a 7805, but it's worth a test run. Now if I only had a scope...

[Brians256]

Yes, it is interesting! The basic circuit isn't hard to understand. However, the touchy items are how to balance C1 and C2 and R1 and R2 in Roman's circuit. I'm not quite sure how to optimize them without a scope. My 1MHz portascope (TPI multimeter with a builtin scope) may be too slow or have too low of an input impedance to accurately characterize the circuit. Plus, I don't have any test loads. High power resistor dummy loads are expensive!

Very cheap though.

As for whether it's better than a 7805, it depends upon whether you value efficiency or not. A 7805 will have less ripple but it will also waste power. If V(in) is 12V and V(out) is 5V, and the output is 500mA of current, then, at best, the wasted power at best is about 3.5W. The 7805 would be supplying 2.5W of power while wasting 3.5W! That gives you an efficiency of about 42%.

The 7805 will, however, provide overtemp and overcurrent protection.