"Hey guys! Let's design an SMPS that performs like a linear PSU!"

[Phaedrus]

This 450W +12V power supply is amazing. 11.36V at full load. I'm not sure this even has properly regulated outputs.

[NiHaoMike]

What is it designed for? I have seen a similarly performing 12V PSU (though only 60W) in a cheap portable refrigerator. (It runs on 120V AC or 12V DC.) Basically just a self oscillating half bridge, no feedback whatsoever. Works just fine for its intended purpose.

[Phaedrus]

This is an ATX PSU.

This one actually wasn't too bad; ripple under 30mV on all rails, and the +5V and +3.3V were in-spec. I tested another one today though... It died when I ran it at full load (450W), but here's the waveform at 80% load:




Still in-spec, but barely.

These aren't the worst PSUs I've tested, but they're pretty poor.

[dr.diesel]

What manufacturer?  There are a ton of ultra el-cheapo PS out there.

[Phaedrus]

First is Thermaltake, OEM by HEC. Second is Rosewill, OEM by Deer/Solytech.

[tom66]

Honestly, it's not a big deal...

I've had motherboard VRMs that still work on 9V. I know, because I had a PSU with bad mains capacitors, and average 12V output was around 9V (along with heavy PSU buzzing.) HDD works too. Only things that give problems are fans spinning slower, and the optical drive didn't want to eject.

Yes it probably leaves the ATX spec boundaries but if it works reasonably well it doesn't matter. Ripple is more important, for lifespan of motherboard capacitors and power supply capacitors.

I'd say the manufacturer could consider having the 12V start around +12.5V and drop to +11.9V under full load, that way people might complain less? Even though it makes virtually no difference?

[NiHaoMike]

Most older PSU designs need some load on the 5V to get good regulation on the 12V.

[mariush]

The atx standard says the 12v output must have a maximum of 120mV ripple. For the 5v and 3.3v, I think it's 50mV or somewhere around this value, I'm not sure.

The voltages also have to be within +/- 5% so if you want to call that power supply an atx power supply, the 12v should stay within 11.4v-12.6v.

The power supplies he mentioned pass these rules and actually managed to reach those 450w of output, which mean they're better than some Chinese power supplies with fake labels (overrated) but in comparison with good power supplies, the scope results are poor and the voltage regulation is horrible.

I think that's what Phaedrus meant to say.

Good power supplies can have as low as 5 mV ripple at 12v and output something like 11.95-12.1v throughout the whole range (550-750 watts).

[Phaedrus]

Pretty close.

I haven't posted full details here, because I do work in the field and these are competitors' products, so it's not entirely ethical to go around bashing them indiscriminately. Also, the company I work for has some stinkers too, so glass houses and all that (though I am trying to weed those out of our product line).

I'll just say that both power supplies came with computer cases that cost <US$30 retail. The Thermaltake one performed better than I expected and only failed spec at full load and just barely; but the Rosewill violated voltage regulation specs at 50% load and died when I put it to a full 450W. They're both pretty poorly built, with minimal EMI filtering, outdated topologies, mediocre soldering, have cheap Chinese caps, and get <75% efficiency.

So, they're pretty poor.

I just posted because the Thermaltake's +12V regulation curve reminded me of LDO regulation curves, just higher wattage. :p

[tszaboo]

How much is the cable drop?

[T3sl4co1l]

If the 5V isn't loaded proportionally, it's rather meaningless.

Honestly, I don't know what good it is anymore to have good regulation; there are still many devices running from 5V (hard drive motors for instance), but almost everything else is just wads of point-of-use converters.  That power connector on the GPU sure as heck doesn't feed 12V directly into the core, it's got a ~1.2V converter on it.

But alas, things still poop the bed when bad PSUs enter the mix, so there must be something that hasn't caught up.

Tim

[amyk]

Honestly, it's not a big deal...

I've had motherboard VRMs that still work on 9V. I know, because I had a PSU with bad mains capacitors, and average 12V output was around 9V (along with heavy PSU buzzing.) HDD works too. Only things that give problems are fans spinning slower, and the optical drive didn't want to eject.

Yes it probably leaves the ATX spec boundaries but if it works reasonably well it doesn't matter. Ripple is more important, for lifespan of motherboard capacitors and power supply capacitors.

I'd say the manufacturer could consider having the 12V start around +12.5V and drop to +11.9V under full load, that way people might complain less? Even though it makes virtually no difference?

Agreed, in fact I'm willing to bet that the majority of PC PSUs out there are barely within spec and yet still powering their loads for years with no issue. No doubt there will be duds and the chance of a dud is more with a generic no-name one, but overall, for the average PC I don't think it makes that much difference.

As for their rated power... this experience I had might be a little enlightening.

[iampoor]

If your spending thousands on PC parts, The minimal cost of buying a well made PSU is worh it, just for piece of mind if no anyhing else!

[Rigby]

So what's a good ATX power supply, then, in terms of rated load vs. performance, ripple, and efficiency?

I imagine that given three specs there will necessarily be three answers, and that there is no supply that is superior in all three aspects.

[Phaedrus]

So what's a good ATX power supply, then, in terms of rated load vs. performance, ripple, and efficiency?

I imagine that given three specs there will necessarily be three answers, and that there is no supply that is superior in all three aspects.

A good PSU would be, for instance, the CoolerMaster V1000

Rated wattage: 1000W
Maximum output wattage: ~1300W
Rated temperature: 0-40 degrees Celsius (OEM rates it for 50 degrees)
+12V regulation: +/-1%
+5V regulation: +/-2%
+3.3V regulation: +/-2%
+12V ripple: <35mV
+5V ripple: <20mV
+3.3V ripple: <20mV
Efficiency @10% load: >85%
Efficiency @20% load: >90%
Efficiency @50% load: >91%
Efficiency @100% load: >89%
+5V standby efficiency: >80%

All electrolytic capacitors are Japanese (NCC and/or Rubycon), and solid polymer capacitors are used where practical
Uses an LLC resonant full bridge topology with the +5V and +3.3V buck regulated from the +12V


Our top competitors have products with similar specs.

[Fsck]

So what's a good ATX power supply, then, in terms of rated load vs. performance, ripple, and efficiency?

I imagine that given three specs there will necessarily be three answers, and that there is no supply that is superior in all three aspects.

A good PSU would be, for instance, the CoolerMaster V1000
...snip...
Our top competitors have products with similar specs.

Looks like a Seasonic platform, but the performance of it says there's likely been a couple of changes.

But a good/awesome/etc PSU isn't the only thing that should be safeguarding your rig, you should have at least a surge protector in front of it, or a UPS, or even better, both!

[Rigby]

But a good/awesome/etc PSU isn't the only thing that should be safeguarding your rig, you should have at least a surge protector in front of it, or a UPS, or even better, both!

well, I wasn't even talking about use with a computer.  I was thinking of using something like an ATX switching supply and passing what I need through a linear regulator so that I can get linear-like ripple and regulation without the massive transformer and as much inefficiency.

[Phaedrus]

If you use only the +12V rail, the +5V/+3.3V buck regulators go into standby modes, and efficiency goes up another couple percent. If you're pulling ~500W,  you can get 92-93% efficiency out of a V1000, using only the +12V rail.

[ajb]

First is Thermaltake, OEM by HEC. Second is Rosewill, OEM by Deer/Solytech.

Ahhh, Deer.  I seem to recall those being infamously terrible, with many forum posts detailing catastrophic failures or just general shittiness.  That was a few years back, so maybe they've changed their ways?

I remember hearing that some manufacturer--may or may not have been Deer--that was so cheap that they used an RC delay for the ATX power OK signal, which usually worked fine. 

As for using ATX PSUs for non-PC loads, my experience is that they are not very good at handling large transients, even when operated below their rated output.  Adding bulk capacitance helps, but even then a very dynamic load is liable to cause nuisance tripping of the PSU's protection systems.  If you really need something in the hundreds-of-watts range with high efficiency a proper single-rail enclosed PSU is a much better bet.

[NiHaoMike]

As for using ATX PSUs for non-PC loads, my experience is that they are not very good at handling large transients, even when operated below their rated output.  Adding bulk capacitance helps, but even then a very dynamic load is liable to cause nuisance tripping of the PSU's protection systems.  If you really need something in the hundreds-of-watts range with high efficiency a proper single-rail enclosed PSU is a much better bet.

At the last two companies I worked at, they had the practice of buying 1kW PSUs to run FPGA boards that use maybe 100W at most. It's apparently cheaper to buy quality PC PSUs than specialized PSU modules. Although I haven't seen it happen, I suppose the main drawback is some rather exciting fireworks should something short circuit without enough fault current to trip the overcurrent protection. (If I were designing the boards, I would at least put in some fuses.)

[smashedProton]

the motherboard already cuts it down to 1.4ish volts so...

[IanB]

well, I wasn't even talking about use with a computer.  I was thinking of using something like an ATX switching supply and passing what I need through a linear regulator so that I can get linear-like ripple and regulation without the massive transformer and as much inefficiency.

The typical linear regulator is good at removing mains frequency ripple (which is practically DC in frequency terms), but is less good at rejecting the high frequency noise from a switching power supply.

Also the ATX supply is grounded, so you don't get isolation.

Lastly, the ATX supply is designed to supply fairly constant, unvarying loads. It will have fairly good regulation in response to supply voltage changes, but it will not be designed to respond well to large load variations, especially load/no load transitions.

Given that you only have a 12 V rail to start with, and the linear regulator is going to subtract 2-3 V from that, it is hard to see the appeal of a big and complex, hybrid 9 V power supply...?

[Rigby]

Well, if you have a lot of 9-10V LEDs, as I do, then this is a good, inexpensive, high current solution.  I could also use power resistors, I suppose.  If a halfway decent power supply is chosen, it won't fluctuate enough to be visible until there is a proper brown/black-out.

Lots of folks always talk about high frequency noise, but why?  Is it common to build circuits as a hobbyist that degrade or fail when fed high frequency noise?  If a high-frequency device like a desktop computer can deal with it, surely hobbyist stuff can, or am I wrong?

[Phaedrus]

HF noise normally won't affect simple hobbyist circuits, especially those using "jellybean" discrete parts. But once you start using MCUs over a few kilohertz, or high-speed/bandwidth op-amps, HF noise becomes a major factor.

Most ATX PSUs emit ripple somewhere in the range of 60-250kHz, and can give off high frequency noise all the way up to 50MHz. ATX ripple measurements are made with a scope in 20MHz mode, for consistency and comparability between brands, but PSU noise doesn't just roll off like that. It can be the mark of quality, between an "OK" ATX PSU and a good PSU when you switch your scope to 100MHz mode and see if the noise increases by 20mV, or by 200mV. Assuming you have the proper leads and connectors of course; or else you'll just pickup the local radio and what-not.

[T3sl4co1l]

Not true: HF noise around jellybean bipolar opamps is very often rectified to the tune of several percent.  So 50mV HF ripple on a supply might not be much for an audio circuit, but a noticeable error for a mic amplifier.  Worse if you're doing something precision or wideband.  And basically screwed if you have one of those awful unfiltered SMPS power adapters from China.

Sensitivity depends on kind; JFET and MOSFET op-amps are more immune than bipolar (which rectify noise right at the input stage).

Sensitivity is most often a matter of input or sensor noise, but (depending on case) there's probably still plenty of opportunity to get the same effect through the power supply.

Tim