linear pc power supply?

[cowanrg]

ok, usually google gives a ton of hits for eveverything. you could put in "frying pan no pants circus banana" and get 200,000+ hits. well, "linear computer PSU" gets 0 hits. i searched around for awhile and it seems no one has tried to do a linear ATX power supply.

i understand its sort of silly, but maybe there are benefits. there is a thread over on audiogon that suggests a linear power supply will make your computer sound and look better, because of cleaner power. well, ill bite i guess. it makes enough sense, being that computer PSU's arent that great, they are probably VERY noisy (RF and electrical noise).

im willing to give it a shot. how would you guys design it? and do you think it would be possible to build a ~300-500W linear PSU in ATX form factor specs? (150mm x 140mm x 86mm). it would be TIGHT. basically, all you need is +3, +5, and 2 +12v lines.

external is always an option, but keeping it small would make for a neat solution.

[mateo88]

I've definitely thought of doing this, but decided it would just be way too inefficient. Figured it would probably help with overclocking somehow and be great for computers meant for audio. I really don't think there's any way it could be done in ATX specs. Even if it was possible to fit all the parts in the case, it would overheat so ridiculously fast it's not even funny.

[luka]

Hi

It is inefficient and it has bad regulation if you use just caps, trafo,diods. But if you use some sort of regulators it will be still too big to put anywhere, because you have high currents.

[AndrewT]

Hi,
a series linear regulator dissipates a lot of heat when delivering lots of current.
Series regs run cooler when the current demand is lower and almost cold at zero output current.

What proportion of the time does a computer demand maximum current from all the supplies at the same time?

What is the real power consumption on EACH of the voltage supplies during normal running?

I suspect those 250W to 300W PSUs are ticking over at less than 50W for 90% of the time.

[cowanrg]

yeah, those were the problems i was encountering in thinking how it would work...

does it HAVE to be regulated? im guessing and unregulated linear supply wouldnt be that much better than a traditional ATX supply...

my guess is that it would have to be external. if that was the case, you could just cheat and do multiple 3A voltage regulators in parallel right? so, for the 30A lines, just do 10 regulators. sure, its not elegant, but it would work and be relatively simple. and since it would have its own case, you could use large heatsinks and just use separate transformers for each voltage requirement.

[cowanrg]

Originally posted by AndrewT
Hi,
a series linear regulator dissipates a lot of heat when delivering lots of current.
Series regs run cooler when the current demand is lower and almost cold at zero output current.

What proportion of the time does a computer demand maximum current from all the supplies at the same time?

What is the real power consumption on EACH of the voltage supplies during normal running?

I suspect those 250W to 300W PSUs are ticking over at less than 50W for 90% of the time.

i had that in the back of my mind... my home theater computer only uses a 330 watt power supply. it runs just fine and is stable as can be. it doesnt have a bunch of hard drive (none, in fact, just 2 flash drives for the OS). the only thing it powers is the motherboard and graphics card. and it doesnt even use fans, so it doesnt have to power a cooling system.

my guess is you could get away with a 300-500 watt supply for most high end audio/video PC's.

[luka]

Hi

does it HAVE to be regulated? im guessing and unregulated linear supply wouldnt be that much better than a traditional ATX supply...

That is not true. It has to be regulated, and very good. In computer there are many things that depend on "solid" voltage, that is why you can't use/have unregulated supply.

I suspect those 250W to 300W PSUs are ticking over at less than 50W for 90% of the time.

I don't aggre with you. It all depends on what you have inside. Like I have dual core proc., graphic card for playing games from time to time, hard drives (many people more than 2), fans,....

So 50w in my case is to low value, but since I have 400w supply it is not working with that power, more like 100-150w. I can't but someone else could measure how much power does computor need.

[Tweeker]

It might be feasable to use transformer based supply feeding switching regulators, though I wouldnt want to try and fit it in an atx psu form factor. A few critical linear regulators could be fed off another winding.

my guess is that it would have to be external. if that was the case, you could just cheat and do multiple 3A voltage regulators in parallel right? so, for the 30A lines, just do 10 regulators. sure, its not elegant, but it would work and be relatively simple. and since it would have its own case, you could use large heatsinks and just use separate transformers for each voltage requirement.

If those regs were LM338s, youd be burning 900+ watts just for the 30A lines. Battery power begins to look sensible.

How about throwing several farads and some batteries at the problem of regulation and varying loads?

[luka]

Hi

Battery would be good, but even battery sags more than you would want + you would need 2. One small for -12 and -5 volts

[Eva]

You should rather forget about such a thing as a linear computer PSU. The computer itself is far noisier than any SMPS, particularly onboard audio is the noisiest thing in the world and the resulting linear PSU will be bigger and heavier than the own computer itself, like a 300W calss A amplifier. Furthermore, very thight regulation is required at least in the +5V and +3.3V rails, and with modern CPU and video chipsets you should expect at least 200W to be drawn continuously while the computer is performing some task.

A simple way to get a very low noise floor is to use an external sound card, although not all of them feature galvanic isolation thus leading to ground loops. This may be solved by taking advantage of digital optical outputs or coaxial outputs with transformer coupling.