DIY Low-power PSU For Home Server Use


[Viktor] decided to replace his old power hungry home server with a model that is much easier on the old electric bill. The new motherboard uses an Intel Atom chip and consumes far less power than its predecessor. He figured there was no reason to use a bulky ATX power supply when all he needed was 12V for the mainboard and a pair of 5V rails for his hard drives, so he decided to build a PSU himself.

He sourced a 100VA toroid transformer as the basis of the power supply due to its popularity with audio amp builders, adding a standard bridge rectifier and smoothing capacitor before regulating the DC output. A pair of switching regulators were added, one for the 6A, 12V, and a second for the 1.5A, 5V supply. The motherboard only requires about 18W at full tilt, so the PSU should be more than sufficient for his needs.

Schematics and board layouts are available for free on his site, if you are in the market for your own DIY low-power PSU.

Looking for more build to suit electronics?  Check out this DIY amp we featured just the other day.

[Thanks, Chris]

26 thoughts on “DIY Low-power PSU For Home Server Use

  1. Or.. You know… You could buy the PicoPSU power supply.

    Hacks aside, sometimes you have to wonder why a more expensive solution is chosen over a cheaper alternative. As this is essentially what the guy made, only in a form factor that can easily be switched out to other boards should the one fail.

  2. I’d be interested to see what a Kill-A-Watt or other power measurement meter clocks his rig w/ a standard PSU vs. his home brew PSU…

    I’d also like to point out that I’m using an 8 year old Pentium-M HP laptop as a server of sorts at my house… it’s been running since 2003 w/ very little interaction (other then network stuff). It consumes 22W under normal operation (lid closed) which includes a 4200rpm HDD.

  3. @hmbemis – I run a pair of laptop boards, each with no LCD, optical drive or anything but a shared 120W 19V adaptor between them… They mount quite nicely on their own LCD back cover with standoffs and hot glue :) One was a DV6000 (chipset repaired- Quite nice with 4GB DDR2-800!) and a GM965/P8100 (but limited to DDR2-667… Lame)

    But +1 for low-power servers instead of 400W idle gaming behemoths left running all day to download torrents and stream video over internets.

  4. Too bad there is still a large power lose in the step-down transformer connected to the AC mains, that reduces the efficiently over a purely switching mode power supply (SMPS).

    The design is a traditional unregulated DC linear power supply as its front end (bulky power transformer, diode rectifiers, and filter capacitor) and two DC-DC modules (switching) from Texas Instruments for voltage regulation.

    It was a well done project, and I understand the design decisions, but it is an unusual power supply design.

  5. I believe this may have missed the efficiency boat. I don’t see a flyback/switch mode circuit on the front end. I see AC mains being stepped down, bridge rectified, smoothed, then put through a switch mode module. So the benefit of a flyback style circuit is lost regardless of the fact that after being stepped down some switch mode modules are being used.

  6. lets take $50 in components and make a psu that would cost $10 on ebay, print pointless PCBs and spend god knows how much more a year on power THAN put it all in a PSU case
    i love his logic

  7. Alright, let me address some of the comments.

    Price: I got the toroidal transformer and the RIFA cap for free and TI boards relatively cheap. Agreed that this probably isn’t worth building if you have to buy all the parts though. But I’m not really telling anyone to do so. This is just a project log.

    Just buying it: See above. A picoPsu or similar bundled with a reasonable quality SMPS adapter brick would’ve cost more than I spent on this project. And while It’s true that you can get a 12v some amp chinese adapter for peanuts I don’t think I’d want to run my 24/7 home server off one ;)

    Efficiency: According to my Kill-a-watt clone the board idled at 12W w/o an HDD. The guys at silentpcreview measued 11-12W in idle with this board and a laptop hdd using a Seasonic SMPS adapter. According to the manufacturer that HDD consumes 0.6W in low-power idle mode (or 1.9W while read/write). Either way the difference is pretty negligible.

  8. the supply he built is just as “bulky” as an atx psu. it even looks like it’s in a atx psu case.

    also, as a few other people have said, even the cheapest atx psu is going to be more efficient than that configuration. atx uses a switcher to step down the 120/240 to 12v where he’s using a toroidal…

  9. Hey now, lets all remember that home projects do not have to be justified like projects in industry. If he want to build a power supply for the experience, then let him.

    Of course buying a COTS component is always preferred, but that is not the point of this website.

    Ok…. I’m stepping off my soap box

  10. @eV: Size was never really an issue for me. You’re right that it’s in an old regular size psu case.
    ATX PSUs are not terribly efficient despite being SMPS (and efficiency tends to go down with price). But I’ve never tested the board with a regular ATX PSU so I can’t really say how much of a difference there is.
    I am quite happy with the power consumption being almost identical to reviews using better SMPS adapter bricks though.

  11. @Viktor: So long as you got what you wanted from it then good job. To you, and interested readers, I would recommend researching flyback circuits at some point though to get the science of why a (properly designed) SMPS is more efficient. You were able to achieve good enough numbers but to squeeze more out of it you’d have to go the SMPS route.

  12. Wow after all that work and it’s still a fail on the objective. I guess it was a good learning experience and a fun hobby project.. Kind of like my 120 watt FM transmitter linear amplifier and notch filter cans… Cant use them legally but it was fun blasting the local rap station out of the water with country music for an hour.

    Atom motherboard + green PS = dual core server that idles at 8 watts and peaks at 21-25 watts. 4 green WD drives deliver 6TB of storage and spin down when not needed. It peaks at 120watts when I get it to ramp up to full load and spin up all the drives at once but that peak lasts only 30 seconds before it settles back to the 21-25 watts.

    The second one I built for zoneminder hovers at 62 watts simply because the 8 chip 120fps bttv card sucks power like no other and zoneminder doing frame analysis is very cpu intensive. But then I have a government class 8 channel video security system that even the $30,000 systems suck compared to it in useability and features.

    Basement 19″ rack went from 3400 watts down to under 240. The whole house audio system still peaks at 1000 watts if I have the garage, living room and basement all cranking along, and I cant reduce the 120 watts that the Crestron procesor uses.

  13. “[Viktor] decided to replace his old power hungry home server with a model that is much easier on the old electric bill.”

    WHAT? Since when is a step-down linear supply more efficient than a switcher?

    Seriously, HaD, you need some quality control. Let’s assume that he is putting a 100% load on this transformer, giving him a power conversion efficiency of about 80%. Now, let’s assume that he is getting ~90% efficiency from his switcher. In this perfect little world, he gets around 70 percent efficiency. However, in the real world, this number will most definitely be lower!

    At best, this is equivalent to a normal switcher. Viktor, I’m not knocking your project, good job on a construction that appears to work well for you. I’m just knocking typical HaD oversight ;)

  14. Kudos to Viktor for building a working power supply, and for taking the time to write up the details so that others might learn.

    But shame to Hack-A-Day. Not for featuring this project, but for falsely representing it as smaller, more efficient, and something other people might want to replicate exactly. For the cost of the parts (or resale price if you already have them) you could buy a PS that really is smaller, more efficient, and more versatile.

    Or better yet, you could use recent “hacks” to power the server from free energy, gathered by homegrown piezo crystals, and rectified by toothpaste. Make sure to document the entire build process with two terapixel cameras, and release it as a 3D video which can only be viewed with eyelid shutter glasses. ;)

  15. I can suggest going to and useing their online tools to suggest a few more efficient suggestions and schematics, their webbench tool is handy for fine tuneing and learning about appropriate components in the right spots, things like low esr or inductance parts….

  16. Honestly, there is really no need for this kind of contraptions anymore. Especially if you are interested in power electronics topics there is ample knowledge around on how to build high efficiency AC-DC converters and how to optimize system power consumption. In general, not just for PC power supplies.

    What Viktor has done here is, and I don’t mean this offensive, use old technology and use it in an outdated way. Toroid transformers, although well-coupled and relatively low-emission, suffer from high magnetic losses especially at low power. They are probably the *worst* choice for low power supply if used in this way (with a huge low-esr capacitor causing peaked current use and a bad power factor). Transformers also suffer from relatively high static (i.e. primary voltage-induced) losses. And then your PCB quality: you’re talking about reliability of that big capacitor, and subsequently soldering low-quality etched PCB with unstabilized statically indeterminate pins on the dc/dc converter module. And if anything, audiophiles are again the worst place to go search for efficiency, those weird people are still using class A/B amps…

    There is SO MUCH knowledge on the internet and it is so easy to come by that it just plain hurts to see people engaging in such designs for all the wrong reasons. And to see them on hackaday.

    Almost all major SMPS controller manufacturers have at least exhaustive application notes, and some even have demo boards, of synchrounous-rectified quasiresonant or LLC resonant switchmode power supplies that can, with a minimum of tinkering or if you are so inclined simulation, attain 90% efficiency or better at these low loads. What is more, you can build such a device either from scratch or by modding a demo board for probably the same money as this device. With input and output protection.

  17. This is not a linear power supply. A linear power supply uses a linear regulator (such as the LM317 or 7805). The efficiency of this supply is mainly governed by the efficiency of the transformer and the efficiency of the switching regulator modules.

    The efficiency of the transformer should be pretty good, considering it’s a bit over spec’d. The cap is a bit large though. Power factor could be lousy.

  18. @Shadyman
    HaD represents themselves as a place that links to other people’s “hacks”, and does a little write-up on them as well. This is a tiny little bit of work to just link to someone else’s project. It isn’t that hard. Therefore, it is surprising to see oversight like this on almost a daily basis. Futhermore, it is surprising to see poor solutions to a problem (like this) published as relevant “hacks” on a daily basis. I can not imagine how many people are misled by this site every day. HaD needs new management.

  19. @Kyle,
    Over-react much? Send them good hacks and get off your high horse. I’m sure the writers don’t have time to do exhaustive research on every project.

    Also, Hacks aren’t always perfect. You might be thinking of an engineering blog. Sure this guy didn’t do an awesome job, but at least he did something.

  20. “But I’ve never tested the board with a regular ATX PSU so I can’t really say how much of a difference there is.”

    @Viktor: Can you do it now? For science sake. If you were right, you can tell all the critics to put a sock in it, and in any case the rest of us will know if we should follow your example or not, or in other words we’ll all gain an insight in how SMPS’s scale their efficiency against power consumption.

  21. OH MY GOD PEOPLE he is using a toroidal transformer connected to two SWITCHING REGULATORS not LINEAR REGULATORS!! he clearly says “Essentially they’re complete switching regulators on a piece of PCB” when describing the regulator modules he used. I guess nobody reads things anymore.

    For the record, transformers are surprisingly efficient and often have efficiencies in the high 90% range. this paired with some high-efficiency switching regulators like he is doing could be quite effective; if the transformer and regulators have 90% efficiency each (he claims the regulators can reach as high as 96%), that’s an 81% overall efficiency (0.9^2) and enough to beat even some high-end Corsair power supplies.

    I’m not saying he actually does reach these efficiencies in real-life since it depends on a number of factors, but come on people, give it a chance!

  22. Nice project. Some people don’t understand the joy of building things yourself. Yes it can cost more than an off the shelf product but you don’t learn anything do it that way. You can read all the books you want but until you build things yourself you will never understand electronics.

    I will be glad when power supplies in pc switch to using one voltage. There is no need for all the different voltages we have now, it is a legacy thing. Nothing is powered directly from 5V or 12V in the pc except the 12V fans. Everything takes the 5v or 12v and converts it to 3.3, 1.8, 1.3 or lower. A single 12VDC supply could power everything in the pc and some things can already do it.

    Some hard drives have converters that while wired to 5V on the connector are within specs that would allow them to run off 12V only without anything but a soldering iron modification. Even external devices like routers and modems don’t use 5v or 12v internally. All these devices convert it to lesser voltages. I repaired a router the other day that actually had stamped on the board, DC input 5~18VDC for the input because it connected directly to a smps supply that output 3.3vdc.

Leave a Reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.