Running from DC (48V up to 60V), pico-psu (DC-ATX) considerations, alternatives.

[Mithril]

Feels like the right forum, but I could be wrong.

One of my "slow burn" projects is moving all of the critical gear in my rack to DC input, using the "normal" UPS output for unmodifiable(or not yet modified) equipment mostly. This allows for much better "on battery" time during power outages, in some cases much better efficiency than a terrible power brick or cheap PSU (yes, I could buy high quality replacement power bricks or PSUs, but if I'm going to spend the money I might as well achieve multiple goals).

I'm currently using a few DC-ATX PSUs, one is a "name brand" (of sorts) pico PSU, and another 2 generic/ebay jobs. I've been looking at getting into Lithium Iron Phosphate batteries which are "48V" (nominal charged voltage is about ~52v and charging voltage can get close to 60V). Most/all of the "wide range" DC-ATX PSUs cap out around 28v/32v. Does anyone know of quality DC-ATX/pico PSUs that can tolerate 60v? Or am I better off with a quality DC-DC to step down to ~12.1-12.2 (to account for some droop) and use 12v DC-ATX PSUs which expect a steady 12v supply.

I would expect feeding even a high quality buck converter into a 12v DC-ATX would require some additional filtering and bulk-caps close to the inputs. Am I better off using a single high quality high capacity buck converter for 12V output, or one converter per PSU? (assuming at this stage we are not dealing with adding redundancy.

Unfortunately most of the more commercial DC-ATX psus seem to be telco standard, which means positive ground (AKA, -48V, not 48V).

 

[edge]

I would suggest a higher capacity UPS over going to DC. You are likely to lose any efficiency gained by DC to the isolation circuitry required in your devices' DC power supplies.

 

[Mithril]

I would suggest a higher capacity UPS over going to DC. You are likely to lose any efficiency gained by DC to the isolation circuitry required in your devices' DC power supplies.

I'm trying to tackle both angles and figuring out where to spend money now. With simi quality DC-AC-DC you're facing a double conversion in the range of 85-95% at both stages, with a final efficiency of 0.72 - 90% (and possibly lower). A decent buck converter, even with a wide difference between Vin and Vout can be 80-90%. The issue with ALL of these is finding quality units. Be that quality DC-ATX (like picopsu) that can tolerate up to 60v for safety, or quality DC-DC buck converters. Or quality inverters, which very much has an efficiency of scale in $ to output ratio. Considering that this system will at some point also have solar and be powering more things, efficiency of power use while on battery is important. The Inverter in my UPS I think is around 65% at lower load and effectively has a cutoff around 10% where efficiency drops rapidly.

the 2 most critical systems are the HA firewall pair, and even with a very expensive platinum PSU are going to be operating in the less efficient range due to low power draw. Keeping the NAS on for 1+ day would be nice but not required.

I don't believe even the "extended run" rackmount UPS that take 48v are intended for multi-day runtimes, and 48v is much easier to deal with once it's outside of the unit.

There is a LOT of quality 48v gear out these days due to popularity of solar systems. I had hoped that included more consumer focused DC-ATX.

 

[colinb]

@Mithril, did you manage to find anything?

At home we are just in the process of installing PV and a very large "48V" (52V in practice) battery bank (24kWh initially for 3days' power, 52.8kWh in a future upgrade for 6days all assuming no changes in behaviour in event of a power cut). Keeping as much on DC as possible and avoiding the losses associated with repeated conversion feels like a no-brainer, but as you've noted, the devil is very much in the quality details!

 

[Mithril]

@Mithril, did you manage to find anything?

At home we are just in the process of installing PV and a very large "48V" (52V in practice) battery bank (24kWh initially for 3days' power, 52.8kWh in a future upgrade for 6days all assuming no changes in behaviour in event of a power cut). Keeping as much on DC as possible and avoiding the losses associated with repeated conversion feels like a no-brainer, but as you've noted, the devil is very much in the quality details!

Nothing yet. I think it's more likely to find quality "48v" (as you noted, higher in practice, and even higher charging!) to 12v/5v/24v DC-DC units.
I've been distracted by other projects

 

[armandh]

Power quit during a storm 3 days ago.
Double conversion [AC-DC-AC] 1KW Liebert UPS [purchased used] held just fine while I hooked up the [large barely portable] generator.
Cable/internet went out with the power for 9 hours. The AC-DC-AC held through the 2 second brown outs when the 5 ton cooling cycled on..
Our second generator use in 8 years at this home. The UPS has connection for optional battery packs. Ive not felt the need.
My home brew XigmaNAS servers have, in the past, been tolerant of power disconnection. For each I have the embedded Xi OS on a SSD.
Those who use a USB stick for this often find they have failed while the OS ran merrily from the RAM-disk. [USB sticks are NG for a swap file]
This tolerance was last tested just B4 replacing the 7 YO UPS batteries. [4 12V in 24V series sets paralleled, each set can be changed live]
Ive never done it that way as I can shut down the servers,

to be sure the gasoline generator would be ready for its next use I ran the cool down on [corn free] low lead 100 Aviation gas then ran it dry.

 

[t123yh]

Take a look as this: GitHub - t123yh/atx48

This might be a perfect solution to power your PC by 48V.

 

[Mithril]

Take a look as this: GitHub - t123yh/atx48

This might be a perfect solution to power your PC by 48V.

Interesting, I couldn't find price for the main module (I assume thats at least 50% of the BOM right there). Any idea on cost for the parts and board if someone was looking to build one or a few? This might be overkill for some of the machines I'd want to use it on be if the price is right.

FWIW my current thinking is a quality 48V to "12V" (with an offset/adjustment so its about 12.2 +/-1 0.1v for line drop etc) "perfect diode" (mosfet modules used in solar, good to ~15amps) OR-ed with an AC powered 12V supply. I don't currently need to keep my nas or file server running super long, they can just gracefully shutdown same as now.

I likely am going to get an inverter for AC loads still, but A) that's still an extra efficiency loss (DC - AC - DC) for computers and B) the larger the capacity the more expensive and higher the standby power (generally). Plus if I have any inductive loads (like a fridge) I'm worried about how "clean" the power stays for sensitive loads.

 

[MBuisson]

if this is still relevant, you can have a look here : HDPLEX 800W DC-ATX with 12V-63VDC Input
I use one myself for my DC solar powered PC for about 3 years plus a 400W one for my NAS, 4 years.

but, as @edge already said, be carefull of isolation. We are used to every PSU beeing isolated. Going DC can changes that.
if you have only one device then you most likely won't care. if you have more than one device then things gets tricky. You get grounds differences between your devices - and those cause curretns to flow trough the data cables that link them.
the above PSU at 50V may cause currents in the 16A to flow (more with efficiency consideration). This will cause a drop in the negative/return line. if you have a second non isolated DC/DC with much less current it will also experience a lower drop. once you connect those two together (eg : my desktop and my NAS trough 2xHDMI cable on a single monitor) then one has a ground of mostly 0V (the NAS, 50W) while the desktop has about 300mV (500W). The HDMI cable connected both reference together: part of the desktop return current started to flow trough the HDMI cable (around 2A DC). My monitor didn't like it and one of its two HDMI input now look funky due to overheating.
So, no HMDI cable equal no issues ?
next in line was the ethernet switch: Lan cable DC current was somewhat lower at 1.5A and didn't cause damages, but I switched to optical fiber just to be sure (no copper mean no ground current !). I was planning to upgrade my 1Gb network to SFP+ anyway, just made me do so earlier. I think POE switch (which have isolation on the data lines) would work too.

that being said, it's great for a solar based PC. Efficiency is similar to a platinum ATX PSU but you have no inverter losses (cable loss can hurt instead). I cool it with a usb 40mm fan that is barely audible to keep temperature below 40°C (it reach 70°C in passive cooling which impact MTBR negatively). Your inverter is also smaller because a 800W load has been taken off its shoulder. That's a lot of money saved. And I have a 12kWh UPS attched to it with 12*0,95=11,4kWh usable by the PC instead of 12*0,95*0,94 = 10,7kWh: 0,7kWh more battery. knowing the price of batteries this is also not small.

 

[gregsachs]

Industrial practice is typically to ground dc negative, I would absolultely do that if I was running multiple pieces of equipment off of DC or using multiple DC supplies. As you found, DC can float and that can be a bad thing...

 

[MBuisson]

Seeing how you went straight into the pitfall I have been warning about I guess my explanation was kind of lacking. To be fair it’s a nasty one - I fell in it myself and I have 14 years of experience as a power supply designer. But AC/DC.

In essence, this in not a AC/DC power supply (alway with insulation) but a non insulated DC/DC. In this regard there is no standard practice because it is also not in use in the industry : the so called “DC” datacenters use insulated DC/DC. This simplify their usage quite a lot.

Floating is not possible with non insulated converter. Everything is grounded from the power supply Input side rigth from the start. Because it’s low voltage / high current it’s even extremely well grounded: the impedance between the two grounds is the cable impedance and is extremely low (0.017Ohm).

And that’s exactly the issue. You may externally improve the grounding but for it to have an effect you need to lower the impedance bellow that of the input return cable. Only then would the voltage drop between the two systems get low enough that your data cable are safe. but then you have current carrying ground cables, which is also not good industry practice.

This lastly bring safety concerns: if the return power cable of the strong system gets interrupted you expect the system to turn off. With such grounding the return of the weaker system will be used to carry the current of the strong one (using the grounding cable) and all will keep operating. The weaker system may however be exceeding its current rating without being aware of it. As a weaker system it may have lower cable gauge and weaker connectors - fire, fire, fire.

to keep it short;
IF you use non insulated power bricks
THEN do not connect electrically two or more system ground together (nothing conductive, optical fiber is fine)
OR use insulated DC/DC and all is like usual (but they’re expensive and ATX one have poor efficiency on top)