Daydreaming (nightdreaming?) of ATX12VO for redundant power delivery with actual 12VDC supply

This ATX12VO stuff is starting to roll out, and I am feeling a bit hopeful (in terms of what my next server build will be)…

This board in particular isn’t part of my shopping list, since I’d rather see an AMD Epyc board with ATX12VO instead (although, initial challenge is that Intel is responsible for that spec - and while the enterprise would benefit from the power efficiency that Linus noted in his video, I can see the traction be slow) for the IPMI creature comforts (because server in unfinished basement with a duck-low ceiling).

I am also working my way into photovoltaic solar (roof panels for grid-tie coming “soon”) so figuring out a comprehensive - and inexpensive - battery bank to introduce at a later point could mean many hours when an inverter need not be involved.

There’s much more I could say about the idea, but this is probably sufficient for now.

Edit to add (and save confusion/controversy): I am not seeking ATX12VO systems to slap in 12V straight to the board and call it a day - I see the rollout of the ATX12VO standard to prompt various solutions for solar PV users, small-scale enterprise, etc - through PSU’s manufactured to accept DC power.

I hadn’t quite realized how nice this spec is for people running off of DC systems, since they are generally all 12v or have 12v output. No need for a possibly sketchy pico PSU anymore, all you need is straight wires/adapters.

Also, the higher idle power efficiency could help, although it depends on your setup, and I am not sure how much applies after you take the PSU out of the equation.

For me, I would see the big benefit out of having the proper ATV12VO PSU to still use for mains power, and then have 12VDC in from a battery bank as the redundant power source.
I was discussing this with a friend who is involved directly with enterprise environments, and he did mention how common 48VDC has been for telecomm purposes anyway. Since the power demand of a server rack will typically exceed a helpful wire-gauge at 12VDC, running more 48VDC in a datacentre to then have a buck-converter as your redundant PSU seems like a useful way to go about it (in commercial setups small enough that may not justify diesel generators set to automatically start up in a power outage).

A cheap way to bulk up your battery capacity would be to find failing UPS units (ones that charge but fail to perform inversion), fully-populate their internals with sealed lead acid batteries (such as the SLA12-9’s) - to tap in “from the back”. I have did a bit of a remixed idea to 8-Bit Guy’s old UPS video, but my change mostly focuses on re-routing the power button to allow it to be controlled remotely (in my situation, the RPi that will be powered by it). Photo below.

Edit: a further note for this project is that the RPi 3B+ that will use this as its redundant supply, that is being done through a MoPi2 HAT, and to be protected with a 5A fuse (though for 12VDC, a 2A fuse would have been plenty - if one exists in a blade form-factor). I plan to add LED lighting off this at a later point (for power outage scenarios), but I plan to be rather careful about load since I don’t want the load to fluctuate as much as I don’t want the batteries to exhaust themselves after only a few hours.

The more expensive solution (but likely one for small datacenters) could be to have the bottom of each cabinet (up to 6u) populated with rack-mountable Lithium Sulfur (Li-S) batteries, like what OXIS sells - those are generally 48V with 3000Wh of rated capacity in a 3u. Compared to a 1000VA APC UPS 2u unit, those would be rated at around 400Wh. Li-S has been one that slipped below my radar (I think it’s popular in Australia which may explain why I - at the other end of the globe - have heard little about it) but it gets touted as chemically tolerant to physical damage (won’t go Note 7 on you), has fewer ingredients (apparently just as the name already give away), which does mean no reliance on cobalt.

I note all this at this point because - I do not plan to buy a new server for a few years (this year I put in a couple used HP Proliant systems, a DL385 G7 and a DL180 G6) and while I know that this is the last run of HP servers I will bother with (anything newer will mean acquiring a new pool of spare parts anyway), my next generation seeks these goals (and having ATV12VO come out sooner means better affordability for cast-off gear when I go to buy it years later).

You’re not getting rid of the DC-PSU, you’re just putting it right on the motherboard, so you have a much more expensive part failing.

Besides that, 12VDC is nasty! Think of cars that drop to 10V when starting, then jump up to 14V when the alternator kicks in… And it’s dirty, noisy power.

And it’s terrible for distribution, too. Huge, thick copper wires for short runs, and expect 30% voltage drop across even a small house. In fact I noticed travel trailers shoot up to 18V to overcome the voltage drop across that modest distance. I wouldn’t let sensitive electronics anywhere near it… Let your cheap pico-PSU take all the damage and supply nice, clean consistent voltage out to your sensitive electronics.

Besides, most folks with PV solar panels skip the DC step entirely… Much higher than 12V coming out of the panels, goes directly into micro-inverters, and the 120/240V ties directly into the grid. No particular ability to produce 12V anywhere.

1 Like

This

Low voltage DC systems only work when you are drawing a few mA.

As soon as you are drawing Amps, it all gets rather wonky with contacts welding shut and all the good stuff.

1 Like

you might want to look into how most thin mini itx boards, 5x5 and deskmini boards have their power jacks

often they have a laptop 19v jack on the IO plate AND a 2pin connector on the board to wire up your own jack


I was thinking more of low power embedded type systems running off of solar power or similar. Say on a boat, or an RV or something.

I’m not taking about a 500w gaming system, but a 50 watt low power system.

1 Like

Yes, possible.
However 50W / 12V = 4A while 50W / 48V = 1A, that is the difference between any piece of wire (21 AWG) or dedicated beefy copper (15 AWG).

Keep in mind that 15AWG “beefy copper” is your bog-standard IEC cable between your PSU and the wall - beyond all the insulation reveals less copper per lead than lamp cord provides.

At any rate, for short runs (up to a few meters along the same server cabinet) 10 AWG would be the functional limit for being any hope of workable. To plan for less than a 2% voltage drop would mean less than a 6A load for a 3m run (round-trip to ~20ft) with 12VDC, or 23A with 48VDC and a buck-converter in the server as its redundant “PSU”.

As was mentioned elsewhere in the thread about “dirty power” - this is a concern to the point as to why most UPS units will refuse to connect to personal generators - if the motor needs the governor to rev it a bit to take up the new load, it’s just too-little-too-late. Traditional PSU’s (including ATV12VO power supplies consequently) have capacitors to smooth this stuff out - if they didn’t, starting your clothes dryer could make your current rig suffer a brownout.

This is probably the more important part of my use-case (that I will pop in an edit up top shortly) - I am not seeking ATX12VO systems to slap in 12V straight to the board and call it a day - I see the rollout of the ATX12VO standard to prompt various solutions for solar PV users (even if they are just trickle-charging their battery-banks from mains), small-scale enterprise, etc, through PSU’s that accept DC power. There are many scenarios that won’t make it practical (such as at your computer desk - unless you have a dozen wall-worts for peripherals that could be consolidated - and even that is a tangential thought) but short runs between the backup/power source and what is to be powered would be more practical than an inverter converting DC back to AC, just for any PSU to convert it back to DC once again.

Good luck to who ever may attempt that. I have at least 6 different voltages comeing from various powerbricks and wallwarts.

Am aware of that. I am just saying, building your own in house grid for more than LED lighting using a 12V DC bus is not a project I would see as worth the effort.


I also have a very strong opinion on Intels 12V-Only PSUs (TL;DR: In a time when motherboards can’t even fit a proper heatsink for a 15W heatload, do we realy trust them with handling 20A of 3.3V and 5V DC-DC stepdown?)

1 Like

Same sentiment here TBH - there’s no point in me trying to pull off a common power in my office (or living room) when I see gains better sought as backup power in a basement rackmount unit.

I hear that - which is why if a PSU manufacturer pandered to people who do - to be more definite about it: if Eaton or APC manufactured rackmount UPS units and ATX12VO PSUs meant to connect to each other via a 48VDC line - marketed as an efficient battery backup solution for computer equipment that desires a redundant supply anyway - there would be edge benefits in other markets.

I feel that is a rather moot point - since the Pentium 4, the 3.3V rail needed to be stepped down further by the motherboard for the CPU anyway. 5V rails usually don’t get supplied more than 3A. If the motherboard VRM was handling most of the low-voltage current for the CPU anyway, what real difference is there to pass off the load for powering SSD’s to the motherboard as well?

1 Like

Not sure about that.
NVMe drives like their share of 3.3V (Samsung 970 Evo says 2.8A @ 3.3V), as do USB ports for all the RGB-nonsense (reminder: USB 3.0 bumped up the current limit to 900mA).
If PSU makers could get away with cheaping out on the minor rails, they would.

Vcore VRM gets 12V from the PSU anyway.

1 Like

My bad. I was glancing at the rails table on an EVGA 750W supply and only saw the +5Vsb load (3A). For completeness, that 100-N1-0750-L1 PSU provides +5V (20A) and +3.3V (24A) for a collective maximum load of 150W, +12V (59A), -12V (0.3A), and that +5Vsb (3A) for that grand total of 750W.

1 Like

I have fallen for that before :wink:

The “my pc is a USB charger” 5V stand by current…

1 Like

Coincidentally, the Lenovo M92p tower I used to use has a 14-pin ATX connector that the M93p uses (and noted on wikipedia as being a system with a 12V-only power supply - though obviously not ATX12VO) and while the power delivery on that was subjectively painful (no connectors for a discrete video card; only two power breakouts from the board - which I think resulted in two molex and two SATA outputs), in the cookie-cutter implementation of an office workstation, the 280W power supply was quite redeeming.