[chimpontherun]

I've noticed that ALL the devices I plug into my UPSes have external power bricks. Most of them are either 5V, 12V, or 19V

So, I replaced all my UPSes with LiFePO4 batteries supplied by Victron AC->12V chargers. Routed the battery contacts directly to all devices that consume 12V (WiFi AP, network hubs, SLA 3d printers). Used 12V -> 5V adapters to supply 5V / USB2 devices (R-Pi servers). For 19V, Drok DC-DC boost converters work great.

Result: threw away 3 UPSes (different APC models). Overall power consumption with AC present dropped by about 40%. Time on batteries (same Wh battery capacity) increased by a factor of about 20 (yes, 20 times: that's not a typo). Evidently, AC waveform generation is extremely power-hungry

[tredre3]

> Evidently, AC waveform generation is extremely power-hungry

I've tested a dozen models from APC. The inverter used in those devices uses roughly 15-20W with no load. Then for any load they have about 85% efficiency. Then you have further losses into any PSU connected there because they tolerate square waves but aren't optimized for it. So yes, in the end, less than 40% of the battery capacity in cheaper UPSes is actually usable.

The reason you're seeing 20x is because obviously you've also greatly increased your battery capacity (typical under-the-desk APC units have 70-150Wh capacity, less than half of which is usable as explained above).

> Overall power consumption with AC present dropped by about 40%.

I'm finding that part harder to understand. The UPS consumes almost nothing when AC is on, so that can't be that. You've replaced multiple PSUs by more efficient, bigger ones, sure that can explain part of your improvement. But 40% drop is wild!

[michaelt]

> The UPS consumes almost nothing when AC is on, so that can't be that.

Back in the 1990s, one could buy a "double conversion" UPS that converted AC to DC then back to AC, at all times. This was, supposedly, the best type of UPS (in my experience they were also the least reliable)

[throw0101c]

> Back in the 1990s, one could buy a "double conversion" UPS that converted AC to DC then back to AC, at all times. This was, supposedly, the best type of UPS (in my experience they were also the least reliable)

They are "best" in the sense that your output is completely decoupled from your input so you got the most protection from any electrical noise. The trade-off is lower efficiency (AC-DC-AC roundtrip) and more battery wear (it's constantly 'in use').

Any >10kVA UPS is probably double-conversion/online.

[dreamcompiler]

If built correctly this design also suffers no transition transients. You can switch the external power off/on all day and downstream equipment will never see a glitch.

[jsiepkes]

As far as I know the more expensive UPS models are all still "online" (ie. double conversion) UPS'es.

These are also the only variants which will protect you against things like a phase ending up on neutral in a 3 phase power system. I've seen this happen twice. Fried a lot of equipment.

[ErroneousBosh]

Our UPSes at work are like that, but the smallest is the size of a sofa and the largest is the size of a minivan.

You can put them on street power, you can throw them over to the diesel genny, back onto a secondary genny, back onto street power, and the output won't even ripple.

[Aurornis]

This is called an online UPS and it's still a thing.

It's not a good option for home use because it's always sending power through an inefficient path. The devices we use have power supplies that can handle transients and fluctuations.

[scottlamb]

Where is it a good option? I'd expect datacenters to be more focused on efficiency, not less.

[felixgallo]

Music, medical, engineering. Wherever the waveform consistency and noise floor are the optimization point.

[rdbl27]

Datacenters are focused on "never letting the equipment go down for any reason."

If they can do that efficiently, that's great.

If they have to choose between efficiency and outage risk, they always drop efficiency. That's why they use exclusively online UPS.

[scottlamb]

> Datacenters are focused on "never letting the equipment go down for any reason."

Disagree. I'm most familiar with Google. For a long time at least their UPSs were built in to each machine and not super reliable. They built entire datacenters without cooling with the understanding that for a few days a year it may be too hot for human access. The reliability is from failing over to other equipment—machine, cluster, even region.

Small/on-prem/co-location datacenters may make totally different choices, but I think the major cloud/AI providers would be similar to Google in this respect.

The sibling comment about other industries was enlightening to me though.

[butvacuum]

it's even worse: SLA's can only discharge to ~half their nameplate capacity or so- and if you're pulling tons of amps, it's even less. LiFePo4 otoh you'll get the nameplate capacity.

Then theres trivialities like LFP's voltage being higher so there's less resistive losses, etc

[exmadscientist]

> Evidently, AC waveform generation is extremely power-hungry

Yes, it's quite ugly. Open up one of these things and you'll find a big block of four transistors (if not more due to doubling up) on a big heatsink. That's the inverter drive bridge, and it's probably the single largest source of heat in the whole thing. It's not hard to find.

[ErroneousBosh]

I've been looking at Victron gear for an off-grid power system up north (initially for a caravan, will get migrated across to a house once the upgrading and repairing is done).

A huge decision on that was that they publish a fat PDF with the full spec of the serial data they emit and can be programmed with, and a service sheet to allow a suitably skilled engineer to fully bench test them if there are any perceived problems.

They're not cheap, but my late father always used to say he didn't have enough money to buy cheap tools.

Edit: handy hint too - many cheap crappy power bricks will work down to 90V and will work just as well off DC as well as AC because the very first thing they do is rectify and smooth the incoming supply.

I do not recommend having 90V DC kicking around unless everyone involved knows what they're getting into.

[cwillu]

Yeah, 90vdc is an immediate nope from me: way too easy to suffer a painful death.

[cyberax]

I have thrown out my Tripplite UPS because its battery has degraded to the unusable levels. I replaced it with a 5kWh LiFePo4 rack-mounted battery and an AIMS rack-mounted inverter. I'm really surprised that there are no off-the-shelf solutions for this. The traditional UPS makers are just neglecting the recent 10 years of battery advances.

I even made an Arduino-based module that provides an SNMP UPS interface for my Synology NAS. It works surprisingly well and has almost 12 hours of autonomy compared to barely 2 hours for the much heavier lead-acid battery.

One trick that I'm kinda proud of: I powered my server directly from the 96V DC. And I periodically switch the current direction using a DPDT relay to avoid wearing out one side of the rectifier inside the PSU.

[exmadscientist]

> to avoid wearing out one side of the rectifier inside the PSU

If you are seriously worried about this then the whole thing is trash. Either the design is marginal or it is not. You cannot possibly switch a relay fast enough to make a difference here (and have the relay survive).

[cyberax]

I'm worried about long-term thermal wear from uneven heating, so I switch the current direction once a day. Simply because that's the maximum time I can set on a time delay relay that controls it.

[exmadscientist]

Don't bother. That's not how these things fail, and one day is so so far beyond the thermal time constants involved that it's not doing anything useful. You would have to switch on timescales of seconds to minutes to do anything meaningful, and that would kill your relay in short order.

If you are truly close to the design failure point of the rectifier, it's not safe to run at all. (You are almost certainly not.)

If you are worried about the fact that you're only using one element of a multi-element package, again, it's a nonissue. We do this all the time. It's often cheaper to add a second bridge rectifier to get a single diode than it is to add another BOM line item for the "proper" part. As long as that diode isn't operating near absolute maximum ratings (it probably isn't), it doesn't matter that there are or aren't three more in the box.

[butvacuum]

note: its the additional component and its knock on effects that are the cost- not the half cent diode.

aka- you add a diode, now you have to add procurement, warehousing, extra time on the pick and place, possibly a more expensive/larger+slower one as well(so even more time), then you have wastage, labor for keeping the machine fed...

[cyberax]

I removed it to see what happens :)

The reason I added it in the first place is because I've seen failures from uneven PCB heating. It's not the components themselves fail, but radiators get unglued and PCBs can get damaged.

[Dylan16807]

I don't understand what you're saying.

If you're specifically worried about wear, then you could switch once a month and it would be enough.

[exmadscientist]

Silicon rectifiers don't wear out.

(At least not on timescales relevant to individual humans.)

So hearing that makes me get suspicious that something else is going on.

[Dylan16807]

I'm also suspicious of the idea of that part wearing out, but if it doesn't matter at all then there's no reason to call things trash.

Your other comment says: If you are truly close to the design failure point of the rectifier, it's not safe to run at all. (You are almost certainly not.)

Well there's no reason to assume it's close to the failure point.

Think of it this way: Draw the line in the sand for where you'll approve the design, but just barely. If someone is running a diode close to that line, then it's not trash but trying to improve longevity isn't crazy either.

[exmadscientist]

The point is that the dance with the relay doesn't move the needle on design acceptance. If it is acceptable with the relay, it will be acceptable without the relay, because the component stresses will be the same.

So if it is unacceptable, it is unacceptable, and needs to be fixed. I said "trash" because it's going to become trash, and with luck just the power supply. Hope there's a fuse inline! Input rectifier failures tend to take down other stuff without one.

[scottlamb]

I toyed with this too, but I guess I have a slightly more diverse set of devices than you do. A few more weird voltages, and some things that expect mains. I looked into finding a DC version of their power supplies (e.g. the pico-box X9-ATX-500 to replace a conventional ATX PSU, tracking down DC versions of network switch hot-swappable PSUs from eBay) but decided it wasn't worth it. I just bought a stock LifePO4 power station. I found that I got most of the benefit [edit: measured in terms of runtime after power outage, not power draw while input power was available] just from switching to LifePO4 rather than from avoiding DC->AC->DC, and it was cheap and easy.

[nomel]

If you get your battery pack up to 48VDC, it opens up a whole world of low voltage power converters, since this is standard in telecom/PoE.

[bb88]

I've been in the process of cutting the DC side of power bricks and crimping anderson power pole (APP) connectors onto both sides of the wire. For camping and ham radio, it's really nice to hook up to a battery without the AC inverter taking DC -> AC and the power brick sending that to DC again.

The only thing to be careful with is connecting different voltages to different connectors, but it's at least possible with the APP connectors to "Build your own" with different color housings and different ways of combining the housings.

So maybe 13.8v is red/black and something that's 5V is black/white, etc.

[butvacuum]

theres also anderson's SB series- APC seems fond of knockoff versions.

[aspbee555]

I did something similar but made the batteries and solar priority, solar charges battery and wall power only used to top off as needed, otherwise always running on batteries

The Drok DC-DC did not work for my minipc that needed 19V/130W supply (would cut off with heavy draw), but the JacobsParts LTC3780 130W has been running my minipc's for almost a year now, gaming minipc, server minipc and networking

before that the solar panels barely charged the solix unit, but now my batteries fully charge and I still sometimes have left over solar I feed into the solix

[dreamcompiler]

This is the way to go IMHO: Keep the outputs DC. Put a bunch of USB-C PD ports on the thing and you're good with a lot of modern equipment.

[dylan604]

> Evidently, AC waveform generation is extremely power-hungry

Evidence is the heat from that conversion