diy solar

diy solar

Can EG4 18KPV work as UPS for sensitive loads like computer servers?

Also, per UL1741 SA, the inverter HAS TO stay on-grid between 88-120% (or 105.6-144v for 120v). A UPS can have a tighter range.

Maybe there is a way to set override the inverter grid profile to have less aggressive ride through. Ride through is anyway only needed for grid support inverters. The application in question is decidedly not supporting grid

How long does it take to disconnect the relay is the question. During that time, the grid will sink the power, and for all practical purposes, 0v will be received by your load.

If the grid is on its way out, doesn't it drop below 88% and therefore eligible to disconnect?

I don't think it goes down to zero in a step function either. Depends a lot on the ambient conditions. For instance, if a line goes down during the day on a branch with a lot of grid tie inverters those inverters might slow down the voltage drop. At night in most situations if distribution line goes down it will likely plummet pretty fast unless there are enough folks on the branch running storage inverters that are prewarmed to catch the drop quickly.
 
Last edited:
Sol-Ark markets their 15K as "5ms UPS no-glitch transfer". EG4 says their 18KPV is a UPS with "10ms transfer". Is the hardware all that much different? Can the Sol-Ark keep servers running during the grid-to-battery transfer while EG4 can't? Anyone know this for sure?


Have you tried contacting SolArk, EG4 support to do a test? Not sure if they will do such testing for a customer since it's not like an enterprise equipment or software level of profit under discussion. But always ok to ask.

Ideally the designer or manufacturer should already know performance parameters like speed of the transfer relay, sequence diagram of the events during the different transfer types, and what the total latency is for different modes under ideal conditions.

Alternatively you could reach out to folks directly via DM or tagging to do the testing. Forum members probably have more time than influencers.

I agree with above post that the most likely info you can get for controlled testing is for a clean transfer from flipping the disconnect with no extra load. I guess you can put a test load between the disconnect and transfer switch (8 hair dryers?) to make it more realistic. Ideally someone would be able to scope the waveform during the transfer instead of just using logs and qualitative descriptions of what happened to the server.

I think it's very unlikely to find someone that has a fully populated server rack they can dedicate to this testing. Unless there's someone on this forum really into hoarding homelab hardware and has the hardware you are interested in benchmarks on.

Your 12kW server rack is also a large % of the inverter capacity. You may need to stack inverters if there turns out to be a surge load as the grid goes down, as diyrich noted above, since you have little surge capacity with if your load uses up that much of a HF inverter. Also from this thread it seems like this class of UPS load and uptime requirement is an outlier so pushing the power performance spec at the same time as the transfer latency spec (in all situations) is probably a risky proposition.
 
Unfortunately I think it is only the OP that could test a Solark instead with his particular server rack because they are all going to be different.
What I can tell you is this, I regularly switch grid on and off to my EG4 18k and the soft start on my AC compressor will cut out supply to the compressor every time on switch over but the fan keeps running. However none of my clocks reset. So it is quick but not that quick.
 
Unfortunately I think it is only the OP that could test a Solark instead with his particular server rack because they are all going to be different.
What I can tell you is this, I regularly switch grid on and off to my EG4 18k and the soft start on my AC compressor will cut out supply to the compressor every time on switch over but the fan keeps running. However none of my clocks reset. So it is quick but not that quick.
Does the compressor cut off going the other way as well - battery to grid?
 
Get a 12v charge controller with 12 lifep04 battery.
Use something like this to power the server directly.
24pin DC ATX PSU 12V DC Input DC 12V 150 watt Pico PSU ATX High Mini-ITX Power Supply with Adapter Board ATX Switch PSU Car Auto for Computer Mainboard and car/Motorcycle https://a.co/d/fHD6ufZ

Very interesting idea , can a PC/server handle the swing in voltage ? Especially from a car alternators

Or do you need some kind of regulated power supply?
 
Very interesting idea , can a PC/server handle the swing in voltage ? Especially from a car alternators

Or do you need some kind of regulated power supply?

Yeah, a lot of the enterprise rack-mount servers use very high demanding and specific-application PSUs, quick-change cartridge-based (may have 2 or more PSUs per chassis). Many of our servers in my lab at work use 1100w PSU x2, and some of the hyperconverged 4-node servers have 2200w PSUs x2, so your average desktop or PICO / ATX power supply will not be an option there.

Good news however, many of the enterprise servers on the market will also have a 48v DC input power supply option available, since there are a handful of datacenters where they prefer to run servers / racks on a DC power bus.

Example:
 
Last edited:
Simple answer is no. The unit might have the transfer time within acceptable levels, but the control logic is not designed to accommodate failures gracefully and with load priority. You could add an external static bypass switch or chain two units together in an isolated/redundant configuration, but a single unit will never be as reliable as a UPS.
 
Does the mini PSU you link regulate the +12v or only the other derived voltages?
The link is for 12v (unregulated 11.4-12.6v input) to ATX PSU, which is regulated 12v (16a). -12v (.1a), 5v (6a), and 3.3v (6a).
Note: I have not used it, so I have no idea the quality.
 
That's a very limited input range


I've used a 500w version of an ATX 12v input PSU (9-18v input spec) from Powerstream on a home server project before, I still have it in a box in my shop somewhere (although I don't use desktops anymore so I decomm'd it and boxed it away in case I ever needed it again)...


It worked fine for years running it direct on a 12v battery bank I used for my home core equipment. I also used a Powerstream 12v DC-DC 6a power supply to stabilize a solid 12v for my modem and home router.

The ATX PSU said it could run on wide input range so I ran it direct off battery bank, but the router and modem I wasn't sure, so I put the 12v DC-DC stabilizer (10-16v input, 12v output) in there to make sure they got the steady 12v all the time, for when the battery bank voltage fluctuates..

As long as the specs sheet of the PSU's input range shows it can go along with what your battery bank will see in real life, then my experience is it will work fine.
 
Last edited:
I need a high capacity UPS in the 10-12kW range to keep a rack of servers up while the backup generator starts. I recently purchased an EG4 18KPV hybrid inverter and four EG4 LL server rack batteries. This is grid connected with no solar and no grid back-feed. I am seeing a "glitch" on the critical loads when the mains go off. During the switch-over I can hear the pitch of the server cooling fans change and some computers reboot. Very disappointing.

EG4 uses the term "UPS" in their marketing literature / spec sheets and it has a so-called high speed relay built in gird-to-battery switching. The EG4 specs for the 18KPV say 10ms transfer time which is what most consumer grade 1000-2000 VA UPS are rated for so I fully expected this to work. There is a poorly documented setting on the 18KPV called "seamless switching" which sounds like what I want. I have this checked and yet the glitch remains. Current load is only about 4-5 kW so not even fully loaded, but not nothing either. I may try catching the glitch on an oscilloscope but getting the right capture trigger to see the actual "event" might be tricky.

Interestingly when the critical load is being partly powered by battery, meaning the inverter is already "on", there is no glitch when the mains go off. This "glitch" in output only occurs when load is entirely supplied by the mains and the mains fails causing the relay to switch. So maybe something slow with the inverter turn-on and/or sync? But why do they spec 10ms switching time then?

If found this thread https://diysolarforum.com/threads/online-ups-functionality.53936/ which appears to be focused on using an all-in-one inverter to provide electrical noise isolation like a double conversion UPS. I don't need noise isolation and would rather eliminate double conversion to avoid the resulting efficiency losses. What I need is a simple stand-by battery back-up UPS in the 12kW range that provides 240VAC. Feels like the 18KPV would be perfect. What I am missing?

Is anyone powering sensitive loads like computers directly with an 18KPV? 120 or 240. Without having a purpose made back-up UPS between the inverter and the computers.


Last night I got some exposure with the 18kpv at a neighbor's house who was having some trouble with their settings, so I had a chance to play with the menus in the display and whatnot. I needed to look up in the manual to check on some setting descriptions, and I stumbled across this setting function called 'Power Backup'...

In the display UI it would be called 'Off-grid Output'.. What is yours set to here?

1699775230129.png

Seems like enabling Off-Grid Output could be a function worth trying out. I haven't tried running this inverter with any of the AC inputs on it yet so I don't know all of the behaviors of it yet, but was just wondering if it may act different than Seamless EPS Switching. May not hurt to try a test.
 
Last edited:
Back
Top