Uninterruptible Power Supply (UPS) for Glowforge


Does anyone use an uninterruptible power supply (UPS) for their glowforge?
If so what is the specifications that I would need for my glowforge?


Some do.

The max power rating is 800 Watts for the Glowforge.

If you are in areas prone to brown/black outs, it would be a good idea.

Make sure to put a battery back up on the wireless router and your ISP’s equipment.


Ok, I see a Cyber Power UPS 1350 VA, 810 Watts
half load - 9 min
full load - 2 min
surge rating - 1500 Joules

would this work for the Glowforge?
if you have experience let me know what you think.


That would be fine, but you’ll only have 2 minutes of runtime.

It would work for taking the brunt of a temporary brownout or protecting from a lightning strike, but you shouldn’t expect it to allow the completion of the majority of print jobs that are in progress.


Yes, I have a 1500VA APC on mine (and a similar one on my 3D printer)


Go with an APC 1500G.

Not only is it a little longer runtime at full load, but the APC 1500G has the ability to add an additional “side car” that will triple the power rating (for those areas where power is really inconsistent or stable).

Amazon will offer you 30 days of Prime (read free 2 day shipping) as a trial. That is the best place to get a 1500G and the the “side car”.


I wonder if the GF processor board has the ability to store where it is up to when power goes. As it is a Linux system they must has some way of handling power off without corrupting the file system so maybe they have a power fail signal from the PSU.


That would be a read-only filesystem, and relying on RAM for anything else. Could they have non-volatile flash? Sure… But that wears out. Just look at engineers on the Curiosity Opportunity rover and what they’ve had to do to keep things working over the years.


Actually, I made a mistake… I always confuse them… I didn’t mean the Curiosity rover, I meant the Opportunity rover.

The Opportunity rover as a whole is testimony of “how many hacks can we squeeze out of a dying machine to keep it running”. It was only slated to last for 90 days, and it’s rolled over 5000+ days now. One of the key hacks they’ve had to do is sidestep corrupted memory. (https://www.theverge.com/2014/12/30/7470743/nasa-mars-opportunity-memory-loss)

It just recently celebrated the 5000-day mark with updates from troubles over the last three years.


Well something more down to earth would be the Raspberry Pi and corrupted SD memory cards. In a nutshell, turning off a RPi without properly executing an orderly OS shutdown will eventually corrupt the system bad enough that the only recovery is to completely wipe the SD card and reinstall.

It’s not something new to RPis, it’s something that has happened to embedded systems I’ve worked on for years. You learn to favor offloading temporary data as quickly as possible in order to prevent data loss.

So in addressing the original question:

The answer (in my mind) would be: send regular progress updates to the cloud and resume from the last reported location.


And that cloud may be on Mars some day.


If I was implementing the system I use a power fail signal from the PSU to write the buffer position to EEPROM or battery backed SRAM before shutting down. That way you can resume in exactly the same place seamlessly. To send it to the cloud needs an order of magnitude more time hold up time. Regular updates would always be out of date so you might engrave a few pixels twice giving a blemish.


It’s going to be a tradeoff of what you’re willing to lose versus wear and tear and servicability, for sure.

With an EEPROM or SRAM you still have a part that’s going to wear out over time to the point where it either needs replacement or becomes inoperable due to limited write cycles. A customer isn’t going to replace socketed/soldered flash memory themselves, that’s likely a complete board replacement.

A replaceable backup battery is a good idea because it’s more user-friendly to repair, but I’d use the battery to run long enough to dump the remaining queued network packet buffers and avoid the need for a non-user serviceable part altogether. People can easily replace batteries.

Yes, it would take longer… but running a CPU stack with a rechargeable battery for the “order of magnitude” longer isn’t a burden, because it will be completed long before the rechargeable battery is drained. Unless you’re going to use button batteries.

Edit: For some reason (probably the hour) I was thinking of nvSRAM not battery backed up SRAM. Anyways. Interesting thought experiment on a theoretical use case that highly likely doesn’t apply to our Glowforges. :stuck_out_tongue:


SRAM doesn’t wear out and you can power it with a super capacitor for long enough to span a power outage.

EEPROM would last at least 10,000 writes. If you see 10,000 power cuts while actually lasering you need to get a better power source!


The new Prusa mk3 uses the remaining power supply to write the info to static ram on power loss. The Einsy motherboard can be powered for 500ms apparently from the power supply caps.