CR2025/CR2032 battery dummy for power consumption tests and a hungry Supermicro board (#P21)
This bugged me for quite some time now: Ever since a very nice dumpster revealed a stripped Supermicro server chassis including a Supermicro X8DTL board (Dual-1366) to me, the BIOS battery of that system seemed to be empty all the time. You see, I use this machine as an offline backup, so once every couple of months I fire it up, transfer my data, and unplug it from power and ethernet. And almost always it doesn’t start right up because the CMOS data is gone. Ideally it should be a headless unit, so that’s highly inconvenient.
I finally got around to test this – the fix will be another project for some day in the future.
As dummy adapters for the CR2032 lithium coin cells commonly used for the RTC backup are nowhere to be found, I made my own. Those cells have 20mm diameter and 3.2mm height, which is no standard PCB thickness. Many manufacturers max out at 2.0mm, some do 2.4mm, some a bit more, especially with multi-layer PCBs. It went for the cheapish route: Two boards that will be glued together.
Here’s what I made and ordered on OSH Park, which seems to be the cheapest option for sub-1mm ENIG PCBs by far. I ordered one standard 1.6mm (now also available with black PCB material which looks fantastic!), and the other one in 0.8mm, both at no extra cost compared to their standard $5/in² in packs of three.
1.6mm plus 0.8mm offers the option to also make a CR2025 dummy when squeezed tight, I added some hot snot for about 2.8mm thickness. That fits fine in my CR2032 holders and might even do for slide-in CR2025 holders, but as I have two more boards each, I would be able to make a slimmer version if needed.
KiCad and Gerber files are available here, but you can also reorder the exact same thing via the OSH Park link above. It’s a shared project and open hardware in general.
The boards have one of the legs plated on both sides to attach a wire to them, and enough routed out space to make room for both leads to escape the holder. This design should be viable for both the top-insertion holders as shown below twice, as well as holders that the battery can slide in from the side.
This is the test unit that went into two boards:
First, the Supermicro X8DTL server board – please excuse the wiring madness of 2 SATA and 16 SAS devices
I wasn’t so sure if adding a amp meter would skew the results, but the HP 6644A offers remote sensing (especially easy with my custom front connectors), so I just put the voltage reading line after the multimeter, set it to 3.15V and enabled output – done.
I’m glad they have identical readings as I feared my Aneng had blown in the µA range before, but apparently it didn’t. 80µA steady consumption, okay…what does that mean?
Well, a typical CR2032 nominal capacity is in the order of 200mAh to 240mAh. Let’s assume the former as we do not know at which voltage the RTC and CMOS content will become unstable, and divide this by our 80µA: That’s 2500 hours of runtime, or about 3.5 months. Now that matches nicely with my previous observation of an empty battery each time I run that machine a few months apart. Dang!
Usually those coin cells last for years, so 80µA standby current can’t be normal. I didn’t like the option of losing BIOS data from another machine, so I pulled out an old Soyo board from storage, still hosting a Pentium 200 MMX. What do you know…
1µA consumption! Okay, this one wanders around quite a bit, it’s usually in that range, but there are occasional bursts up to 8µA and some downtime at 0.8µA (not sure about the accuracy of these readings!). But it shouldn’t be far off some 1µA-ish average current. 1/80th of the Supermicro, or about 20 years of battery life, meaning the self discharge of the cell likely will drain it faster than those pesky nano-amps for keeping the RTC going.
Well, now I know. I’ll be making a rechargeable power supply to feed that hungry monster by some means. I bet Supermicro never tested standby consumption, as having a server off the grid for months likely isn’t one of their design goals. As so often with my stuff, unusual usage requires unusual solutions for unusual problems..
[…] I also asked if anyone wanted one of my old server boards, Supermicro X7DBE and X8DTL-iF (as recently diagnosed with high battery drain!). I was going to keep one, and random.org decided it was the much less […]
[…] for fully depleting identical batteries within four months on their X8DTL server board, see #P21 for […]