Did you try 4.5v on the external battery header ?
Most Varta are 3.7v, CR2032 are 3.2v and with a diode drops to about 2.6-2.8v and too low to fully function the 5818 cmos (needs 3.0+ volts).
Check VDD pin 24 of the 5818 ic with power off (on battery only) and see if you have 3.0v, then check with power on and should be between 3v and 5v...

I did check it, and the power is indeed too low. With the CR2032 I'm getting 2.5V at the IC. When I connect a 3.6V battery to the external header I'm getting 2.8V at the IC, so still too low. Incidentally, when I measure pin 24 when the computer is switched on, I'm getting 4.3V at pin 24 (I'm getting a good 5V elsewhere on the board), so somewhere along the line I'm losing power.
Could I connect the positive from the external header directly to pin 24, or is that going to blow up something?

I found where I'm losing power: It's right after the first diode.
I took the diodes from a spare parts 386 board and put them in place on this mainboard, but it's the same result.
I'm getting 3.08V before the diode, 2.9V when lifting the other side out of the circuit. It's going down to 2.7V with the diode full inserted in the circuit. I'm getting 2.7V at pin 24.

naujoks wrote on 2024-04-22, 13:16:

With the CR2032 I'm getting 2.5V at the IC. When I connect a 3.6V battery to the external header I'm getting 2.8V at the IC,
PSU getting 4.3V at pin 24 (I'm getting a good 5V elsewhere on the board), so somewhere along the line I'm losing power.
Could I connect the positive from the external header directly to pin 24, or is that going to blow up something?

Ok lets go down the list.
The orig Varta is 3.6v, the cr2032 is ~3v. So with the Varta rtc would be getting 3.1v (+0.6v more) which would work.
The external header was designed for 4.5v ext battery and has two diodes in series giving 3.5v to pin 24
No you cannot connect direct to pin 24 !!!
Not all old boards are capable of using a cr2032 as the battery, most 286 and 386 cannot due to the rtc chips used and use a 3-AA battery pack for the external or a true replacement 3.6/3.7v varta

I have a 386 board with same M5818 and traced/voltaged it out.
From EXT battery header are two diodes in series dropping the 4.5v to 3.5v then to pin 24.
From psu +5v line the is one diode in series (near 5818) which drops the +5 to 4.3 to pin 24 just like your measurement. (needed to prevent battery touching main +5v which goes everywhere 😀)
Off the 3.6v + Varta is one diode to pin 24 (plus the charge circuit or transistors/resistors) for a 3.1v reading.

My 386 board is this: https://theretroweb.com/motherboards/s/ecs-8517-v3.3 same ALI M5818 RTC ic

I connected a 4.5V battery pack to the external header. With it, I'm measuring 3.5V at pin 24. It's not giving me any different behaviour though.
What's really strange is that when switching the computer on the first time, the system hangs at POST code C0 (which, with AMIBIOS, apparently is something RTC related). I then have to disconnect the VGA cable from the graphics card and reconnect it again and then start the computer again for the system to boot normally. Like, what the heck?

Post high res pics of your board, it'll help.

I do not see a post code for C0 for any 1990-1994 AMI bios in my archives. Where did you get that info ? Generally a rtc failure does not halt a boot but KB controller error <0C code in AMI> can (and is near the old battery)
The video glitch is probably from the ISA vid card and what ever monitor you are using not fully being compatible (had that with a Nokia 19"crt and a GF4 years ago, was the vid card bios)
yes pictures would help

Here's a couple of pictures. One of them is of the place for the RTC before putting the socket in.

Have you removed the socket of the keyboard controller and the bottom 2-4 RAM sockets? In most cases there are also interrupted traces to be found.
The RTC not only needs battery voltage, but also has a reset circuitry and more to make it work properly.

As it says DS1287 on the silk screen, just for fun I actually put in a DS1287, and what do you know, it's working! Incidentally, a DS12887 does not work.
So that solution works for me, although I'd still be interested in finding out why the original IC doesn't work. I'm assuming that, since the DS1287 works, and I was getting alright voltages with the battery, damaged traces can be excluded as a fault? Maybe the 5818 is kaput after all?

The DS1287 has its own internal battery (and crystal) so the CR2032 won't be doing anything.

The symptom of the RTC losing all its settings except for the calender (date/time) would indicate the rclr pin (21) isn't being held high when the RTC is powered.

So I should be able to measure the battery voltage on pin 21 when the computer is off?

One thing I just noted is that the DS1287 only works when I have the external 4.5V battery connected. It doesn't work without it and also not with the CR2032 in place of the original 3.6V Varta.
I wonder if that tells us something?

naujoks wrote on 2024-04-26, 06:08:

One thing I just noted is that the DS1287 only works when I have the external 4.5V battery connected. It doesn't work without it and also not with the CR2032 in place of the original 3.6V Varta.
I wonder if that tells us something?

Probably a voltage drop too steep with only a starting +V of 3V?

What is the external battery's power needed for in this case, since the DS1287 has its own battery?

Maybe the battery in 1287 is near dead. Try jumper on the middle two external battery pins.