I had a Gigabyte GA-5AA the caps looked fine but were bad anyway. After replacing the five caps next to the CPU socket and the two between the socket and the battery the board started working again. Try underclocking you cpu to reduce the stress on the power delivery circuit. If that improves stability then you should replace the capacitors.

I can give this a shot before recapping, sure. Some time ago I recapped a MSI K8T Neo2 that had a couple of blown caps next to the CPU and was working completely fine. I guess you can never be sure with that old stuff.

Edit: In case I'm going to replace the caps, any hints for what type is needed? Can I go low esr for all of them? Unfortunately, the caps don't have any type number as far as I can see, and I could not find any datasheets.

Do you have a POST card you could use to see where it stops (if it even shows life) when it doesn't POST? I doubt it's your problem, but the kernel32 error would make me run a few passes of memtest86 to rule out bad RAM.

Unfortunately, I don't have a POST card. I will try memtest86 and maybe test with different RAM.

Note on that kernel32.dll error: I had a similar one when closing Internet Explorer recently. I did some tests with 7-zips integrated benchmark and no issues there. I'd like to run prime95 too. Any idea what is the last version that runs on Win98?

technokater wrote on 2023-11-02, 12:24:

I can give this a shot before recapping, sure.

After.
Repair practice and common sense say that you first need to get rid of problem areas.
Only after then you must out diagnostics.
Most often the problem goes away.

Well, I would need to order (correct) type of caps first and then hope I don't mess up the motherboard while trying to get the old caps out. That's why I'm a bit reluctant to rush the recapping. But yeah, considering that this hardware is decades old it is a potential source of problems... Any tips for finding replacement caps? I googled both brands, GCS and Stone, and can't really find much about it. On my last recapping I did find the datasheets for the caps, so it was easy to find replacements.

technokater wrote on 2023-11-02, 14:50:

source of problems... Any tips for finding replacement caps?

If I lived in Germany, this would be a very big problem.....
https://www.we-online.com/de/components/products/WCAP-PTG5

Thanks for the link! I can just use polymer ones? That's good I guess 😀

Now there will be many opinions that no.
The rest simply change the capacitors.

I see. I didn't use polymers yet, even though I have read about it. So, the original ones are GSC 1000 uF 10 V 105 C and Stone 330 uF 25 V 105 C. I need to stick to the voltage rating of course, at least not go below, but I didn't see the capacity available with the Würth polymers. For electrolytics I can go higher, but not lower. How is it with polymers?

I didn’t know that I couldn’t use polymer ones, so my Aopen S370 now looks like this. 🙁
But the S754 is still nicer under Windows 2000, so the P3-S Tualatin went on the shelf.

technokater wrote on 2023-11-02, 15:20:

I see. I didn't use polymers yet, even though I have read about it. So, the original ones are GSC 1000 uF 10 V 105 C and Stone 330 uF 25 V 105 C. I need to stick to the voltage rating of course, at least not go below, but I didn't see the capacity available with the Würth polymers. For electrolytics I can go higher, but not lower. How is it with polymers?

On such boards everything is usually powered from +5V.
10V capacitors are used to improve ESR. Typically, the higher the voltage, the lower the ESR.
For 8mm - 820uF*6.3V is suitable as a universal capacitor.
For 10 mm - 1000*6.3V.

Nice, thanks a lot! I'll do some RAM testing and then decide what to do about the caps. It is really weird for me that everything seems fine once the PC has actually booted.

This is a combo AT/ATX motherboard, on which the 3.3V bus is made locally from +5V.
For proper repairs, you need to change the capacitors after the CPU converter and before it.
And also after the 5->3.3V converter(or linear regulator) and before it.

Good catch. I'd probably change all of them anyway, I did that in my previous recappings too. Just to make sure.

Well, a short update: I thought I saw some corrosion on the external cache chip, so I did a quick cleanup and reflow with fresh solder. No change at all. But then I reset the BIOS and forgot to turn on the ECC setting for the SDRAM. What shall I say, it boots every time since then. I did a little research and it seems like the Aladdin V chipset does not support ECC beyond 83 MHz! I'm running 100 MHz, doh 🙁 I'm still not confident in all the caps but I will observe the situation. Thanks for the help though.

Edit: This didn't fix the KERNEL32.DLL crash when trying to install the ALi AGP driver though. But that might be unrelated after all anyway. I found a prime95 version that still runs on Windows 98 and on blend profile it basically crashes instantly. So yeah, retro computers are fun 😁 definitely something wrong with the memory (I swapped RAM sticks already)

Edit 2: So I tested this prime95 version on another system which didn't show any issues before. Turns out that the Large FFT and blend tests crash with the same error message on that system. I switched to an older version of prime95 (v25.11) and this is running fine since a couple of hours. The error message is about a stack fault in KERNEL32.DLL. I suspect that the newer prime95 version is either compiled using an incompatible runtime (i.e. something doesn't match with what I have installed on my Win98 SE) or it is some hardware incompatibilty (unsupported CPU instructions maybe).

Edit 3: I just ran prime95 on blend for 5 hours without any issues, memtest86 is on its third pass right now, no issues either. But Windows 2000 Setup crashes with STOP 0x7F UNEXPECTED_KERNEL_MODE_TRAP (divide by zero if I look at the exception params) right after booting from the CD. I think this might be related to the kernel32.dll errors that I've been seeing on Windows 98 SE with the ULi AGP driver setup. I tried reducing the memory timings and fiddled with some other BIOS settings, but no luck. I'm slowly running out of ideas. I will test with only 1 stick of memory afterwards, see if that helps. Even though I find it hard to believe that the system is running prime95, memtest86, and 3DMark 99 rock stable on Windows 98 otherwise. Unfortunately, I do not have a second K6-2 CPU to test if something is wrong with my CPU. Might also try to use a real HDD instead of the CF card.

Edit 4: Yep, somehow the CF card or the IDE adapter causes the crash when trying to install Windows 2000. I plugged in a 40 GB IDE HDD (which is ATA133, so much newer than the system) and what shall I say, no crash and the setup is running right now. I have the feeling that the other exceptions that I've been seeing might be gone as well, but I need to install Windows 98 on the HDD first to confirm. I didn't suspect the CF card/adapter because it really seemed to be working. Unfortunately I only have CF cards of the same type/brand, so I cannot check right now if changing the CF card will help the situation.