You never have any guarantees with overclocking, but this is still interesting.

Hypothesis:
- even though it was stable at 33MHz, some component on the motherboard was running at the very edge of what it could do
- adding the 100Mb NIC pushed whatever part it was over that edge
- after removing the NIC it would work again at lower speeds, but with insufficient headroom to run at 33MHz (even though 30MHz does work)

Assuming this hypothesis to be correct, the obvious culprit would be a capacitor near the point of failure, most likely in the power delivery part of the board (as that would be shared between ISA bus and CPU), but in any event somewhere along the 5V line. Any suspects visible?

Good remark. Pls let me check that later

Might be CPU, might be the chipset. If either can't handle 66MHz clock you'll get a dead or broken system. Most of the 386SX mobos I've seen are meant to be cheap first and foremost, the design is very light on decoupling caps on power rails near the big chips. I've seen 386SX with pretty much two 100nF and one electrolytic farther away, and this was a 2-layer design IIRC. That's not a very good overclocking base even if the PSU is recapped with modern low-ESR parts...

> I was then changing network cards to test an intel 100mbit.
> when I removed the card again the system didn’t boot anymore.

so it was working with old NIC, stopped with it removed, and you havent tried with same configuration of cards again? as Deunan says bad filtering/decoupling is a given on 386SX/budged 386DX mobos, that old LAN card might have been the lynchpin keeping your config stable 😀

Yes indeed. Also tried with the old card setup up to with Graphic card only.

Ok I will check all these electrical units. I am absolutely not doubting that as I don’t have another explanation but also made the experience that a cap or whatever is broken OR is working. Never had something in between. I can also imagine that visually you won’t notice anything on them but I will give it of course a try.

Thanks all

PS. As stated it was fully stable at 66mhz osci. You quite fast run into issues with windows 95 in case sth isn’t working properly.

Is that your board https://theretroweb.com/motherboards/s/dataex … 1-386sx-ver-1.0 ?
Try exact combination of cards and drives again. You can also try soldering additional caps on the board. Sprinkle few ~220uF electrolytics/tantalums and 100nF on 5V near last ISA slot and near CPU (pins 8-9-10 and 11-12-13 is obvious spot).

Yes that’s exactly the board. One of the tantalum I already replaced.

Exact combo of cards and drives have already been tested without success. Oh ok I check the doc and will make a proposal for what you have just recommended

I checked the board and the picture of the page you mentioned. But I have no clue where I should or better where I could put in additional caps.
Would you mind to highlight the spots on the picture? I checked the pins but don’t know where there to add them. 🙁
Thanks a lot really upfront.

Ok I think you mean these for the cpu.

„ 220uF electrolytics/tantalums and 100nF on 5V near last ISA slot and near CPU“

Im not sure which capacities to put where exactly. Is this something you further go in detail ?

Fully stable doesn't neccessarily mean that it was more than 0.0001% away from unstable.
If it's related to capacitors, and as you have swapped and removed all cards (which do contain capacitors themselves), maybe it works better if you move this or that card closest to the cpu.

Here is CPU pinout.

Red is +, blue -. Ideally you would glue like three or seven 100nF ceramic capacitors on top of the CPU in a star pattern all connected to each other on one end. Wire that center to ground on the motherboard (any blue pin), and connect left over capacitor ends to red pins of cpu with thin wires distributing capacitors around ignoring 8-14 pins as there is already one cap there. something like this:

+ one ~100-680 uF electrolytic on top of all this for good measure between any blue/red :]

Without putting scope on this board I give 0 warranty that CPU voltage fluctuations are your main issue, but more decoupling never hurts.

I say thank you very much for your effort. I will plan this project !!!

Thanks again!!

PS: as I am not the full e-expert I just want to explain some thoughts I have:
All worked well with 66mhz and the current set of caps. Taking this perspective I would have assumed that I should replace (maybe extend) the existing caps and not to add some on places where none have been so far. But I am not the expert and am thankful for your hint to follow up on

Indeed it will be much easier to solder big electrolytic caps to already present caps next to ISA slots (blue ones). The only voltage rail you should bother with is +5V connected to Pin B3, B29 and D16. Since you mention it doesnt start with no cards at all I would try adding caps closest to chipset/cpu.
Easiest soldering would be on the back of the board directly to ISA slot pins between B1-B3 B31-29 D18-16.

Clear and understood. Thanks a lot for your effort. So you wouldn’t start with replacing the existing blue ones near power supply I understand.
Thanks and I will plan this.

I will also test with replacing cards more in detail - as proposed here in the thread.
Thanks also for that

Blue caps look like Tantalums, those either work fine or due to age and moisture ingress go short and blow up 😀 no middle ground.

Ahhh

Short update here: I could scope the issue further:

The whole error chain is depending on the run time of my pc (in hrs):

With 30mhz all is fine. But I experience that after some hrs the systems won’t boot after hard-resetting. Even power off and on won’t work and I then have to wait 30seconds. Afterwards it works again.

Trying now with 32mhz - yes I slowly decreased from 33 MHz - the systems runs well. When I power off the system after several hours of usage it won’t power on again. I have to wait then several hrs before it will power on again.

Isn’t that strange?! I definitely could reproduce these behaviours.

Seems like some components do need I don’t know a recovery time or something? But talking about minutes or hours (for 32mhz) it would be new to me that capacitors could need such cooldown / de-load time.

What do you think?

That’s what I found as a possible reason:

The error is due to the CMOS latch-up: after a long period of operation, residual charges remain in the CMOS circuits, preventing a complete reset. Only when these residual charges have dropped - often after hours - does the PC start again, which occurs more frequently at higher clock speeds.