Strange. Perhaps you could point a fan at areas of the board (e.g. near the CPU) while it's in its "slow" state, to see if it might speed up?

Some boards require turbo header to be explicitly set for normal operation. Untied inputs can and will randomly toggle. The scan does show a jumper in place though...

That's a good point - however I forgot to mention I've tried with or without the jumper, and also with a turbo switch header from a case. Turbo makes a difference in both cases - when board is working normally the turbo switch make the board go from 100% speed to 50% speed (figuratively speaking) and when the boards gets in that slow defective state, the switch makes the board go from 10% normal speed to 5%. I'll see if I can investigate further or I might just give up on the board for now.

Based on the picture you've posted, there's no backup battery.
This might cause the board to lose some of its CMOS settings which, in turn, might cause it to revert to default settings.
Try to fit an external battery.

Thanks - I've already tried it with an external battery but it doesn't seem to make any difference so far

MMaximus wrote:

Thanks - I've already tried it with an external battery but it doesn't seem to make any difference so far

Is possible then that one (or more) of the chips or chipsets has a problem. Once it heats up, then it causes the PC to slow down. Try changing out your memory first.
EDIT: Does any of the chips(ets) get excessively hot (too hot to touch) once the PC starts to slow down?

A temperature related change seems plausible. Are you running the board bare with no fans or heatsinks? The system probably isn't sophisticated enough to slow itself down, so maybe the clock is actually slowing down (does heat make the crystal oscillator's output frequency decrease?). Maybe just stick a heatsink on your 386 and see if it helps.

Do unplug the board between tests? Do you unplug it at night? Perhap it's capacitor related?

Have you run speedsys at full speed and when it slows down? I'm curious what the slowdown value is, if it's a consistent value and if that value can be tied to anything? I have a hard time believing a 386 can generate enough heat to cause something like this due to lack of cooling. You've already tried reducing variables, but have you tried different RAM? Or different amounts of RAM?

Also check the board for cleanliness. Any oils, etc over traces could maybe cause electrical bleed once things heat up?

Get a program that reliably tells you the CPU frequency (I think chkcpu can do that, but I'm sure there are others as well). What values does it report when the system is in "slowdown mode"?

Lots of good suggestions - thanks guys. 😀

I don't really understand what's happening. I've turned on the system again today and now the RAM count is very slow even on a cold start. Maybe this is not heat related after all but more like a random fault? I've looked around but it doesn't seem I have any 30-pin compatible memory (I've got some other 30-pin sticks but they don't have 9 chips on each like those from the board, and they're also slower-rated so I was reluctant to try them). The board refuses to post if I use only 2 sticks, so I'm guessing a whole bank needs to be populated with 4 sticks for the board to function.

After a long wait to access the BIOS setup I managed to adjust some settings like "no HDD" and the floppy drive type. The board remembers the settings, the RAM count is still very slow (however it's possible to skip it by pressing ESC) BUT... the system boots to DOS on a floppy. From there I've managed to launch Dr. HARD and do a quick benchmark and... it looks like the performance is normal 😕

I started a game from a floppy (Test Drive 3) and it plays ok. I just don't understand why the RAM count was much faster a few days ago... I guess I need to find other 30-pin sticks to try. 😒

Here is the Dr. HARD benchmark: 4231 "Hardstones" and 132 "Softstones". I'm not sure if that score is in the acceptable range for a 386DX25 - for comparison my 286-20 gives 3069 "Hardstones" and 99 "Softstones".

That benchmark sounds fine to me because if I remember correctly 286 is bit faster at same clock speed compared to 386dx.

After some more testing:
I managed to source some RAMs and swapped all the sticks, it doesn't make a difference with the slow ram count. OTOH I've noticed that when the turbo header is not jumpered, the system won't read floppies. So as Tiido was correctly pointing it, I guess the board does need to be jumpered for normal operation.

I've run a few more benchmarks:

NSSI: 5882 Dhrystones

Checkit 3.0: 5262 Dhrystones

The checkit performance seems right. My 386 DX-40 scores about 32 times the IBM PC-XT, so a 25mhz version without cache seems about right at 15-16 times the XT.

Does your CMOS setup have an option for something like "quick memory test" or "disable entire memory testing"?