BitWrangler wrote on 2024-03-06, 16:38:

Haven't got 2x256kB 30 pin around have you? might help bracket what lines to look at if they work in bank 0 and 1.

No, but I do have two 30 pin 512 kb modules:

Would that work?

EDIT: Ah. Checked stason.org. It won't work.

boggit wrote on 2024-03-06, 16:51:

No, but I do have two 30 pin 512 kb modules: https://i.imgur.com/1G18ctO.jpg […]
Show full quote

BitWrangler wrote on 2024-03-06, 16:38:

Haven't got 2x256kB 30 pin around have you? might help bracket what lines to look at if they work in bank 0 and 1.

No, but I do have two 30 pin 512 kb modules:

Would that work?

Actually, [this] seems like a better bet for motherboard ID. This one, incidentally, lists bank 0 as the slots originally occupied by the SIMMs.

boggit wrote on 2024-03-06, 16:51:

No, but I do have two 30 pin 512 kb modules: https://i.imgur.com/1G18ctO.jpg […]
Show full quote

BitWrangler wrote on 2024-03-06, 16:38:

Haven't got 2x256kB 30 pin around have you? might help bracket what lines to look at if they work in bank 0 and 1.

No, but I do have two 30 pin 512 kb modules:

Would that work?

I think it should, that looks like a 1MB chip - HY514400A is 1,048,576 x 4-bit per chip. So with a pair you should have 2MB of RAM 😀

Thermalwrong wrote on 2024-03-06, 17:06:

I think it should, that looks like a 1MB chip - HY514400A is 1,048,576 x 4-bit per chip. So with a pair you should have 2MB of RAM 😀

Yes, you are right! Brain fart.

However, I cannot find a legitimate RAM configuration with 2x4mb + 2x1mb. Or do you suggest that I try using only the 1mb ones?

Well yah try them alone either way as then we rule out flaky modules maybe.

Though I am not sure the second board given is any better match, since there should be more of those jumpers showing in your first pic. Wider photo of the board would help others have a go at IDing it correctly. Hard to do it off those microhouse diagrams though, they are more like stylised subway system maps than actual maps.

BitWrangler wrote on 2024-03-06, 17:18:

Well yah try them alone either way as then we rule out flaky modules maybe.

Though I am not sure the second board given is any better match, since there should be more of those jumpers showing in your first pic. Wider photo of the board would help others have a go at IDing it correctly. Hard to do it off those microhouse diagrams though, they are more like stylised subway system maps than actual maps.

Never hurts to run Memtest+ 4.1 overnight. These things can be caused by edge case memory issues.

BitWrangler wrote on 2024-03-06, 17:18:

Well yah try them alone either way as then we rule out flaky modules maybe.

Fair enough. It's a pain to unslot and slot in these modules with the board screwed into the case though, but it's an even bigger pain to remove the board from the case.

douglar wrote on 2024-03-06, 17:20:

BitWrangler wrote on 2024-03-06, 17:18:

Well yah try them alone either way as then we rule out flaky modules maybe.

Though I am not sure the second board given is any better match, since there should be more of those jumpers showing in your first pic. Wider photo of the board would help others have a go at IDing it correctly. Hard to do it off those microhouse diagrams though, they are more like stylised subway system maps than actual maps.

Never hurts to run Memtest+ 4.1 overnight. These things can be caused by edge case memory issues.

Not a bad idea, though seeing test above 1MB disabled in setup makes me wonder if the 4MBs pass the BIOS memtest even. Starting wondering whether this is a gussied up 286 design and doesn't like 4MBs

BitWrangler wrote on 2024-03-06, 17:29:

Not a bad idea, though seeing test above 1MB disabled in setup makes me wonder if the 4MBs pass the BIOS memtest even. Starting wondering whether this is a gussied up 286 design and doesn't like 4MBs

I enabled test above 1MB as well as memory parity check. The RAM cards passed that without issues.

There's a whole whack of boards look identical vis a vis the BIOS, crystal, resistor and cap layout in front of the SIMM sockets, but none of them have the 244 buffer chips installed...

https://theretroweb.com/motherboards/s/dfi-386sx-25-33-40ane
https://theretroweb.com/motherboards/s/kaimei-kmx-al-8
https://theretroweb.com/motherboards/s/kaimei-kmx-al-8517
https://theretroweb.com/motherboards/s/lucky- … -386sxa-ver-4-0
https://theretroweb.com/motherboards/s/unknown-612-3035-001

edit: ah these two do... (have buffers)
https://theretroweb.com/motherboards/s/auva-c … mputer-npm16-a0
https://theretroweb.com/motherboards/s/mg-8517-rev-2-2

But yeah, looks like all members of the AlI M1217 6 slot anonymous club have bank 0 labelled towards inside.

EditII: That ADS delay J5 3 pin is a bit of a mystery setting, not least because it's labelled J rather than JP leading some people to connect front panel stuff to it and causing weird problems. Should be set 1-2 enabled or 2-3 disabled with jumper cap. Disabled seems to be default for Intel CPU maybe enabled does something for Cyrix compatibility, unclear.

Gate A20 is controlled by the keyboard controller, right? I would check that over the ram.

kingcake wrote on 2024-03-06, 20:19:

Gate A20 is controlled by the keyboard controller, right? I would check that over the ram.

Working A20 is all tested during POST, I think, so it wouldn't seem like a hardware issue to me

jakethompson1 wrote on 2024-03-06, 21:28:

kingcake wrote on 2024-03-06, 20:19:

Gate A20 is controlled by the keyboard controller, right? I would check that over the ram.

Working A20 is all tested during POST, I think, so it wouldn't seem like a hardware issue to me

I don't know if you can count on that without disassembling the bios. I just don't see how bad ram would prevent an A20 handler from working.

kingcake wrote on 2024-03-06, 21:45:

I don't know if you can count on that without disassembling the bios. I just don't see how bad ram would prevent an A20 handler from working.

Bad RAM can not prevent the A20 maskinig circuit from working - but bad RAM surely can just ignore the address bit that is mapped to A20. If the RAM ignores A20, it looks to the processor as if the A20 gate is permanently closed. If the bad RAM does care about A21, memory sizing might ignore the missing A20 bit, as memory sizing code does not expect A20 to be ignored if A21 is respected, so it might not check whether A20 is respected.

I found [this] article through googling, written by someone also using an AM386SX processor, who also experienced the same A20 error. This seems to strengthen the case for faulty RAM being the villain of this drama.

BitWrangler wrote on 2024-03-06, 17:43:

There's a whole whack of boards look identical vis a vis the BIOS, crystal, resistor and cap layout in front of the SIMM sockets […]
Show full quote

There's a whole whack of boards look identical vis a vis the BIOS, crystal, resistor and cap layout in front of the SIMM sockets, but none of them have the 244 buffer chips installed...

https://theretroweb.com/motherboards/s/dfi-386sx-25-33-40ane
https://theretroweb.com/motherboards/s/kaimei-kmx-al-8
https://theretroweb.com/motherboards/s/kaimei-kmx-al-8517
https://theretroweb.com/motherboards/s/lucky- … -386sxa-ver-4-0
https://theretroweb.com/motherboards/s/unknown-612-3035-001

edit: ah these two do... (have buffers)
https://theretroweb.com/motherboards/s/auva-c … mputer-npm16-a0
https://theretroweb.com/motherboards/s/mg-8517-rev-2-2

But yeah, looks like all members of the AlI M1217 6 slot anonymous club have bank 0 labelled towards inside.

EditII: That ADS delay J5 3 pin is a bit of a mystery setting, not least because it's labelled J rather than JP leading some people to connect front panel stuff to it and causing weird problems. Should be set 1-2 enabled or 2-3 disabled with jumper cap. Disabled seems to be default for Intel CPU maybe enabled does something for Cyrix compatibility, unclear.

Good call! I agree that those seem like even better fits.

mkarcher wrote on 2024-03-06, 21:52:

Bad RAM can not prevent the A20 maskinig circuit from working - but bad RAM surely can just ignore the address bit that is mapped to A20. If the RAM ignores A20, it looks to the processor as if the A20 gate is permanently closed. If the bad RAM does care about A21, memory sizing might ignore the missing A20 bit, as memory sizing code does not expect A20 to be ignored if A21 is respected, so it might not check whether A20 is respected.

I think I will try going with just one of the two 4mb SIMMs that I have, trying them both out separately, to see if that makes any difference when it comes to the A20 issue.

Before switching SIMM cards, I installed MS-DOS 6.22 to see if that would work any better.

Booting it up in MS-DOS got me the same A20 error as before.

Interestingly, however, running HIMEM.SYS /M:1 sort of works at first, but before the config.sys process is finished, it reports the following (translated from Swedish):

ERROR: HIMEM.SYS has detected unreliable XMS memory at adress 00100000h.
The driver for XMS was not installed.

UPDATE: Okay, this is quite a head-scratcher. I switched out the two 4MB SIMM cards, replacing them with two 1MB SIMMs from my 286 mobo (as one can not, according the jumper manuals of all the likely motherboard suspects listed by BitWrangler [here], run 4MB using just one 4MB SIMM card).

Same A20 error as before and - here is the kicker - with /M:1, DOS reports the same unreliability error on the exact same address (00100000h).

New thought: Could this issue be PSU related? I am using an IBM XT power supply to feed the motherboard. The pinout for the P8+P9 connectors are identical, except for the fact that while the XT P8 connector has a KEY pin after the POWER GOOD one, an AT P8 provides +5v on the corresponding pin. As all +5v pins are connected together on the motherboard (if I have understood things correctly), this just means that an AT PSU provides extra current to +5 voltage.

As the 386 motherboard, small as it is and with much fewer discrete components than, for example, the 5160 motherboard (and since I lack power-hungry peripherals like MFM hard drives, et cetera), I thought that this should work fine, but perhaps this is what creates the A20 issue?

I have run boards like that on XT PSU before, it shouldn't be using more than a quarter of it's capacity, very low demands. However, XT supplies aren't getting any younger, best check you are getting something resembling 5V out of it, and not lower than 4.5.

About now, if I had the board here, I'd be inspecting the backside in detail for any minor dings and scratches, checking none of the pins of the SIMM sockets are dull or can be moved. Rule out simple mechanical/electrical problems first.

BitWrangler wrote on 2024-03-07, 15:27:

I have run boards like that on XT PSU before, it shouldn't be using more than a quarter of it's capacity, very low demands.

Yeah, that's what I thought as well.

BitWrangler wrote on 2024-03-07, 15:27:

However, XT supplies aren't getting any younger, best check you are getting something resembling 5V out of it, and not lower than 4.5.

Good point. I do have another XT case+PSU here. I could try the motherboard in that one, to see if that makes any difference.

BitWrangler wrote on 2024-03-07, 15:27:

About now, if I had the board here, I'd be inspecting the backside in detail for any minor dings and scratches, checking none of the pins of the SIMM sockets are dull or can be moved. Rule out simple mechanical/electrical problems first.

I will do that, and take a picture as well.

[This] is what the front of the motherboard looks like, by the way. I realized I never posted an actual photo of it earlier.