Sorry for the delay, gardening came up 😀

@mkarcher
Memory chips on the 4meg module:
First two lines are the same:Panasonic MN414100ASJ-07
Bottom line: 93465B 049 (first three); 93465B 012 (rest six)

@Horun
It's not a branded board. (or it's brand is TD). Below the sticker [TD(printed in blue, stylish font, rest is normal black font) S/N 357095] is a board print that I guess the board type: TD60K
Native 25Mhz board with a harris25mhz cpu, 50MHz oscillator near the CPU, 32Mhz near the FPU socket (probably for the 32/3 clock for the fpu).
Chipset has a sticker:
CITYGATE
D70-275
9215 Z00
JAPAN
The mobo has 4 sockets for simms, and 16 sockets for chip ram (20 pin ones).
The IDE controller is made by Acer, A2 REV-C print on it and its S/N (L9220300).
Using MSDOS 6.22

@Grzyb
It's called TestExt
https://archive.org/details/msdos_TESX34_shareware
I was thinking about using a 34pin sony with it's own controller, but yet to find it in the bess, same with the SCSI drive. Found a working 16bit controller with option rom so I can boot from it, yet to find my old ultraplex40, but working on it.

@maxtherabbit
The chipset should support it since it does work if I don't load the CDROM driver. Starts windows, plays dune with midi and digitized sound samples, vibra16c is working fine too.

@rmay635703
bios does not try to move the 384k theoretically. There is an option for that in bios to relocate the 384k and it is disabled.
Not sure about the rest though, will try to troubleshoot the less painful routes before having to take out the logic analyzer, that is a ROYAL pain 😀

@Anonymus Coward
As mentioned in the previous test result post:
2x4 megs in lower address space:
oakcdrom.sys - soft freeze, as always
vide-cdd.sys - hard freeze
gscdrom.sys - hard freeze

2x4 megs in upper address space:
no POST

@Jo22
Can you give me some ideas on that?
I've been trying with a Cirrus Logic card. Chip print:
[CL logo]Cirrus Logic
CL-GD5420-750C-C
26730-277CG
9413 T
KOREA-B

Will have to try the emu386 I guess, time for experimenting!

Thanks all for these smart hints, now I have an idea what to test next. Hopefully not with the logic analyzer (^.^);

I've also got a 286 board that also only recognises 4MB modules in two of the four sockets as well - it's a bit wacky, that one.

I'd be interested to see how well that system performs - my experience of 25MHz boards is that they nearly always insert wait states; I don't think 70ns modules (which is what your Panasonic chips appear to be) will stand up to 25MHz 0ws. Some of us have pushed 25MHz 286 systems to the limit, and properly configured, with zero-wait states and hand picked memory, they can absolutely fly. A lot of 20 and 25MHz designs are crippled as standard though.

There's quite a few more cdrom drivers here to try: https://www.hiren.info/downloads/dos-files

matskatsaba wrote on 2021-05-09, 09:28:

@Jo22 Can you give me some ideas on that? I've been trying with a Cirrus Logic card. Chip print: [CL logo]Cirrus Logic CL-GD5420 […]
Show full quote

@Jo22
Can you give me some ideas on that?
I've been trying with a Cirrus Logic card. Chip print:
[CL logo]Cirrus Logic
CL-GD5420-750C-C
26730-277CG
9413 T
KOREA-B

Will have to try the emu386 I guess, time for experimenting!

Hi, good luck! ^^ In the past EMU386 helped me to get some TSRs and other stuff running.
-> Re: Timeline of MS-DOS for NEC PC98 and more
-> Re: New game: Loonies 8192 (386+, VGA, OPL-2)
Maybe it helps you, too.

matskatsaba wrote on 2021-05-09, 09:28:

The chipset should support it since it does work if I don't load the CDROM driver. Starts windows, plays dune with midi and digitized sound samples, vibra16c is working fine too.

To 100% ensure that it's working properly here's what I would do:
1) Make a RAM Drive that's the size of all of your extended memory (excluding the HMA)
2) Load up the mTCP FTP server
3) connect to your 286 with a modern machine and flog the RAM drive with a file or files that come close to filling it 100%

I'd transfer over the files, download them back as a separate copy, and then do a binary comparison to ensure they survived the round trip intact. This will also verify that the RAM is being refreshed adequately.

matskatsaba wrote on 2021-05-09, 09:28:

@mkarcher
Memory chips on the 4meg module:
First two lines are the same:Panasonic MN414100ASJ-07

OK, the 4M modules need a 10-bit refresh counter, as I expected. What kind of chips are on your 1M modules? I would like to look up whether they can work with a 9-bit refresh counter (unlikely, but possible). I couldn't find a datasheet for the citygate d70 chipset, but I didn't try very hard to find one. If your BIOS setup allows you to select "hidden refresh", try enabling it. Hidden Refresh is a special kind of CAS-before-RAS-refresh, which is independent from the refresh counter on the board.

mkarcher wrote on 2021-05-09, 17:48:

OK, the 4M modules need a 10-bit refresh counter, as I expected. What kind of chips are on your 1M modules? I would like to look up whether they can work with a 9-bit refresh counter (unlikely, but possible). I couldn't find a datasheet for the citygate d70 chipset, but I didn't try very hard to find one. If your BIOS setup allows you to select "hidden refresh", try enabling it. Hidden Refresh is a special kind of CAS-before-RAS-refresh, which is independent from the refresh counter on the board.

I'm trying to find your reasoning here into how the memory modules make the cdrom driver fail ?
Care to explain.

weedeewee wrote on 2021-05-09, 18:24:

mkarcher wrote on 2021-05-09, 17:48:

OK, the 4M modules need a 10-bit refresh counter, as I expected. What kind of chips are on your 1M modules? I would like to look up whether they can work with a 9-bit refresh counter (unlikely, but possible). I couldn't find a datasheet for the citygate d70 chipset, but I didn't try very hard to find one. If your BIOS setup allows you to select "hidden refresh", try enabling it. Hidden Refresh is a special kind of CAS-before-RAS-refresh, which is independent from the refresh counter on the board.

I'm trying to find your reasoning here into how the memory modules make the cdrom driver fail ?
Care to explain.

Improper DRAM refresh can make ANY memory resident software or driver fail, because it corrupts the code while it sits in memory.

Some people in this thread are taking the statement that the system "works perfectly" other than the CD driver at face value. I (and I assume mkarcher also) am not

maxtherabbit wrote on 2021-05-09, 18:32:

Improper DRAM refresh can make ANY memory resident software or driver fail, because it corrupts the code while it sits in memory.

Some people in this thread are taking the statement that the system "works perfectly" other than the CD driver at face value. I (and I assume mkarcher also) am not

true, though then you would see more instability than just the cdrom driver, since the whole os also resides in that same memory, as well as the interrupt routines that get called very frequently.

but it's not all the "same memory" it's possible parts of the RAM are being refreshed and others are not

maxtherabbit wrote on 2021-05-09, 18:59:

but it's not all the "same memory" it's possible parts of the RAM are being refreshed and others are not

Considering it's dos and it's a 286, it's all the same first 1MB of memory with a 16bit memory bus.
512KB of two memory modules would be used. in which the cdrom driver would be loaded, the same space as which the os is loaded etc. blah

What would be the easiest way to test this 'refresh problem' idea, considering op linked to a memory test program that would've tested all of the memory.
then we could discuss the efficacy of the test program or if op actually ran it.
So
ramdrive in extended memory.
copy files over
wait
compare files on disk to files in ramdrive.
if different there is a memory problem,
if not...

Right

I was running the TestExt program I have linked previously, and left it overnight.
All that said, it is possible that there is a read/write error that the tester program does not notice, so the first thing I will try, is the ramdrive.

The 1meg simms:
2x KMM44C1000-CJ-6 and one BP41C1000A-6

I am fairly certain that the 4meg modules are fine, since the symptomes did not change if I swapped the modules around (in a 2x1+2x4 combination).
There might be a hardware related and/or refresh problem since it's an old board, and it does have capacitors. So this is the plan for now:
0. Checking out the hidden refresh option
1. Going to try the ramdrive copy-reread test in a few passes.
2. Trying the setup with a few more different video cards (in case of conflict)
3. Trying different IDE controllers
4. Getting back to you with the results (And also probably will recap the board with rubycon caps, just to be on the safe side.

matskatsaba wrote on 2021-05-09, 19:56:

I am fairly certain that the 4meg modules are fine, since the symptomes did not change if I swapped the modules around (in a 2x1 […]
Show full quote

I am fairly certain that the 4meg modules are fine, since the symptomes did not change if I swapped the modules around (in a 2x1+2x4 combination).
There might be a hardware related and/or refresh problem since it's an old board, and it does have capacitors. So this is the plan for now:
0. Checking out the hidden refresh option
1. Going to try the ramdrive copy-reread test in a few passes.
2. Trying the setup with a few more different video cards (in case of conflict)
3. Trying different IDE controllers
4. Getting back to you with the results (And also probably will recap the board with rubycon caps, just to be on the safe side.

While you're at it, try freedos as well.

Justr put it back to the original setup and be happy with that if it was working fine as per youre original post.

Keep the 30 pin 4meg simms for another system.

Yeah will try, freedos as well, just got overwhelmed with work today.

I could chose not to max out the memory, I know. But the point of this build is to max this 286 out as much as possible, so while your advice would work, it is not an option. The reason I want to find out the reason for this fault, because if it's a hardware fault on the mobo, then it needs to be addressed by either a repair or a swap (though it's not easy to find native 25mhz 286 boards that can see the modules).

I certainly need to get more 4meg modules (or at worst case scenario, PCBway them since I have tons of 1meg 3chip simms), since these were from a 486 that will need memory to run too. But that's a project for later.

matskatsaba wrote on 2021-05-09, 09:28:

@Horun It's not a branded board. (or it's brand is TD). Below the sticker [TD(printed in blue, stylish font, rest is normal blac […]
Show full quote

@Horun
It's not a branded board. (or it's brand is TD). Below the sticker [TD(printed in blue, stylish font, rest is normal black font) S/N 357095] is a board print that I guess the board type: TD60K
Native 25Mhz board with a harris25mhz cpu, 50MHz oscillator near the CPU, 32Mhz near the FPU socket (probably for the 32/3 clock for the fpu).
Chipset has a sticker:
CITYGATE
D70-275
9215 Z00
JAPAN
The mobo has 4 sockets for simms, and 16 sockets for chip ram (20 pin ones).
The IDE controller is made by Acer, A2 REV-C print on it and its S/N (L9220300).
Using MSDOS 6.22

Thanks ! Wish you could take a good picture of the board and sounds like a real nice one. As others have mentioned many 286 had a 16Mb limit but it also depended what ISA cards and what the BIOS settings are.
Without knowing the exact chipset makes me wonder about full 25Mhz ability since you also have DRAM sockets, as most mature 286 chipsets (like Headland fast chipsets) only had simms.
If only you had the original manual....

I'm in the middle of recapping the board right now (16 caps left).
Will upload a pic as soon as I'm done with this.
If nothing helps, I will try to remove the sticker from the chipset and dump the eprom. (not an otp rom, jsut a regular uv-eprom)
Edit:
Quick update:
Recapping the board did nothing.
Coming up: Different video card and IDE controller

matskatsaba wrote on 2021-05-09, 19:56:

The 1meg simms:
2x KMM44C1000-CJ-6 and one BP41C1000A-6

Thanks for the part numbers. This confirms that the specified refresh requirements of both types of modules are identical. These chips also need a 10-bit refresh counter and an average refresh timing of a refresh every 16µs (i.e. 16ms for the whole chip).

weedeewee wrote on 2021-05-09, 18:24:

I'm trying to find your reasoning here into how the memory modules make the cdrom driver fail ?
Care to explain.

You did very thourough experimenting with different kinds of modules, and those experiments have a clear pattern: The one CD driver that works at all works always if 1M modules are in the low bank, and never works if 4M modules are in the low bank. While the behaviour of simple DOS drivers shouldn't depend on the amount of memory installed, as they use the low 1M only, your experiments show that the amount of memory is not the factor that determines whether the system crashes or not: Your system works perfectly with the 1M SIMMs and the extended memory cards. Your system works perfectly with the 1M SIMMs in the low bank, and 4M SIMMs in the high bank (10M total), but it crashes with only the 4M SIMMs in the low bank (just 8M total).

For me, this clearly shows that the 4M modules are incompatible with your mainboard. The questions I was asking were trying to gather more information that can tell the root cause of the incompatibility. Refresh problems are very hideous to troubleshoot, because refresh problems are only visible with certain access patterns. Especially, many memory test programs access all the rows of the memory often enough that the test access itself is enough to keep the RAM refreshed. On the other hand, when there are delays (like probing for IDE CD-ROMs), the RAM relies on refreshing. That's why your symptoms could be explained by refresh issues. As your 4M sticks and your 1M sticks require the same signalling for refreshing, the theory that refresh problems cause your issue looses some support, though. It's still possible that the 1M modules are more tolerant towards insufficient refresh than the 4M modules, but other reasons for your issues are likely more likely.

mkarcher wrote on 2021-05-10, 18:12:

For me, this clearly shows that the 4M modules are incompatible with your mainboard. The questions I was asking were trying to gather more information that can tell the root cause of the incompatibility. Refresh problems are very hideous to troubleshoot, because refresh problems are only visible with certain access patterns. Especially, many memory test programs access all the rows of the memory often enough that the test access itself is enough to keep the RAM refreshed. On the other hand, when there are delays (like probing for IDE CD-ROMs), the RAM relies on refreshing. That's why your symptoms could be explained by refresh issues. As your 4M sticks and your 1M sticks require the same signalling for refreshing, the theory that refresh problems cause your issue looses some support, though. It's still possible that the 1M modules are more tolerant towards insufficient refresh than the 4M modules, but other reasons for your issues are likely more likely.

Agreed, but I still advocate for the RAMDrive test to put this to bed once and for all

mkarcher wrote on 2021-05-10, 18:12:

You did very thourough experimenting with different kinds of modules, and those experiments have a clear pattern: The one CD driver that works at all works always if 1M modules are in the low bank, and never works if 4M modules are in the low bank. While the behaviour of simple DOS drivers shouldn't depend on the amount of memory installed, as they use the low 1M only, your experiments show that the amount of memory is not the factor that determines whether the system crashes or not: Your system works perfectly with the 1M SIMMs and the extended memory cards. Your system works perfectly with the 1M SIMMs in the low bank, and 4M SIMMs in the high bank (10M total), but it crashes with only the 4M SIMMs in the low bank (just 8M total).

For me, this clearly shows that the 4M modules are incompatible with your mainboard. The questions I was asking were trying to gather more information that can tell the root cause of the incompatibility. Refresh problems are very hideous to troubleshoot, because refresh problems are only visible with certain access patterns. Especially, many memory test programs access all the rows of the memory often enough that the test access itself is enough to keep the RAM refreshed. On the other hand, when there are delays (like probing for IDE CD-ROMs), the RAM relies on refreshing. That's why your symptoms could be explained by refresh issues. As your 4M sticks and your 1M sticks require the same signalling for refreshing, the theory that refresh problems cause your issue looses some support, though. It's still possible that the 1M modules are more tolerant towards insufficient refresh than the 4M modules, but other reasons for your issues are likely more likely.

This is somehow difficult to read for me, though I think I get what you mean.

I'm waiting for the results with the freedos cdrom driver 🤞