mmmh, I forget. How much physical memory does a 386sx support? I though it was the same as a 286, which would be 16MB. 24 address lines.

we'll just have to wait and see what someone who has a mainboard with the same chipset can verify is supported.
or
maybe mkarcher wants to chime in 😀

no clue about what your supposed to see on the signal lines. I only assume now that since A11 isn't connected, that 16M simms won't work. 😒
I wonder if the chipset actually supports more than 4MB and the listings that mention mainboards supporting more than 4 aren't just in error.

weedeewee wrote on 2024-08-17, 05:24:

Have you tried the "4x 16MB 8 chip. ICs: KM41C16000AK-6. PCB: no marking " on the board ?

Yes, this was my first attempt, wherein I (re-)learned that this thing is a 386sx and therefore can't address more than 16MB of memory and needs to see SIMMs in pairs. This was the post I (re-)learned that from and now use as a reference. 😀

And, just to be sure (hey I do this stuff really late at night and my memory isn't great anymore), I just tried it again.

It didn't boot on the first try with all four sticks. I thought to myself "Did I ever try this with just two?" And so I tried with just two. It also didn't boot, but I thought "I should make sure everything is seated well" and went to power it off and flubbed pressing the switch and just sorta power cycled it briefly.

And what do you know? I got a VGA bios and it POSTed, counting 8MB of RAM. Once I got through its complaints about the RAM size changing, I tried booting to a MemTest floppy. It got stuck accessing the floppy. But at this point, I knew the magic spell, and power cycled it quickly again. This time it got all the way to loading MemTest and crashed. And again, this time it got all the way to starting MemTest, examined some memory, and started having wild errors like "writing garbage characters all over the screen".

So I had a theory at this point. I knew that at least one pin of those SIMMs was just straight up floating. So I was likely to have which quarter of the SIMM I was talking to just change randomly.

So then I went down the path of "well maybe this RAM is now bad for some reason" and swapped in all the sticks in pair combinations. All of them behaved the same.

Then I tried putting all four in and doing the quick power cycle. And it actually counted 16MB of memory... and did all the weird crashing stuff too.

Note that it was counting 8MB when it had 16MB installed and counting 16MB when it had 64MB installed.

So I think this confirms it. This board is probably only refreshing the bottom 10 address pins' worth of rows.

Now I want to make a board that sits between a SIMM and a socket and tells you how many rows the memory controller is refreshing.

AND some kind of bodge-wire-and-logic-chip thing to add additional refresh rows for SIMMs without 8 or 9 chips. Hopefully that might keep people from tearing up parity SIMMs.

I have a lot better understanding of how I'll design my RAM disk card, too.

wierd_w wrote on 2024-08-17, 06:12:

Why dead bug?

There's a perfectly good pc104 header and mounting holes. Nobody says the stuff on the bottom-most card HAS to be connected to the PC104 header.

Very good point! I was thinking I wanted to do something out of the way, and the back of the board has some open space. But using a PC104 card with pass-through headers is definitely the way to go!

NoTrueSpaceman wrote on 2024-08-17, 07:07:

Yes, this was my first attempt, wherein I (re-)learned that this thing is a 386sx and therefore can't address more than 16MB of […]
Show full quote

weedeewee wrote on 2024-08-17, 05:24:

Have you tried the "4x 16MB 8 chip. ICs: KM41C16000AK-6. PCB: no marking " on the board ?

Yes, this was my first attempt, wherein I (re-)learned that this thing is a 386sx and therefore can't address more than 16MB of memory and needs to see SIMMs in pairs. This was the post I (re-)learned that from and now use as a reference. 😀

And, just to be sure (hey I do this stuff really late at night and my memory isn't great anymore), I just tried it again.

It didn't boot on the first try with all four sticks. I thought to myself "Did I ever try this with just two?" And so I tried with just two. It also didn't boot, but I thought "I should make sure everything is seated well" and went to power it off and flubbed pressing the switch and just sorta power cycled it briefly.

And what do you know? I got a VGA bios and it POSTed, counting 8MB of RAM. Once I got through its complaints about the RAM size changing, I tried booting to a MemTest floppy. It got stuck accessing the floppy. But at this point, I knew the magic spell, and power cycled it quickly again. This time it got all the way to loading MemTest and crashed. And again, this time it got all the way to starting MemTest, examined some memory, and started having wild errors like "writing garbage characters all over the screen".

So I had a theory at this point. I knew that at least one pin of those SIMMs was just straight up floating. So I was likely to have which quarter of the SIMM I was talking to just change randomly.

So then I went down the path of "well maybe this RAM is now bad for some reason" and swapped in all the sticks in pair combinations. All of them behaved the same.

Then I tried putting all four in and doing the quick power cycle. And it actually counted 16MB of memory... and did all the weird crashing stuff too.

Note that it was counting 8MB when it had 16MB installed and counting 16MB when it had 64MB installed.

So I think this confirms it. This board is probably only refreshing the bottom 10 address pins' worth of rows.

Now I want to make a board that sits between a SIMM and a socket and tells you how many rows the memory controller is refreshing.

AND some kind of bodge-wire-and-logic-chip thing to add additional refresh rows for SIMMs without 8 or 9 chips. Hopefully that might keep people from tearing up parity SIMMs.

I have a lot better understanding of how I'll design my RAM disk card, too.

Well, that's some... improvement/insights. The refresh cycle is done by the chipset if i'm not mistaken.
Doubt you'll find a connection for that one the ISA or PC/104 connector.

weedeewee wrote on 2024-08-17, 06:18:

mmmh, I forget. How much physical memory does a 386sx support? I though it was the same as a 286, which would be 16MB. 24 address lines.

Yes, 16MB total because it has 24 address lines. And in case you were wondering, those top four are pins 75, 76, 79, and 80 (A20 through A23, in that order) on this CPU.

weedeewee wrote on 2024-08-17, 06:18:

no clue about what your supposed to see on the signal lines. I only assume now that since A11 isn't connected, that 16M simms won't work. 😒
I wonder if the chipset actually supports more than 4MB and the listings that mention mainboards supporting more than 4 aren't just in error.

My previous post shows that with some finagling, the don't exactly "work" but I could see the board come alive.

Considering that I could get it to count up to both 8MB and 16MB, I think it supports it, but probably only using 8 or 9 chip SIMMs. The question now is if there is some modification (ideally in the BIOS setup of the ALI M1217!) that would add support for 2 and 3 chips SIMMs. And if not in the BIOS, maybe a (relatively) simple hardware mod.

That would be a big help for people trying to get these low end 386s working to their highest potential. Which frankly is a lot - 16MB is plenty to run lots of cool operating systems. I know this is a video-game oriented site but for some of us, running "ancient" NetBSD is a game 😀

NoTrueSpaceman wrote on 2024-08-17, 07:42:

weedeewee wrote on 2024-08-17, 06:18:

mmmh, I forget. How much physical memory does a 386sx support? I though it was the same as a 286, which would be 16MB. 24 address lines.

Yes, 16MB total because it has 24 address lines. And in case you were wondering, those top four are pins 75, 76, 79, and 80 (A20 through A23, in that order) on this CPU.

No idea where you're going with that, considering the simm slots do not see those address lines.

My previous post shows that with some finagling, the don't exactly "work" but I could see the board come alive. […]
Show full quote

weedeewee wrote on 2024-08-17, 06:18:

no clue about what your supposed to see on the signal lines. I only assume now that since A11 isn't connected, that 16M simms won't work. 😒
I wonder if the chipset actually supports more than 4MB and the listings that mention mainboards supporting more than 4 aren't just in error.

My previous post shows that with some finagling, the don't exactly "work" but I could see the board come alive.

Considering that I could get it to count up to both 8MB and 16MB, I think it supports it, but probably only using 8 or 9 chip SIMMs. The question now is if there is some modification (ideally in the BIOS setup of the ALI M1217!) that would add support for 2 and 3 chips SIMMs. And if not in the BIOS, maybe a (relatively) simple hardware mod.

That would be a big help for people trying to get these low end 386s working to their highest potential. Which frankly is a lot - 16MB is plenty to run lots of cool operating systems. I know this is a video-game oriented site but for some of us, running "ancient" NetBSD is a game 😀

I think re-reading the mkarcher post or finding some more of his posts related to ram might be helpful.
Indeed having max memory on those old systems is very nice.

I am confused about why you want to design a hardware ramdisk, when they already exist.

ACARD made one that uses 68pin DIMMs, and is SATA.

https://www.reddit.com/r/retrobattlestations/ … lete_the_acard/

I cant seem to find any still being made though.

weedeewee wrote on 2024-08-17, 07:33:

NoTrueSpaceman wrote on 2024-08-17, 07:07:

Yes, this was my first attempt, wherein I (re-)learned that this thing is a 386sx and therefore can't address more than 16MB of […]
Show full quote

weedeewee wrote on 2024-08-17, 05:24:

Have you tried the "4x 16MB 8 chip. ICs: KM41C16000AK-6. PCB: no marking " on the board ?

Yes, this was my first attempt, wherein I (re-)learned that this thing is a 386sx and therefore can't address more than 16MB of memory and needs to see SIMMs in pairs. This was the post I (re-)learned that from and now use as a reference. 😀

And, just to be sure (hey I do this stuff really late at night and my memory isn't great anymore), I just tried it again.

It didn't boot on the first try with all four sticks. I thought to myself "Did I ever try this with just two?" And so I tried with just two. It also didn't boot, but I thought "I should make sure everything is seated well" and went to power it off and flubbed pressing the switch and just sorta power cycled it briefly.

And what do you know? I got a VGA bios and it POSTed, counting 8MB of RAM. Once I got through its complaints about the RAM size changing, I tried booting to a MemTest floppy. It got stuck accessing the floppy. But at this point, I knew the magic spell, and power cycled it quickly again. This time it got all the way to loading MemTest and crashed. And again, this time it got all the way to starting MemTest, examined some memory, and started having wild errors like "writing garbage characters all over the screen".

So I had a theory at this point. I knew that at least one pin of those SIMMs was just straight up floating. So I was likely to have which quarter of the SIMM I was talking to just change randomly.

So then I went down the path of "well maybe this RAM is now bad for some reason" and swapped in all the sticks in pair combinations. All of them behaved the same.

Then I tried putting all four in and doing the quick power cycle. And it actually counted 16MB of memory... and did all the weird crashing stuff too.

Note that it was counting 8MB when it had 16MB installed and counting 16MB when it had 64MB installed.

So I think this confirms it. This board is probably only refreshing the bottom 10 address pins' worth of rows.

Now I want to make a board that sits between a SIMM and a socket and tells you how many rows the memory controller is refreshing.

AND some kind of bodge-wire-and-logic-chip thing to add additional refresh rows for SIMMs without 8 or 9 chips. Hopefully that might keep people from tearing up parity SIMMs.

I have a lot better understanding of how I'll design my RAM disk card, too.

Well, that's some... improvement/insights. The refresh cycle is done by the chipset if i'm not mistaken.
Doubt you'll find a connection for that one the ISA or PC/104 connector.

From what I am grasping, he wants to implement a timed RAS/CAS circuit that does the RAM refresh signals that are missing, (Maybe tripped by catching the edge of the last refresh generated by the chipset?) on the card. The PC104 socket would just be a convenient place to place such driver chips, and would not be attached to the ISA bus. It would take some IDC or DuPont leads going to bodges on the motherboard, to get / deliver these to the SIMM sockets. Straight up bodging the missing data bus lines that are missing above A10 to the CPU on the underside, in combination with the above (and some way to delay the next RAS/CAS cycle so it stays properly synchronized?) might be enough to get around the problem.

(he will *NEED* to bodge the A11-A23 address lines in order to count to 16mb on the present SIMM sockets correctly anyway, regardless. Their being NC means there is no way to signal the chips to respond at all!!)

I suggested the PC104 header as a nice place to get GND and +5v.

Ok tested some 70nS 4Mb simms, both 3 chip and 9 chip types. Tested with 2 in bank 0 (8Mb) and then with 4 filling all banks (16Mb). Even tested some 80nS 4Mb simms and all worked !!
Also tested some 8 chip simms and they worked also, but had "parity check" disabled in bios for those.....
Ran Memtest86+ v4.10 and when tests got over 50% w/o error stopped them (had previously tested all the simms on a cyrix 486dx2-80 based board) no issues other than slow memory speeds 😁 .
So the ALI M1217-40 chipset will work with 4 x 4Mb simms if the address traces are there......
Yes I have a newer Varta battery so not worried about corrosion yet...

Horun wrote on 2024-08-17, 20:11:

Ok tested some 70nS 4Mb simms, both 3 chip and 9 chip types.
Also tested some 8 chip simms and they worked also

Thanks for testing your board, Horun. You don't happen to have any 2 chip 4MB SIMMs to test, do you?

Horun wrote on 2024-08-17, 20:11:

So the ALI M1217-40 chipset will work with 4 x 4Mb simms if the address traces are there......

It seems my board has all the traces connected as well (other than A11, which is okay).

So if the chipset is okay with it, could it be a difference in the BIOS? Perhaps my BIOS doesn't set up the DRAM controller on the M1217 correctly? I guess now is the time to dig into that earlier datasheet for the M6117D. Reading through the section on the DRAM controller, I see things that relate to options in the BIOS menu, so maybe I can disassemble my BIOS to see what it's doing. But I also don't see anything that seems to control the width of the row refresh.

And it also mentions supporting up to 64MB of memory, so this just might be too different of a device to use as reference. 🙁

wierd_w wrote on 2024-08-17, 09:25:

I am confused about why you want to design a hardware ramdisk, when they already exist.
I cant seem to find any still being made though.

Because of that last part right there!

And personally, I have to ignore my own feelings of "why do this if it's already been made?" because nearly everything has already been made. Especially in the retro space, for enough money. If I let my own self criticisms win, I'd not be bothering with any of this.

Earlier in this thread I mentioned "a RPi daughter board" and just this morning I stumbled across the PicoMem on Texelec... it's already been made! Already doing all the things I thought I could have fun figuring out! Now what do I do? Make my own version? Abandon the whole idea? Honestly I don't know. It certainly took some of the wind out of my sails.

So maybe I'm re-inventing things but doing it in a different way that gives a bit back to the community or inspires the next person to try a cool project or helps bootstrap some new project.

I'm talking mostly to myself at this point. 😀

Any how - thanks for helping me figure out what was going on with my machine.

Sorry no, do not have any two chip 4Mb 30 pin simms.