And here’s the link to the thread with more details about the card.
Single board computer 386-sx40 skipping VGA BIOS?

Your card SoB is probably not designed to put that much RAM on the main system.

I see that it has a PC104 header connector. You might consider looking into things to put on that.

You will likely be relegated to having to use something like this, to put more RAM in. Bear in mind, that these are going to be DOG SLOW, as they have to go over the ISA bus. (386sx should be more or less the top end I would EVER consider using one of these in!! It should be enough to get some of the more cranky windows programs that wanted more than 4mb present to run, even if these things ARE super $$$ these days.)

https://www.ebay.com/itm/135074723562?_nkw=16 … %3ABFBMmp6C2Kpk

For general purpose use though, that PC104 connector is basically just another exposed ISA bus, and quite a few peripheral devices were made for that format. You literally can just slap it on top of your SoB.

https://pdf1.alldatasheet.com/datasheet-pdf/d … 80386SX-25.html

Dunno if that helps.

Might need the datasheets for the memory too.
But I honestly don't know enough to help.

There's a guy on YouTube who made switchable fast page/edi Simms who went into fair detail about his issues.

I'll see if I can find it

Found https://www.youtube.com/watch?v=FNOqEWzRiBQ

Anyway. If you think A10 isn't connected, you could always do a diode mode measurement, if you have a multimeter.

edit: also a very rough guess. It could be a problem with one of the hct245 near the cpu.

While this, https://theretroweb.com/chipsets/169#docs ,is absolutely not what is on your board,
Even though from the retroweb page it looks like it.
It can give some insights into the software side of it, considering the m6117D apparently is the m1217B combined with a 386sx
Though it leaves one to wonder what the functional differences are between the M1217 & M1217B considering the M6117D apparently supports EDO.
And reading that datasheet isa bit mindboggling if you read it supports 16M bytes of memory and 16M SIMMs (it's mindboggling since it's a 16bit memory path and a 30p simm is only 8 bit... oh wait it is possible if the hardware splits it up which would mean it's F'n slow.)

Your card SoB is probably not designed to put that much RAM on the main system.

I was afraid of that.

I see that it has a PC104 header connector. You might consider looking into things to put on that.

One thing that interested me about this system-on-a-card is that it opens up the possibility of using weird peripherals that were only/better supported by PC104 cards.

Also it seems easier to design boards for PC104 because there is no need for card edge connectors or back plates.

You will likely be relegated to having to use something like this, to put more RAM in. Bear in mind, that these are going to be DOG SLOW, as they have to go over the ISA bus.

Okay, thanks for the info. At least it’s a way out. Have you ever heard of a card that supports many MB of RAM? I think I’ve only seen cards that go up to 2-3MB. Is there any restriction why one could support the full 24bit 16MB address space?

… that PC104 connector is basically just another exposed ISA bus … you literally can just slap it on top of your SoB.

PC104 Raspberry Pi daughter card anyone? 😀

myne wrote on 2024-08-15, 10:23:

https://pdf1.alldatasheet.com/datasheet-pdf/d … 80386SX-25.html

Dunno if that helps.

Actually it does. I read through the pin descriptions and it seems like the 386sx is still pretty similar to the “8 bit microprocessor” style of the 8088, 8085, and Z80. It doesn’t have any specific memory interface or mode pins as far as I can tell. It looks like it literally just sits on the bus like anything else.

That’s good news to me as I’m pretty familiar with that world. When things like cache and stuff start getting involved my brain leaks out my ears at the circuit level.

I’ll check out that video too. I pretty much understand the physical difference in the interface between EDO and FPM DRAM so I should be able to follow along.

NoTrueSpaceman wrote on 2024-08-15, 16:22:

Okay, thanks for the info. At least it’s a way out. Have you ever heard of a card that supports many MB of RAM? I think I’ve only seen cards that go up to 2-3MB. Is there any restriction why one could support the full 24bit 16MB address space?

There were ISA boards that supported 8mb, you had to install 2x to hit the limit. There were also 386sx “like” CPUs that could use more than 16mb like the 386sl with 32mb

The ISA Boards I saw that supported more than 8mb were usually very late EMS 4 (sometimes EMS/XMScombo) boards.

The trouble you will encounter is many motherboards flake out with much more than 12mb unless they support a memory hole because theoretically 16bit memory mapped stuff can sit up there making a chunk unusable.

Some systems handled it gracefully others just wouldn’t use anything up there, others would lock. Some of the oldest dumbest AT class systems had no issue with 16mb so long as you didn’t have anything using that memory area.

So support is all over the map

4 MB was a common RAM limit of 286 chipsets (some supported more).

Installing an ISA RAM card is possible any time.

Performance penalty is relative, it depends on how good the chipset is and thus how far the gap between ISA bus is.

NoTrueSpaceman wrote on 2024-08-15, 16:22:

I was afraid of that. […]
Show full quote

Your card SoB is probably not designed to put that much RAM on the main system.

I was afraid of that.

I see that it has a PC104 header connector. You might consider looking into things to put on that.

One thing that interested me about this system-on-a-card is that it opens up the possibility of using weird peripherals that were only/better supported by PC104 cards.

Also it seems easier to design boards for PC104 because there is no need for card edge connectors or back plates.

You will likely be relegated to having to use something like this, to put more RAM in. Bear in mind, that these are going to be DOG SLOW, as they have to go over the ISA bus.

Okay, thanks for the info. At least it’s a way out. Have you ever heard of a card that supports many MB of RAM? I think I’ve only seen cards that go up to 2-3MB. Is there any restriction why one could support the full 24bit 16MB address space?

… that PC104 connector is basically just another exposed ISA bus … you literally can just slap it on top of your SoB.

PC104 Raspberry Pi daughter card anyone? 😀

vintage ones usually top out at 8mb, but combined with the 4 on the memory bus, that is 12. That's plenty for a 386.

"Bocaram AT Plus"for example, can go up to 8.

There is precisely ZERO reason that a homebrewed ISA RAM module would be limited like that though, with the goodies readily available these days. Hell, you could slap a particularly large parallel SRAM in there, and make it dead-nuts simple, electronically speaking. The system wouldn't care. XMS memory does not need fancy windowing or latching circuits, it just needs some bus logic glue, and these days, you could probably do that with a dipswitch block and an EEPROM. (in place of GAL/PAL chips. You can have quite a few address pins on larger EEPROMs, to provide your conditional logic on.)

Some are even kinda inexpensive, like this 64mbit pseudo SRAM. (Too bad its BGA.. and 3.3v only. It'd need a regulator circuit.)
https://www.mouser.com/ProductDetail/ISSI/IS6 … 52BLvIqsw%3D%3D

If I was going to do that, I would make the module for the PC104 header, so the total bus travel distance is very low.

Thanks everyone! It’s great that y’all share your knowledge so I know I’m not barking up the wrong tree or heading towards a cliff.

Honestly for all the DOS stuff I’ve done, 4MB has been plenty. But I’m interested in other operating systems and usually there things jsut get better and better the more RAM you have if only so you can cache things into a RAM disk off the network. Linux, NetBSD, even a shot at the world’s worst Windows 95 machine.

(I have also thought about homebrewing a hardware RAM disk that presents as a HDD or floppy controller but that’s another story.)

I’ll look into what SRAM I can get. It seems like all the large capacity SRAM is actually PseudoSRAM (which is fine). There’s also the SPI PSRAM, which would be gluing together two very distant technologies! But I suspect I’ll go with some kind of DRAM as a chance to learn more about it.

There’s still the question that perhaps this could be a fix of adding a few wires. So I’ll start by mapping out the connections on the board to see if there is an easy win for this chipset. I’ll go looking once I have time, but if anyone has the datasheet for the ALI M1217, I’d greatly appreciate it.

Wow the data sheet for the M1217 is really hard to find.

Documentation of other boards with the chip indicate that they supported some of the configurations I tried unsuccessfully.

(The docs on this site aren’t always to be trusted but I suspect in this case they’re probably okay)
https://stason.org/TULARC/pc/motherboards/U/U … -ALI-M1217.html

I wonder if this is a 8/9 chip versus 2/3 chip SIMM problem. It works fine with two chip 1MB sticks.

I really really _really_ don’t want to buy even more RAM but I guess if I’m going to be designing RAM disks I need a fair set to test with. This would be the third set for this card….

Jo22 wrote on 2024-08-15, 17:43:

4 MB was a common RAM limit of 286 chipsets (some supported more).

Installing an ISA RAM card is possible any time.

Performance penalty is relative, it depends on how good the chipset is and thus how far the gap between ISA bus is.

386sx systems usually topped out around 12mbs depending on if there was some form of cache or ram interleaving and whether the board could actually run decent timing.

He would need to test the memory speed to know.

From what I remember ISA ram is 8.33mbs both up and down (without overclocking) which isn’t that far off the performance you were usually stuck with anyway unless your motherboard has advanced features or cache.

NoTrueSpaceman wrote on 2024-08-15, 23:08:

Wow the data sheet for the M1217 is really hard to find. […]
Show full quote

Wow the data sheet for the M1217 is really hard to find.

Documentation of other boards with the chip indicate that they supported some of the configurations I tried unsuccessfully.

(The docs on this site aren’t always to be trusted but I suspect in this case they’re probably okay)
https://stason.org/TULARC/pc/motherboards/U/U … -ALI-M1217.html

I wonder if this is a 8/9 chip versus 2/3 chip SIMM problem. It works fine with two chip 1MB sticks.

I really really _really_ don’t want to buy even more RAM but I guess if I’m going to be designing RAM disks I need a fair set to test with. This would be the third set for this card….

It probably is the chip arrangement, but also, you said you tested it on a mac.
You know what Apple are like for rewiring simple things to force you to buy their branded crap.

Watch the vid. He made pcbs for simms. I'm sure it will give you some clues.

I have this: https://theretroweb.com/motherboards/s/ecs-8517-v3.3
Same ALI chipset and cpu. Will dig it out and test it with some 4Mb 9chip simms this weekend (unless called into work). It's AMI BIOS does have up to 16Mb in the ram strings.....

NoTrueSpaceman wrote on 2024-08-15, 23:08:

Wow the data sheet for the M1217 is really hard to find. […]
Show full quote

Wow the data sheet for the M1217 is really hard to find.

Documentation of other boards with the chip indicate that they supported some of the configurations I tried unsuccessfully.

(The docs on this site aren’t always to be trusted but I suspect in this case they’re probably okay)
https://stason.org/TULARC/pc/motherboards/U/U … -ALI-M1217.html

I wonder if this is a 8/9 chip versus 2/3 chip SIMM problem. It works fine with two chip 1MB sticks.

I really really _really_ don’t want to buy even more RAM but I guess if I’m going to be designing RAM disks I need a fair set to test with. This would be the third set for this card….

Which kind of simms do you have at the moment?
and
Which test equipment do you have at your disposal?

Horun wrote on 2024-08-16, 02:36:

Same ALI chipset and cpu. Will dig it out and test it

Thank you! With the pin out or datasheet to the ALI M1217, I could figure this out in like 20 minutes. Maybe I'll start making my own pin out to help the next person along.

weedeewee wrote on 2024-08-16, 05:06:

Which kind of simms do you have at the moment?

30 pin SIMMs:

• 4x 4MB 2 chip that originally had a third chip but someone has desoldered the parity chips. ICs: HY5117400JC-70. PCB: PNY
• 4x 1MB 2 chip with boards for 3 chip, but never had one soldered on. ICs: NEC 424400-80. PCB: NEC MC-421000A8BB-80
• 4x 1MB 2 chip with boards for 3 chip, also never added. ICs: HM514400AS8. PCB: TKK-5 HB56G18B-8A
• 3x 256k 3 chip. ICs: two M5M44256BJ-7 and one BJ1X1000-6 PCB: B3140 REV.A 94V-0
• 1x 256K 3 chip. ICs: two KM44C256AP-8 and one V53C256P80 PCB: Daewoo P/N23050-05501
• 1x 256K 3 chip. ICs: two HM514256AJP10 and one MN41256A-08 PCB: SSM251A9B J0345BA IBI,M2V0
• 1x 256K 2 chip. ICs: two KM44C256AP-10 PCB: Copyright (c)1988 Network Computing Devices Inc. 256K SIMM SBC-2A 94V-0 25-90

Irrelevant / Oddballs

• 4x 16MB 9 chip (current in a Mac IIsi, and I don't want to dig them out)
• 4x 16MB 8 chip. ICs: KM41C16000AK-6. PCB: no marking
• 7x 256k 3 chip boards. Socketed for 2x 18pin DIPs and 1x 16pin DIP. ICs: none (because I pulled them out of the socket to use in a memory expansion card). PCB: Olympic Computer 441. 5 have bodge wires. 2 don't
• 1x 256k 3 chip boards. Same as above but with pins soldered on to make it into a SIPP. (I got it that way!)

Additionally,

• 4x 4MB 9 chip sticks on their way....

weedeewee wrote on 2024-08-16, 05:06:

Which test equipment do you have at your disposal?

I have

• Rigol DS1102E 2ch 100MHz 1GSa/s digital oscope
• BK 2704B DMM
• a bachelors degree in electrical engineering from about 20 years ago
• all the various kinds of microcontrollers and dev boards you'd accumulate after 20+ years of messing around in electronics as a hobbyist

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

weedeewee wrote on 2024-08-15, 13:22:

Anyway. If you think A10 isn't connected, you could always do a diode mode measurement, if you have a multimeter.

edit: also a very rough guess. It could be a problem with one of the hct245 near the cpu.

I used a DMM to trace some connections.

First, A10 of the SIMMs is connected. A11, is NOT. Not even to ground or VCC. I think a quick hack would be to connect all the A11 lines of the SIMMs together to ground and see if a 16MB SIMM will show up as a 4MB SIMM. It seems weird they left it N/C.

Second, the "HCT245 near the CPU" was a good guess, but it was actually the 'F244s on the other side of the CPU.

U29 is a 'F244
SIMM A9 (pin 18) -- U29 1Y2 (pin 16)
the matching input to that bus transceiver output is 1A2 (pin 4)
U29 1A2 (pin 4) -- M1217 (pin 26)

SIMM A10 (pin 19) -- U29 1Y1 (pin 18)
that corresponds to 1A1 (pin 2)
U29 1A1 (pin 2) -- M1217 (pin 25)

So A10 is connected "the same way" as A9 and I would expect it to work.

I was curious as to the enable lines of the 'F244 bus transceiver. The relevant one is 1/OE (pin 1). It is connected to ground, which means this transceiver is always enabled. Since it's a non-inverting transceiver, I suspect it is simply there to provide enough current to drive the SIMMs.

Now I should be able to use my oscilloscope to see if the signal is wiggling on both sides of that transceiver on those pins. Luckily it is a fairly large SMD package I have some hope of desoldering and replacing if it comes to it.

Since the machine doesn't boot with 4MB SIMMs installed, should I simply monitor those lines at power up? What should I expect to see? I guess I can look at the signal path for the 1MB address line SIMM A9 (U29 1Y2 & 1A2) with 1MB SIMMs installed and working to get an understanding.

Also, I still haven't watched those videos yet. I stumbled across this post and I think it is way easier for me to understand than slogging through videos. Videos are always way too slow and tedious (they get paid more if they're a certain length / longer. I get it.)

At least as that post describes it, it isn't that complicated of an issue. Using my oscilloscope, I should be able to see if this chipset is only refreshing the first 10 bits of rows.

It also explains why some of the 2 chip 4MB SIMMs I bought have CLEARLY had their parity chips inexpertly desoldered. MemoryMasters (who I bought them from) is stripping 1-bit parity DRAMs off to make 9 chip SIMMs and selling those at a mark up. I expected it was something like that. I guess I should be glad they're not just throwing away the rest of the SIMM after stripping off the parity chip.

Now I wonder if there is a way to add a little glue logic with some fly wires and dead bug style to get that wider refresh. This week has been too long to think this problem through at nearly midnight on Friday.

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.

A 'permanently installed' PC104 card on the bottom could very neatly hold anything you want, and not at all compromise the utility of that connector; it's designed to stack.

forward thinking would have your bodges lead to IDC style terminal pins on the card, so you can in fact detach it later if you need to.

The header would be an excellent source to get voltage and ground lines from, as well.