ISA All-Memory board

[MicroCoreLabs]

I was considering a relatively simple project in which 100% of the memory (both RAM and ROM) was on a single ISA card with a micro controller responding to these address ranges. This way, a non-functional motherboard could possibly be brought back to life if the fault was either with the RAM, ROM, or one of the associated control or data path ICs. The micro has more than enough memory and speed to support tho whole 640KB plus ROM images... It would be similar to my EPROM emulator project, except ISA-slot based and supporting more memory.

The only complication would be how to disable the on-board data buffers for the RAM/ROM... One thought could be to socket the associated motherboard ICs and not populate them when using this card. Another could be to simply cut the buffer direction pin(s)... Something would need to be done to keep the motherboard buffers from conflicting with the All-Memory ISA card... Maybe someone has a better solution...

Could possibly add a few ICs to implement XT-IDE as well.. or connectors for serial/parallel to match the AST six-pack functionality...

Just brainstorming... There are a number of non-functional motherboards out there whose users are unable to debug which may end up scrapped. If there was a board which replaced the functionality of all off the motherboard's memory components, it could bring these systems back to life.

 

[Eudimorphodon]

Not to sound too negative, but are “hard-to-diagnose” memory component errors really that common of a failure mode for XT motherboards? Wholesale memory replacement shims have been pretty useful for machines like the Commodore PET where the correct 1970’s era chips are both really failure prone and hard to find, but the memory used in PC/XT class computers is still reasonably plentiful and the BIOS self test code usually is pretty good at at least hinting where the problem is.

Also it might get really nontrivial to try to disable the onboard memory on any arbitrary clone machine. Some machines are going to have a single buffer with both memory and I/O devices behind it, which means you can’t just cut a direction pin, you’ll have to attack upstream of that. And newer machines might not have discreet circuitry that’s easily modified, you’ll have buffer enable signals emerging from an implementation using PALs or even a full-blown ASIC.

 

[modem7]

Putting aside the issue of catering for the applicable motherboard transceivers/drivers, I see that cards for ROM and RAM are at [here] and [here].

As for catering for the applicable motherboard transceivers/drivers. Even for the IBM 5155 and 5160, where a jumper pad (E1) exists to do that catering for the ROM sockets, soldering is still required (to bridge E1). For the IBM PC family, circuit diagrams exist, and so users can be instructed on how to cater for the applicable motherboard transceivers/drivers via cutting/soldering. Problematic will be clone motherboards for which a circuit diagram has yet to be located.

 

[Malc]

I seem to remember somebody used a lo-tech ram board in a, i think it was a 5150, They disabled the onboard memory / circuitry and the 5150 worked fine using the lo-tech ram board, It was some time back, there should be a post on it somwhere on this forum. But as already mentioned clones / non-obtanium circuit diagrams would be a problem.

 

[Eudimorphodon]

I seem to remember somebody used a lo-tech ram board in a, i think it was a 5150, They disabled the onboard memory / circuitry and the 5150 worked fine using the lo-tech ram board, It was some time back, there should be a post on it somwhere on this forum.

I found two threads about this here and here. The long and short of it is it does seem to be "possible" on the IBM hardware but there's actually more to it than just isolating the memory sockets, you also need to lobotomize the parity checking circuitry.

Obviously the problem is going to get further complicated if you're looking at a "Turbo XT" clone, especially if it's one that uses an 8086 and/or has circuitry to run the expansion bus at a different speed from the CPU. The machine I specifically had in mind when I mentioned the possibility of there being a single buffer pinch point through which both I/O and memory drives the bus was a Tandy 1000-family machine; I actually figured out a way to override built-in ROM on that computer with memory on a card through a combination of a jumper wire to a pin on a (thankfully documented) PAL and abusing a proprietary DMA control line on the bus, so it's definitely a thing you can do if you have the documentation, but in the case of a machine that just has some bad RAM chips I'm not sure it's the optimal way to go.

 

[MicroCoreLabs]

Very interesting. It sounds like the LoTech board and a few motherboard mods can already cover what I was proposing. Well, at least for the 5156/5160... and uncertain if not impossible for clones with consolidated and/or undocumented data paths.

I would suggest making a card which sat between the 8088 and its socket, but I don't think DMA would work at all in this case.

Anyway, always fun to brainstorm and bounce ideas off form members!

 

[pearce_jj]

I seem to remember somebody used a lo-tech ram board in a, i think it was a 5150

The board was tested with a minimum RAM configuration in a 5150, I think 64KB but it should also work with 16KB on the board. As said, getting rid of all the system board RAM will need some modifications.