The BIOS on this machine is awful.
Apparently some tools are needed that maybe long gone? to set a custom harddrive settings? USERHDD.exe

Added is the original BIOS i read off the chip.

517-000672
BIOS VER2.0.3.00
U19

Chip is an Intel P28F010 DIP 32.

Using Claude I have a UserHDD.exe - recreated to test out.
I yanked the battery off for now, before it pops.
Will test more as I ordered a new chip to flash and play with as Claude decompiled everything down and figured it out pretty well.

That´s truly a hyper-proprietary cache module, probably bought by tens of people 😀
But even stuff like this shows up on ebay from time to time....
This would be a really nice machine with the 486DX2 plus 256K cache.

I think this cache-module is piece of unobtainium

Best way to get more performance is use 5V Intel 486DX2 66Mhz (maybe overclock it to 80Mhz on 40MHz bus)
For disk subsystem you can use ISA IDE or SCSI card with its own BIOS extension

majestyk wrote on Yesterday, 06:18:

That´s truly a hyper-proprietary cache module, probably bought by tens of people 😀
But even stuff like this shows up on ebay from time to time....
This would be a really nice machine with the 486DX2 plus 256K cache.

What I came to realize... I found a possible part numbers... 017-0049967 / 017-0049983

I have A 66mhz CPU coming... so hopefully that will work out.

did someone say AT&T Globalyst unobtanium cache? 😀
https://theretroweb.com/motherboards/s/ncr-mo … -3230-a4923-isa
That edge connector looks suspiciously familiar, like VLB being reused for a Cache slot. Hmm where have I seen something like this before? 😀 https://github.com/raszpl/FIC-486-GAC-2-Cache-Module

Wouldnt it be funny if all you need is a right-angle riser? Of course there is a difference of chipsets, so we would first have to verify that pinout matches.

Babasha wrote on Yesterday, 08:31:

I think this cache-module is piece of unobtainium

Best way to get more performance is use 5V Intel 486DX2 66Mhz (maybe overclock it to 80Mhz on 40MHz bus)
For disk subsystem you can use ISA IDE or SCSI card with its own BIOS extension

Yah. I got a 66mhz cpu coming. The DX2ODPR66 I have coming should drop in and work ....
Intel DX4 OverDrive and AMD 5x86-133 on a regulator adapter (Kingston TurboChip, and Evergreen) would probably work too....
Just did see any on eBay for a price I liked....

Also.. interesting enough the 33mhz crystal, is just zip tied to a socket.
While other pictures i've seen is soldered in place. Not that it changes anything... just thought.. huh... ok...

While digging for NCR related stuff, it seems their boards were probably built by 2 or more places. And those places used whatever.
And constantly changing/tweaking the boards. Like mine. It has no 486SX soldered to the board. But every picture I found does.
And some other chips are different, moved around ect.

Interesting thing is...these came with 25 / 33 / 50 mhz setups.... but there are no jumpers on the motherboard? So not sure what that will take.

The BIOS has code for 32/64/128/256kb modules... but messages only for 64 and 256...which from what i found is what they only sold.
Any other size would work but show up as 64kb message. ( bug in their code sorta... but i guess didnt care because only sold 64 / 256 )

Once i get my new P28F010 chip... ill be flashing my modified BIOS and see how things go.

rasz_pl wrote on Yesterday, 14:13:

did someone say AT&T Globalyst unobtanium cache? :-) https://theretroweb.com/motherboards/s/ncr-mo … -3230-a4923-isa That edge c […]
Show full quote

did someone say AT&T Globalyst unobtanium cache? 😀
https://theretroweb.com/motherboards/s/ncr-mo … -3230-a4923-isa
That edge connector looks suspiciously familiar, like VLB being reused for a Cache slot. Hmm where have I seen something like this before? 😀 https://github.com/raszpl/FIC-486-GAC-2-Cache-Module

Wouldnt it be funny if all you need is a right-angle riser? Of course there is a difference of chipsets, so we would first have to verify that pinout matches.

I've pondered the same... very goofy of them to not use a standard socket setup of the day... i guess follow in IBM/Apple ways of trying to be fancy and make more $$ of selling their own upgrades.
I ran across raszpl repo too about this... which got me thinking more.

My AI Slop for this @ https://github.com/jdredd87/ncr-486-3230-bios

Work on the BIOS itself... to address some things...
New USERHDD.exe - since i can't find the original out there...
Some doc stuff on the P1 cache adapter... got a mapping in there to help me here eventually map the pins on the board to the pins on the CPU socket...

as far as slops go your p1-connector-map.html isnt the worst. LLM slopped too hard and thinks you will be building whole cache controller from scratch and is suggesting tracing 486 bus Cache handshake pins 😀 thats idiotic of course, L2 controller is already inside UM82C480. tell LLM :
- chipset type UM82C480 and upload both pdfs from retroweb https://theretroweb.com/chips/2664
- that pesky raszpls repo with cache module as example https://github.com/raszpl/FIC-486-GAC-2-Cache-Module and close enough chipset https://www.bitsavers.org/components/viaTechn … 06MV_199504.pdf so it can figure out what pins are needed and what they do. Pinout wont match from just glancing on mobo pictures (no pin where TAG_Q5 goes on that prticular L2 cache module), but at least grounds and power appear to match so thats a start.
- For example module TAG_WE goes to pin 39 TRWR#, Bank0_OE Bank1_OE are pins 64 65 CROEA# CROEB#, Bank0_A3 Bank1_A3 pins 55 56 CA3A CA3B, VIA chipset can internally remap Bank0_A3 into A2 but your UMC has dedicated pin 54 for CA2 in case you would want to implement single bank (slower) cache - in two bank confix A2 is unused, tagram requires 10 bits meaning you will need two SRAM chips to for it, and so on.
- FIC 486-GAC-2 proprietary cache module - open source reproduction *Completed*
- similar based on previous one https://github.com/wichers/Compaq-Cache-Module-197005-001 and Compaq (197005-001) cache module - open source reproduction

should be enough for a clever clanker to figure all this out

rasz_pl wrote on Today, 04:00:

as far as slops go your p1-connector-map.html isnt the worst. LLM slopped too hard and thinks you will be building whole cache c […]
Show full quote

as far as slops go your p1-connector-map.html isnt the worst. LLM slopped too hard and thinks you will be building whole cache controller from scratch and is suggesting tracing 486 bus Cache handshake pins 😀 thats idiotic of course, L2 controller is already inside UM82C480. tell LLM :
- chipset type UM82C480 and upload both pdfs from retroweb https://theretroweb.com/chips/2664
- that pesky raszpls repo with cache module as example https://github.com/raszpl/FIC-486-GAC-2-Cache-Module and close enough chipset https://www.bitsavers.org/components/viaTechn … 06MV_199504.pdf so it can figure out what pins are needed and what they do. Pinout wont match from just glancing on mobo pictures (no pin where TAG_Q5 goes on that prticular L2 cache module), but at least grounds and power appear to match so thats a start.
- For example module TAG_WE goes to pin 39 TRWR#, Bank0_OE Bank1_OE are pins 64 65 CROEA# CROEB#, Bank0_A3 Bank1_A3 pins 55 56 CA3A CA3B, VIA chipset can internally remap Bank0_A3 into A2 but your UMC has dedicated pin 54 for CA2 in case you would want to implement single bank (slower) cache - in two bank confix A2 is unused, tagram requires 10 bits meaning you will need two SRAM chips to for it, and so on.
- FIC 486-GAC-2 proprietary cache module - open source reproduction *Completed*
- similar based on previous one https://github.com/wichers/Compaq-Cache-Module-197005-001 and Compaq (197005-001) cache module - open source reproduction

should be enough for a clever clanker to figure all this out

hah yah... It was the last thing I gave "the machine" to ask ... and looked at it.. went neat... and went to bed. And yes, i won't be making anything from scratch.
My hands shake to much Ive come to find out on doing small components. Plus after a surgery, i went from 20/20 vision to ... i dunno, but it aint good.. i lost alot of close up vision 🙁

My hope was existing cache module of some sort... and at worst.. just the "adapter" to make it work is all I wanna do then.

Will be interesting .... biggest issue is gonna be me.... burn out. I have 50 things going on in life and my own dev work for a living.. so gotta make sure I can keep plugging away 😀

Waiting on ram / cpu / sound card / ch375 USB card / new eeprom chip to program for BIOS / new battery + battery holder ... before I mess with this again more..

I have a suite of AI Slop from last few weeks, working with my NEC v30 IBM ps2 ... where I used the CH375 card in it for Networking, serial adapters, keyboard/mouse ... USB Camera! and junk.. plus tools and junk for PicoMEM cards.
Such as improved EMS .. and EMS to XMS ... and other fun stuff .. Using my "dosBridge" project... Makes doing stuff stupid simple now. Also a new CH375 dos driver. All in 1 driver with all the optimizations... and tests out a bit better than FeddyV's ... at least on my NEC V30 8086 😀 ....

https://github.com/jdredd87/DOSBridge
https://github.com/jdredd87/CH375USBTools
https://github.com/jdredd87/CH375

One thing the 486 has been tasked with the past few days is this project...

https://github.com/jdredd87/Wolf3D-V30

I wanted an optimized playable Wolf3d on my V30 box... so I have DosBridge talking to both systems. 486 is used to compile via Borland C , run tests on it... then dosBridge deploys it over to the V30 machine, to run tests on and repeat.
The V30 pumps video into a USB 3.0 capture device.. so DosBridge can even record/take pics/ect of actual video output to get a real eye on it... pretty cool I think .. fun to watch all my machines working together to complete a task.
I wanted real machines..not DosBox emulators involved. So to make sure real results work on real hardware 😀

Anyways... enough rambling...

jdredd wrote on Today, 14:58:

My hands shake to much Ive come to find out on doing small components

cool thing about smd is you can just hot plate whole pcb all at once without the need for manually holding individual components
if you decide to try and design/make module for that board I can help with PCB

jdredd wrote on Today, 14:58:

USB Camera!

saw it on YT last week, neat 😀

jdredd wrote on Today, 14:58:

https://github.com/jdredd87/Wolf3D-V30

private repo

jdredd wrote on Today, 14:58:

I wanted an optimized playable Wolf3d on my V30 box....on real hardware

2.7fps seems quite fast already for 8086, around same perf as Doom8088 Doom8088: Doom on a 286 / 8088 might be bottlenecked on amount of video data moved 1 pixel per byte, https://github.com/jhhoward/WolfensteinCGA seems to be almost 50% faster on slow computers maybe thanks to packing pixels
I wouldnt let LLM burn $ leaving it to "innovate" on its own. Id start by instrumenting engine and gathering detailed timing data from main loop to identify hot path and real bottlenecks. For all you know it might be optimizing useless meaningless things one at a time.
This is a good read "Writing Rust code that's faster than state-of-the-art libraries by asking agents to make the code faster" https://minimaxir.com/2026/09/agentic-iteration/ https://news.ycombinator.com/item?id=49803085 :] "just try harder bro" in a loop seems to be unreasonably effective against LLMs 😁
Those might be useful to adapt for your project
AGENTS.md https://gist.github.com/minimaxir/86de3cc8f62 … 337e70924b3411d
prompt.md https://gist.github.com/minimaxir/933cd6354d9 … 45bea13e0940952

jdredd wrote on Today, 14:58:

enough rambling...

its the coolest kind of rambling!

rasz_pl wrote on Today, 16:02:

private repo […]
Show full quote

private repo

jdredd wrote on Today, 14:58:

I wanted an optimized playable Wolf3d on my V30 box....on real hardware

2.7fps seems quite fast already for 8086, around same perf as Doom8088 Doom8088: Doom on a 286 / 8088 might be bottlenecked on amount of video data moved 1 pixel per byte, https://github.com/jhhoward/WolfensteinCGA seems to be almost 50% faster on slow computers maybe thanks to packing pixels
I wouldnt let LLM burn $ leaving it to "innovate" on its own. Id start by instrumenting engine and gathering detailed timing data from main loop to identify hot path and real bottlenecks. For all you know it might be optimizing useless meaningless things one at a time.
This is a good read "Writing Rust code that's faster than state-of-the-art libraries by asking agents to make the code faster" https://minimaxir.com/2026/09/agentic-iteration/ https://news.ycombinator.com/item?id=49803085 :] "just try harder bro" in a loop seems to be unreasonably effective against LLMs 😁
Those might be useful to adapt for your project
AGENTS.md https://gist.github.com/minimaxir/86de3cc8f62 … 337e70924b3411d
prompt.md https://gist.github.com/minimaxir/933cd6354d9 … 45bea13e0940952

jdredd wrote on Today, 14:58:

enough rambling...

its the coolest kind of rambling!

Repo should be public now. I've gotten it to where I can hit 5-7 FPS depending on options... or lack of options. Can play it full deal with no graphic tricks. But also have options...
This was for me mostly.. V30 CPU and use its instructions to help... i have a 1MB VGA card... i have PicoMem2 card... so using it for EMS/XMS memory and Adlib.
For anyone else.. its kinda "use at your own risk and don't expect things to work or perform the same" .... this NCR 486 DX machine while testing hits 50-90+ FPS depending on options 😀

It is all for giggles for me...

With it at "max" options turned on, it was very playable to me... played first few levels.. and kinda enjoyed it ..

I'll have A video of progress tonight maybe...

Progress on the Wolf3D for MY... Nec V30 system...

Using the NCR 486 to compile and test and then deploy to the V30 to test.... using my DosBridge crud.
I know reinventingthe wheel here to some degree, with another Wolf3D on old hardware .... but this was a good test of DosBridge/Claude and using two real machines at once to do a task.
Alot of C .. and alot of ASM changes... and alot of C to ASM changes... ect... And lots of testing it all had to do.
And it went really well! DosBridge had to only use the two smart plugs to power cycle the systems just a few times ( maybe 2 each ) to get them recovered from a hard lock state.

https://www.youtube.com/watch?v=kTTmaO8T79I&feature=youtu.be