Did you try the Award BIOS for the Ford Lian 9885AW board?
https://theretroweb.com/motherboards/s/ford-l … 885aw#downloads
This 12/19/95-UMC-881E/886B-2A4X5CF9C-00 BIOS looks compatible with your M919 V3.x board and has support for the Am5x86.

I found two more boards that look promising. I will follow-up with these when I’ve checked their BIOSes.

Cheers, Jan

Hi there, Jan.
VLB video is garbled and memory test stops at 64MB and jumps to 128MB with "memory test fail" with the FordLian BIOS. With 1MB L2 cache installed, PCI video does not work.
Without any cache stick PCI video works, 256MB RAM are detected and it reports 256K cache at the summary screen, updates "ESCD" for a minute. With VLB video it garbles and stops at 64 MB.

The FordLian has no VLB slots, I´m afraid the BIOS doesn´t support it.

majestyk wrote on 2023-09-14, 07:27:

At the moment I´m experimenting with different BIOS files for my M919 V3.3B/F. I replaced the 4 x 30-pin SIMM sockets with 2 x 72-pin sockets, because 72-pin sticks are cheaper and because I would like to have 256MB EDO RAM. It´s basically a 3.4B/F now and I can keep the fake "Write Back" cache chips 😉

Nice job! Congratulations on picking this legend 😀

majestyk wrote on 2023-09-14, 07:27:

The known original AMI BIOSes all seem to have the infamous gap in the cacheable area between 56MB and 64MB.
Plus as soon as a VLB video card is installed I´m limited th 64MB RAM. When there´s more RAM, the screen gets garbeled when memory count passes the 64MB limit.

I've faced the same problem, CTCM reports:
56 Mbyte, Cache Area < Main memory!!
I wasn't aware that this was a known bug. Is it a M919's only thing? Why... 🤔 ?
I own 4 M919s and some seem to be more stable/compatible than others, and I think it depends on the quality of each particular board. Not all were made the same.
Using 64MB (32x2) EDO RAM only works with a BUS of 50MHz or less. Using a BUS of 60MHz results in NO POST with more than 32MB of RAM.

majestyk wrote on 2023-09-14, 07:27:

This board came with just one voltage regulator populated, so I replaced it by two new ones with very similar electrical characteristics.
The transistors operate in parallel mode so as soon as they differ electrically, one of them takes all the load and overheats while the second one idles.
The electrolytics got beefed up a bit to make the 3.3V line more stable locally.

Interesting info. Thanks for sharing. By beefed up what do you mean exactly?

I also own a M921 which performs even better than the M919. It has the same chipset, so maybe its BIOS will be compatible too? The AWARD one is a bit faster on RAM performance.
With a 60MHz BUS, Speedsys reports:
Memory Bandwidth: 182MB/s (AMI) vs. 202MB/s (AWARD)

But this board doesn't have a VLB slot, so I don't know if it will be VLB compatible.

Chkcpu wrote on 2023-09-15, 16:11:

Did you try the Award BIOS for the Ford Lian 9885AW board?
https://theretroweb.com/motherboards/s/ford-l … 885aw#downloads
This 12/19/95-UMC-881E/886B-2A4X5CF9C-00 BIOS looks compatible with your M919 V3.x board and has support for the Am5x86.

This one doesn't have VLB either and Jan said it could work 🤔

- CTCM doesn´t report correctly here, if you run "cachechk" you see the MBs that don´t get cached: 56-64MB. All the rest is o.k.
This probably is some BIOS bug...

- I replaced all the 10µF 25V caps by 47µF 63V Rubycon ZL. The 330µF near the "COAST" slot is now 1000µF and the 10µF for 3.3V left of the VRM is now 470µF to have a better buffer against transients directly at the VRM. This seems to be best practice in such situations. But you never know exactly because it´s an interdependent system.

I think 60MHz or 66 MHz FSB are a bit over top for this chipset. I didn´t try it and I´m totally contempt with my moderate 40MHz x 4. That doesn´t mean it´s not possible or even stable, it´s just not my thing.

majestyk wrote on 2023-09-15, 17:13:

Hi there, Jan. VLB video is garbled and memory test stops at 64MB and jumps to 128MB with "memory test fail" with the FordLian B […]
Show full quote

Hi there, Jan.
VLB video is garbled and memory test stops at 64MB and jumps to 128MB with "memory test fail" with the FordLian BIOS. With 1MB L2 cache installed, PCI video does not work.
Without any cache stick PCI video works, 256MB RAM are detected and it reports 256K cache at the summary screen, updates "ESCD" for a minute. With VLB video it garbles and stops at 64 MB.

The FordLian has no VLB slots, I´m afraid the BIOS doesn´t support it.

Hi majestyk,

Bummer, I expected the FordLian BIOS to work.
I was looking for boards with the same UM8881/8886A/B chipset with integrated IDE, the same UM8663 multi I/O chip, 3 PCI slots, no Dallas/Odin RTC, comparable keyboard controller, and a BIOS with Am5x86 support.

The FordLian ticked all boxes, but indeed has no VLB.
The UM8881 being a PCI chipset, I assumed the VLB was just a hardware add-on directly onto the FSB, so without any chipset and BIOS control. Apparently I was mistaken.

I will keep looking for a better M919 v3.x BIOS, but without the chipset datasheet this will be difficult. 😉

Jan

majestyk wrote on 2023-09-15, 17:51:

- CTCM doesn´t report correctly here, if you run "cachechk" you see the MBs that don´t get cached: 56-64MB. All the rest is o.k.
This probably is some BIOS bug...

Doesn't the message "56 Mbyte, Cache Area < Main memory!!" mean that? I'm using CTCM 5/98 v1.6n
It's surprising that they didn't test it properly with more than 56MB of RAM.

majestyk wrote on 2023-09-15, 17:51:

- I replaced all the 10µF 25V caps by 47µF 63V Rubycon ZL. The 330µF near the "COAST" slot is now 1000µF and the 10µF for 3.3V left of the VRM is now 470µF to have a better buffer against transients directly at the VRM. This seems to be best practice in such situations. But you never know exactly because it´s an interdependent system.

That's very cool! Great info. Maybe I'll implement it.

majestyk wrote on 2023-09-15, 17:51:

I think 60MHz or 66 MHz FSB are a bit over top for this chipset. I didn´t try it and I´m totally contempt with my moderate 40MHz x 4. That doesn´t mean it´s not possible or even stable, it´s just not my thing.

Yes, it's pushing it to its limits. 60MHz could work OK if you are preparing a specific build. 66MHz is probably too much. I had better stability with the M921.

So I tried a couple of other BIOS and I have good news.

M919 09/15/1998 S --- 56MB gap.
M921 11/08/95 ------- 56MB gap.
M921 (SUNNYLAB) ---- With EDO POSTS but NO boot, with FPM it boots, it seems to be OK (even L2 works) but I don't have enough RAM to test it above 56MB. This is the only AWARD bios.
SYL8884PCI --------- No 56MB bug. L2 works too. There might be some problems but I haven't done extensive testing.
M919 05/06/1996 -- With EDO, 50MHz BUS, 64MB no noncacheable Areas found! If you leave ram/cache timings in AUTO then, everything works perfectly.
With a 60MHz BUS there are stability problems. This is the original BIOS that came with this board (V3.4B/F, green PCB).

With 96MB I need to reduce the bus to 40MHz for it to boot. I even tried 128MB EDO and CTCM v1.6n reports L1 no noncacheable areas found 😉

Maybe improving the electrolytics would provide more stability in order to run a more stable 50MHz/60MHz BUS... I'll probably try this in the future.

Either way, it seems like this BIOS is quite good! Hopefully it works well for you! 💻 🏃