I got the Dallas in the other board. First, the 32-bit IDE port issue is specific to the first board. It works fine on this other one. I did not mention previously, the first board insists on putting the parallel port at 278h in automatic mode and claims there is a resource conflict when set manually to 378h. So something is wrong with that one.

I made some progress with the 3-2-2-2 DRAM issue. I modified the new board to 2X clock mode as well. A 74F244 has pin 4 - CLK2OUT, pin 6 - SCLKOUT, pin 16 - CLK2IN so it is properly buffered for 2X clock.

First, booting into Linux and running a small program to switch to it works fine!
Lacking a better memory test, hdparm -T averages 12.88 MB/sec before and 14.39 MB/sec after.

The issue has something to do with executing code from shadow RAM while in 3-2-2-2 mode. But not always. Rebooting with Ctrl-Alt-Del after flipping the bit in Linux causes the BIOS to hang at post code 13h.
EMM386.EXE ROM=F000-FFFF is enough to at least be able to set the 3-2-2-2 bit from DOS. But the common cachechk, speedsys, etc. don't want to run under EMM386 so it's hard to say immediately whether it's effective.

I have tried various forms of this in debug:
cli
mov ax,1010
out 22,al
in al,23
and al,fe
xchg al,ah
out 22,al
xchg al,ah
out 23,al
sti
<some other code>
then: out 80,al / inc ax / jmp to out 80,al
or: mov ax,e41 / int 10 / jmp to mov ax,e41

Both of those work fine until I press a key on the keyboard at which point the system freezes. The IRQ 1 handler must be enough to freeze the system, but whatever BIOS code is in the timer interrupt is not.
<some other code> can be: call f000:ef31 or call c000:48 (both RETFs on my system), neither of which freezes the system.
mov ah,88 / int 15 doesn't freeze it either.

In Linux, dd if=/dev/mem of=bios.bin bs=65536 skip=15 count=1 / dd if=/dev/mem of=vbios.bin bs=32768 skip=24 count=1 don't freeze the system, and produce the same results before or after.

I have messed with the shadow cacheability settings to no avail. And the BIOS is new enough it's impossible to disable shadowing of the system BIOS as it's compressed. That it can get multiple POST codes in on Ctrl-Alt-Del (causing the BIOS to re-execute but from shadow RAM) seems to indicate it's some esoteric but reproducible situation that causes the system to freeze.

Nice effort, but results were unfortunate. Is the thought that the system is too slow at anything but 3-2-2-2 DRAM to be of interest? Maybe 3x60 MHz or 2x66 could still be of interest. What FSB's are available on the board?

This was all tested with the Am5x86? Any change with the i486DX33?

Which Linux flavour do you use on these old 486 systems?

Even at 4-3-3-3, it's a perfectly adequate VLB 486 system, just inferior to the SiS 471. The onboard I/O and PS/2 mouse are nice, although the mouse requires the DOS TSR as the AMIBIOS has the PS/2 mouse bit locked out (and it's WinBIOS, so I haven't figured out how to edit one of these). Someone I talk to on Discord said the RAM numbers looked similar to the OPTi 895. The "move" speed in Speedsys is awful though.

I have tested using: Am5x86, Enhanced Am486DX2-66 (WB), Intel 486DX2-66 (WT), Intel 486DX-33. No difference.

I am using Slackware 7.1. Red Hat 7.0 would be fine also. It needs to be something that still comes with XFree86 3.3.6, and the xf86config for 3.3.6 without jumping through hoops, since 4.x dropped a lot of ISA/VLB video support.

I don't know enough about hardware design, but I suspect here that there is some mistake in the shadow RAM logic, such that in some cases whatever outputs it needs to generate aren't ready in time for the 3-2-2-2 DRAM read cycle. Some edge case like the Cyrix coma bug in a CPU, but in a chipset's logic instead.

Does it have a 60 MHz FSB option? Unfortunately, I cannot uncover the PLL's part number from online photos.

The user manual for the board specifically mentions PS/2 support when set to Enabled in the BIOS. Maybe your BIOS version has PS/2 disabled. There are 2 BIOS options on the retroweb you can test out.

I have only seen one 486 AMI WINBIOS with working PS/2 mouse support, and it was for the HOT-433. However, the newer BIOSes for the HOT-433 had PS/2 disabled. Maybe a similar case for the board you are testing now.

How well does GNOME run in Slackware 7.1 on your 486's? Or do you not mess with a GUI?

feipoa wrote on 2025-06-30, 10:17:

Does it have a 60 MHz FSB option? Unfortunately, I cannot uncover the PLL's part number from online photos.

The clock generator is a 48C55A-61, which seems to go up to 60 MHz (or 100 MHz on 2XCPU, so you could presumably keep using that pin, but move the resistor to put the chipset back into 1X mode, and achieve 66, 80, or 100 MHz bus speed).
The chipset in some places states 50 MHz maximum, but in others, says it can handle up to 80 MHz 1X or 160 MHz 2X. The ISA divider can go up to 20, for example.

feipoa wrote on 2025-06-30, 10:17:

The user manual for the board specifically mentions PS/2 support when set to Enabled in the BIOS. Maybe your BIOS version has PS/2 disabled. There are 2 BIOS options on the retroweb you can test out.

I have only seen one 486 AMI WINBIOS with working PS/2 mouse support, and it was for the HOT-433. However, the newer BIOSes for the HOT-433 had PS/2 disabled. Maybe a similar case for the board you are testing now.

The post string is 41-1000-000000-01100111-071595-CT4041; the 01100111 should mean that PS/2 mouse is locked out using AMIBCP.

feipoa wrote on 2025-06-30, 10:17:

How well does GNOME run in Slackware 7.1 on your 486's? Or do you not mess with a GUI?

GNOME 1.2+gmc is more usable than GNOME 1.4+Nautilus, still, since it's slow and unpolished anyway, I just use fvwm.

To try 50 MHz operation, you have to revert the 2X clock mod. Probably, the BIOS code doesn't account for the doubled clock when setting its initial ISA divider.

I installed JP28 and successfully upgraded to 512K cache (UMC 15ns, plus the original 15ns tag). I put 2x 180 ohm and 1x1000 ohm resistors in series and connected pins 1 and 2 of the VRM using them to raise the voltage from 3.42 to 3.60 V.

At 3x50 MHz operation, cache has to be cut to 3-2-2-2 read & write in order to run Linux.
At 3x50 MHz operation, IDE has to be cut from PIO speed auto to something slower (I tried 0 just to move past it) to avoid massive disk corruption. Reading from a floppy hard locks the system (ISA DMA issue?) and I haven't found a solution. Disabling cache fixes neither of these. This is still with the VLB speed jumpers set to <= 33 MHz

4x40 MHz operation is better. Cache runs at maximum settings and RAM at max except the 4-3-3-3 of course.
cachechk is 7 us/KB L1, 15 us/KB L2, 28 us/KB RAM.
doom -timedemo demo3 (full screen with status panel): 2134 gametics in 1760 realtics. I am using the Diamond Speedstar Pro CL-GD5430 VLB which is not a very fast card apparently.

So JP28 allowed 512k? how about JP29 sitting right next to it? I am currently awaiting some 15ns chips to fill up what remains of my empty slots to get 256k total, though with all the games I intend to play, I don't know that going more will do much, I built this setup for late 80s and early 90s games. Keeping the DX4 100, I might try and snag up an AMD 5x86 133 one of these days.

JuddSandage wrote on 2025-07-03, 03:22:

So JP28 allowed 512k? how about JP29 sitting right next to it? I am currently awaiting some 15ns chips to fill up what remains of my empty slots to get 256k total, though with all the games I intend to play, I don't know that going more will do much, I built this setup for late 80s and early 90s games. Keeping the DX4 100, I might try and snag up an AMD 5x86 133 one of these days.

I posted the details earlier Re: Latest datecode on 486 chipsets but basically, JP28 1-2 and JP29 open would be 512K, and JP28 1-2 & JP29 1-2 would be for 1 meg. There are some other possibilities because it's apparent they wanted to accommodate unusual 32Kx9 tag chips

Because of the two tag RAM chips (8+1 and 10+1 tag support, versus a typical 486 chipset having 7+1 support only), you shouldn't even need to go past 256K for the full 128MB cacheable area, I just wanted to see if it was possible.

Seems like 33 & 40 MHz operation should work out pretty well. That 50 doesn't, isn't that unusual versus plenty of other boards. I just can't definitively point to the board, the chipset, the RAM, the cache, or some other component as the reason.

Here is the PS/2 mouse pinout on mine; the manual is wrong. I still haven't figured out how to change the BIOS yet to fully enable it. I suspect there is an AMIBCP out there for these.

bought a two pack of PS2 adapters on amazon, and they worked fine, and a PS2 mouse on amazon as well, I have mouse access in the BIOS and as long as I have the QT drivers loaded I get mouse support in all the games I tried. if you want to look, its CablesOnline AD-108.

I also have just 64MB of RAM installed, and with what I am going to use this system for, that seems a little overkill.

JuddSandage wrote on 2025-07-03, 04:05:

bought a two pack of PS2 adapters on amazon, and they worked fine, and a PS2 mouse on amazon as well, I have mouse access in the BIOS and as long as I have the QT drivers loaded I get mouse support in all the games I tried. if you want to look, its CablesOnline AD-108.

I also have just 64MB of RAM installed, and with what I am going to use this system for, that seems a little overkill.

I already had PS/2 adapters, but I had to move the wires around. There is no standard for PS/2 mouse headers.
What does your AMIBIOS sign-on string and post string say? Mine says TEMPTRONIC CORP. and 41-1000-000000-01100111-071595-CT4041; maybe there are multiple variations.

Here are cachechk runs comparing this chipset to an SiS 471, both of them with 4x40 MHz. The SiS 471 is able to run on maximum DRAM settings. The DRAM speed (inability to set 3-2-2-2 DRAM burst) hinders the CS4041 but they are otherwise comparable. The cache on CS4041 is just as fast, it's just the DRAM that is slower. Conclusion: I won't be dismantling the SiS 471 system I have set up to replace it with this motherboard, although the onboard I/O would make for a neater case inside.

Well, the string on POST screen is (C) American Megatrends Inc. , 41-1000-000000-00111111-071595-CT4041-0

The DRAM read/write speeds compared to the SiS 471 are quite a bit slower. So unless there's a working 60 MHz FSB, my interest in this board has dwindled. What was the DOOM result between the two systems at 160 MHz?

Are you able to use JuddSandage's BIOS to get your PS/2 mouse working?

I can dump the BIOS if anyone wants it. just let me know.

feipoa wrote on 2025-07-03, 10:00:

The DRAM read/write speeds compared to the SiS 471 are quite a bit slower. So unless there's a working 60 MHz FSB, my interest in this board has dwindled. What was the DOOM result between the two systems at 160 MHz?

Are you able to use JuddSandage's BIOS to get your PS/2 mouse working?

The BIOS is on TheRetroWeb and I tried it. After programming it onto a modern flash chip (SST 39SF010A), after the first boot, the chip is corrupt and it goes into the boot block. I assume it's something related to trying to update its ISAPNP configuration.

On the first boot, the mouse pointer did appear in SETUP, but didn't move. I'm sure there is a way to get it going one way or another, since the TSR works.

I don't have the SiS 471 system set up right now aside from floppy boot, so I need to do that at some point to compare "dooms -timedemo demo3". As that system has a Trio64V+ in it, it had better bet faster obviously.

Right now, at 50 MHz on the CHIPS board, I can't even get dooms to complete with cache off, so I think it may be pointless to consider 60 MHz.

Well, I got my extra 128k of Cache, and after a little bit of issues (I somehow only got one side of a chips pins in, and two other pins bent while putting them in) its working fine now. Also found an eBay listing for an old NEC 4x4 cdrom drive, going to install that as well.