Could it have something to do with the "antivirus" option in bios getting turned on, or misbehaving?

Since my board boots from a floppy, here is a photo of my jumper configuration for the off chance something is wrong. My board is configured for a cyrix 5x86c at 2x60 so the fsb/divider/multiplier (black jumper)may differ, but the rest should be the same for an am5x86.

feipoa wrote on 2025-12-29, 02:37:

For anyone who owns a PC Chips M918 v1.2 motherboard, were you able to boot from a 1.44M floppy diskette? When I try to boot t […]
Show full quote

For anyone who owns a PC Chips M918 v1.2 motherboard, were you able to boot from a 1.44M floppy diskette? When I try to boot to the floppy disk, it reads the disk for about 4 seconds, then stops reading. The floppy drive light stays on, but nothing else happens. If I boot from the HDD, I can read the floppy disk within DOS just fine.

I have tried two different floppy cables, two different brands of floppy drives, and two different brands of boot diskettes (brand A was a w95 startup disk and brand B was Memtest 4.20). These boot disks work fine on every other system I've used. The floppy drives work fine on all other motherboard's I tested. I've tried all 4 M918 BIOS revisions.

There's likely something going on with motherboard hardware or BIOS. I tried swapping the UM8663AF chip, but it didn't help. I reflowed the floppy pins and measured the 10K pull-ups on the floppy header.

Full list of floppy drives tested:
Mitsumi D359T5
Mitsumi D63119
Mitsumi D359M3D
Epsom SMD-300 (F401819729)
Panasonic JU-256A798PC
Panasonic JU-257A137P
Panasonic JU-256A198P
Teac FD-235HF

Has anyone else run into this issue? How did you fix it? Thanks!

😀 Usual 486 mobo with write-back CPU cache problem

Link (sorry in russian) https://phantom.sannata.org/viewtopic.php?t=9954&hilit=4045

I got the same on some VIA and OPTI mobos
Try to catch correct BIOS or CPU type jumpers settings 😀

Hi feipoa,

I agree with Babasha that this looks like an L1 cache WB issue, either due to incorrect jumper setting or lack of BIOS support.
You can easily test this by setting the L1 cache WT/WB jumper to the WT position. I believe this is JP29 1-2 closed on the M918.
You can use my CHKCPU tool to confirm the L1 Cache mode.

If L1 cache in WT mode fixes the floppy boot issue, then you know there is an L1 cache coherency problem in WB mode.
In my old Am5x86 upgrade article, you can read more about this issue.
http://www.steunebrink.info/amd5x86.htm

Cheers, Jan

Thanks for your replies. I agree there seems to be some sort of L1 cache coherency problem, however I've yet to pin it down for the Cyrix 5x86. I did find a solution for the Am5x86.

I tried altering JP29 per chkcpu's recollection, but it didn't help. Next, I decided to trace out where CPU pin WB/WT' goes. The WB/WT' pin goes to JP30 pins 14 and 10. If I place a jumper on JP30 7-14, it forces the Am5x86 in to WT mode. This is expected because JP30 pin 7 goes to GND, and pin 14 is WB/WT'. However, JP30 pin 14 also has a 10 K-ohm pull-up to 5 V (on the motherboard). The Am5x86 CPU datasheet states that WB/WT' has a weak pull-down to GND, so how can removing the jumper from pin 7-14 set WB/WT' high (into WB mode)? The internal pull-down resistor must be of substantially higher resistance than 10K.

Although CHKCPU indicated that the Am5x86 was running in L1: Write-Back mode, the floppy still could not be booted to. I tried to force WB/WT' to 5 V to enable WB, similar to how the M919 does it, but this didn't resolve the boot floppy issue. These are the two convenient points on the motherboard I used to force the WB/WT' pin:

Next, I traced out the other jumper header pin which went to WB/WT', namely JP30 pin 10. The mating end of jumper, JP30 pin 3, goes to the CPU pin SRESET. When I set a jumper on JP30 pins 3-10, I am now able to boot from a floppy diskette! Success at last, but why? Note that there aren't any L1: WB/WT options in the BIOS for the Am5x86. The jumper shown in blue with the red arrow is the one which resolved floppy booting with the Am5x86:

Unfortunately, the jumper on JP30 pin 3-10 did not help booting when a write-back Cyrix 5x86 was installed. M918 BIOS versions 6/2/1995 and 7/21/1995 both have a Cyrix L1: WB/WT BIOS option. If I set L1:WT, the M918 w/Cyrix is able to boot from a floppy. For BIOS versions 11/28/1995 and 1/1/1996, there is no user configurable BIOS option for Cyrix L1:WT/WB - it's always on WB mode. CHKCPU indicates write-back mode, or write-through mode if I sent BIOS to WT. This is where I am stuck now. Thinking a QFP lead might have desoldered, I've reflowed the chipsets after adding flux, but didn't help.

Upon further testing, if I boot from HDD w/Cyrix and try to read the floppy, the DOS can display the floppy's directory, but will hang-up when pressing <ENTER> to see page 2 of the directory contents. It was the Am5x86 which didn't hang-up at this point in testing. Nonetheless, I tried all 4 BIOS versions but could not get a Cyrix 5x86 to boot from a floppy diskette with write-back mode enabled. I even tried installing L2 cache, thinking some funny business might be going on because this motherboard often came with fake L2 cache, but installing L2 installed didn't help resolve the issue w/floppy boots. I tried Cyrix CPU revisions S1R3 and S0R5.

For anyone else with an M918 who has floppy-boot success with a Cyrix 5x86, could you share a photo of your jumper configuration? Similarly, if you are also able to floppy-boot with an Am5x86 and do not have JP30 3-10 jumpered, could you share a photo of your jumpers?

Hmm there is definitely something wrong with that motherboard then.

Nor my running board nor my parts board had the blue jumper on and both booted from floppy with both cyrix or amd

Interestingly, I seem to remember BobocoCz mentioning that he was unable to switch L1 (I think) to WB no mater what he did when he used an am5x86 on the 918. It's sounding like there may be some sort of problem with the circuitry responsible.

I read through a whack of conversation you had with Pshipkov on this board at one point re cache etc, yikes, no wonder this one gets a bad reputation and ends up back in the closet time and time again, seems like the only consistent experience is inconsistencies and frustration, I am actually amazed that mine is so-far-so-good this time around

This may seem an odd suggestion - can you get to DOS from a HD boot (Real DOS or W9x restart in DOS)

If yes, try ImageDisk and see what it can read and what diagnostics it reports.

Set floppy type in BIOS to "none".

ImageDisk talks directly to the floppy controler and does not use BIOS to access the drive... Might at least provide a bit more information/insight.

bertrammatrix wrote on 2025-12-29, 16:37:

Interestingly, I seem to remember BobocoCz mentioning that he was unable to switch L1 (I think) to WB no mater what he did when he used an am5x86 on the 918. It's sounding like there may be some sort of problem with the circuitry responsible.

I don't think BobocoCz shared an image of his jumpers. Maybe he had a jumper on JP30 7-14, which forces WT mode.

bertrammatrix wrote on 2025-12-29, 16:37:

I read through a whack of conversation you had with Pshipkov on this board at one point re cache etc, yikes, no wonder this one gets a bad reputation and ends up back in the closet time and time again, seems like the only consistent experience is inconsistencies and frustration, I am actually amazed that mine is so-far-so-good this time around

Try running it in Windows 9x with a Cyrix 5x86 and some late 90's software. I'm not sure how much more patience I have with this motherboard. I had wanted to replicate BobocoCz's results with the Cyrix at 150 MHz, but if L1:WB isn't working well, then I see little purpose. However, you say he couldn't use L1:WB, so I'm not sure how meaningful this replication will be.

DaveDDS wrote on 2025-12-29, 16:58:

This may seem an odd suggestion - can you get to DOS from a HD boot (Real DOS or W9x restart in DOS) […]
Show full quote

This may seem an odd suggestion - can you get to DOS from a HD boot (Real DOS or W9x restart in DOS)

If yes, try ImageDisk and see what it can read and what diagnostics it reports.

Set floppy type in BIOS to "none".

ImageDisk talks directly to the floppy controler and does not use BIOS to access the drive... Might at least provide a bit more information/insight.

Yes, I can boot from the HDD just fine. I've never used ImageDisk before, but I assume it is v1.20 from here: http://dunfield.classiccmp.org/img/index.htm What do you want me to run after I extract the files: TESTFDC.COM ?

Yes, my site will always have the latest (1.20 is right)

Big thing about ImageDisk is it was designed to read any disk the PC hardware could read which includes many non-pc formats, and all PCs are not created equal. Many PCs can't do oddball formats, things like single-density, sectors not 512 bytes, non-pc sectors/track etc.

Testfdc is to figure out what your PC can do, most useful to help pick the best one to use... not so much to detect/diagnosed hatdware problems.

But .. it can't hurt.. the whole idea is to have it accessed/used by non-BIOS - any extra information you can glean may help figure things out.

Chkcpu wrote on 2025-12-29, 10:19:

Hi feipoa, […]
Show full quote

Hi feipoa,

I agree with Babasha that this looks like an L1 cache WB issue, either due to incorrect jumper setting or lack of BIOS support.
You can easily test this by setting the L1 cache WT/WB jumper to the WT position. I believe this is JP29 1-2 closed on the M918.
You can use my CHKCPU tool to confirm the L1 Cache mode.

If L1 cache in WT mode fixes the floppy boot issue, then you know there is an L1 cache coherency problem in WB mode.
In my old Am5x86 upgrade article, you can read more about this issue.
http://www.steunebrink.info/amd5x86.htm

Cheers, Jan

Hi Jan,

I read through the URL you provided. I can confirm the following:

CPU's CACHE# pin goes to the northbridge (M1489) via JP29 2-3

CPU's HITM# pin goes to the northbridge (M1489) via JP31 6-13

CPU's INV pin goes to the northbridge (M1489) via JP31 7-14

DaveDDS wrote on 2025-12-30, 01:49:

Yes, my site will always have the latest (1.20 is right) […]
Show full quote

Yes, my site will always have the latest (1.20 is right)

Big thing about ImageDisk is it was designed to read any disk the PC hardware could read which includes many non-pc formats, and all PCs are not created equal. Many PCs can't do oddball formats, things like single-density, sectors not 512 bytes, non-pc sectors/track etc.

Testfdc is to figure out what your PC can do, most useful to help pick the best one to use... not so much to detect/diagnosed hatdware problems.

But .. it can't hurt.. the whole idea is to have it accessed/used by non-BIOS - any extra information you can glean may help figure things out.

Hi DaveDDS. With the A:\ drive (3.5" 1.44M) set to 'None' in the BIOS and with the FDC still enabled, I ran:

testfdc a: 1.44 /S /X

250K Double-Density PASSED
500K Double-Densith PASSED

I get the same results if I leave A: setup in the BIOS. I tested with an Am5x86 and Cx5x86 installed. I reconfirmed that these CPUs cannot boot to a floppy drive, although they pass Image Disk FDC check.

Small item of note, I am using M918 PCB revison 1.2. Not 1.2A. Most photos online are of 1.2A or sometimes 3.0. The manual for the Amptron M918i - DX9300 state,

AMD 5x86-133 needs the bios to be at least dated 1/1/96 (V1.2A) or 12/28/95 (V3.0A) to be supported. Set the processor as an enhanced AMD DX2 at 3.3V and 33 MHz.

Maybe PCB version 1.2 never got WB working with Cyrix 5x86? Even though the original BIOS version (but not 1/1/96 BIOS) has a L1: WB/WT option.

There are two unidentified jumpers, JP16 and JP17. Manual states "JP16 factory configured", but it has a removable jumper. if I remove this jumper, the BIOS freaks out say CMOS state low and to enter a password. This jumper sets some pin on the southbridge to GND. JP17 sets another pin on the southbridge (M1487) to GND. JP18, when jumpered, sets another pin on the M1487 to GND via a 1K resistor.

It's normal for IMD and it's tools to work the same weather or not the drive is set up in BIOS. (IMD doesn't go through BIOS) - I keep it set to "none" on my ImageDisk system just to keep anything else (incl BIOS) from messing with the drive.

Try running IMD itself, and see if you can read one of your known goot boot disks into a .IMD file - if that works, try writing it to another disk, and if that works, see if that newly written disk boots on other systems.

Of course pay attention to and report to us any errors or other diagnostics that IMD reports during these attempts.

I have decided to give up on this motherboard. It is too particular. I ran some tests at 3x50 MHz with a Cyrix 5x86, and now it lets me boot from the floppy drive, provided that I have the ISA clock at FSB/6, or 8.3 MHz and I use a particular 64 MB EDO stick. However, if I set the FSB back to 33.3 MHz and run ISA at FSB/4=8.3 MHz, it won't boot from the floppy. Perhaps there are some behind-the-scene timings that the BIOS is adjusting.

I could not wait to get this motherboard back into the box. What a recipe for frustration.

I might get back Image Disk software in the future. I have a 5.25" floppy disk from an old Apple that I need to pull some data off of, but I don't have an Apple floppy drive or any old MacOS still running. The disk is a Maxell MD1.D. I have no idea what density or format this disk is. Can Image Disk auto detect the format and write the disk's contents to a folder using MS-DOS system?

Unfortunately... Apple invented their own low-level format (IIRC it was called GCR or something like that) and these disks can't be read by the PC(or any standard) FDC

You will need something like a greaseweasle to read/record the raw fata bit stream

I think Mac disks were more normal...

Btw.. Apple][ didn't even use standard drives.. the interface electronics were different- even to the way the heads were driven.

There were 3rd party boards to replace the one in the drive to convert a std drive to Apple][ compatible.

I think early Macs also used GCR, and later ones used normal MFM... but it really has been years since I looked at Apple drives.

feipoa wrote on 2025-12-30, 14:39:

Hi Jan, […]
Show full quote

Chkcpu wrote on 2025-12-29, 10:19:

Hi feipoa, […]
Show full quote

Hi feipoa,

I agree with Babasha that this looks like an L1 cache WB issue, either due to incorrect jumper setting or lack of BIOS support.
You can easily test this by setting the L1 cache WT/WB jumper to the WT position. I believe this is JP29 1-2 closed on the M918.
You can use my CHKCPU tool to confirm the L1 Cache mode.

If L1 cache in WT mode fixes the floppy boot issue, then you know there is an L1 cache coherency problem in WB mode.
In my old Am5x86 upgrade article, you can read more about this issue.
http://www.steunebrink.info/amd5x86.htm

Cheers, Jan

Hi Jan,

I read through the URL you provided. I can confirm the following:

CPU's CACHE# pin goes to the northbridge (M1489) via JP29 2-3

CPU's HITM# pin goes to the northbridge (M1489) via JP31 6-13

CPU's INV pin goes to the northbridge (M1489) via JP31 7-14

Small item of note, I am using M918 PCB revison 1.2. Not 1.2A. Most photos online are of 1.2A or sometimes 3.0. The manual for the Amptron M918i - DX9300 state,

AMD 5x86-133 needs the bios to be at least dated 1/1/96 (V1.2A) or 12/28/95 (V3.0A) to be supported. Set the processor as an enhanced AMD DX2 at 3.3V and 33 MHz.

Maybe PCB version 1.2 never got WB working with Cyrix 5x86? Even though the original BIOS version (but not 1/1/96 BIOS) has a L1: WB/WT option.

Hi feipoa,

I can feel your frustration with this M918 board and don’t blame you for tossing it aside.
For what it’s worth, the above jumper settings look fine for a L1 cache in WB mode, so the issue can be either the board’s traces, the chipset revision, or the BIOS.

We do have the preliminary M1489_1487 datasheet at TRW, which could help investigate the BIOS’s programming of the chipset registers. I can write an M1489.CFG file for CTCHIPZ to help with this, if you decide to continue to work on this board afterall. 😉

Happy New Year!
Jan

Hi Jan,

Happy New Year - 2026!

A CTCM34 config file would certainly come in handy for the next time I pull out this troublesome motherboard. Even for other FinALi-based boards, an M1489/M1487 CTCHIP config file could be useful.

One shortcoming of the FinALi is the inability to set L2 in Write-Through mode. We normally set L2:WT when overclocking CPU speed and memory timings to their limits. The chipset preliminary mentions:

M1489/M1487 support write-back L2 cache only. In order to increase cacheable region, M1489/M1487 support 8bit TAG (Index 16h D3='1') with "Dirty bit" = '1', i.e. force dirty.

I have another FinALi board I was testing, the Gigabyte GA-5486AL, but the BIOS doesn't have an option for setting L2 as 7+1 or 8+0. The board only contains sockets for single-banked L2 cache (with one TAG socket), so when I have L2:512K installed with 128 MB DRAM, CTCM7 shows only the first 64 MB are cached. So perhaps the system is stuck in 7-bit mode.