I think most motherboards that support the lower voltage should work. Do you mean pure isa/vlb motherboards or do you also consider ones that have both vlb and pci slots?
I have a FIC 486-VIP-IO motherboard that has both vlb and pci slots and officially supports those dx4 cpus.

This may just be part of the answer, but:
The intel Write-Back CPU's are dual purpose, in that they are Write-Through by default. So they are always "accepted" as in what you asked.

The motherboard needs to select WB mode with the WB CPU pin, and have BIOS and cache line support to actually use WB succesfully. And maybe even ditch VLB usage. I never got that far myself.

Thanks people! 😀
Also, that's a good point, regarding the "bus mixture". I think it would be nice to concentrate on VLB/ISA first.
Plus, I've recently read about the FIC 486-VIP-IO mobo. Have you tried it with a DX4-100 WB CPU, and has its write back functionality been successfully "utilised" / tested?

gerwin, that's interesting that you say "And maybe even ditch VLB usage." Do you think that, in order to get a DX4-100 WB to fully work in WB mode, one must avoid using the vesa local bus slots?

Even with more modern 486 pci motherboard, I don't see any difference in performance if cpu is in WB or WT mode and for L2 cache WT mode seems to be faster than WB.
That makes it harder to actually test if WB is turned on except to believe what cpuchk says about the mode.

retro games 100 wrote:

gerwin, that's interesting that you say "And maybe even ditch VLB usage." Do you think that, in order to get a DX4-100 WB to fully work in WB mode, one must avoid using the vesa local bus slots?

I wrote 'Maybe' for a reason there. I am not sure about VLB versus Write-Back at all.
Anonymous coward mentioned that VLB interfacing and Write-Back are trouble. But I can't find the post anymore, so it is not really proper to quote him without source... Feipoa mentioned he got things working on an "Asus VL/I-486SV2GX4" though. That was in this topic, page 4.

(My single 486 system is right were I want it, with 33, 66 or 100MHz core speed, 33MHz FSB, VLB, Write-Through cache and 16MB RAM. Anything more strays from 486 reference that I like it to be.)

My Genoa Turboexpress X4 motherboard is VLB and supports writeback mode for L1 cache. (It doesn't have onboard voltage adjustment, but I use an "upgrade" chip which works in a 5V socket)

I benchmarked DOOM with and without WB, and there was a noticable difference. 4758 realtics with WT, 4122 realtics with WB (at 100MHz). Snooper showed only a small difference, it went from 274 to 282. So it depends on the workload.

PCChips M912 is also a VLB board that supports WB L1.

The Acer/AOpen VI15g also supports L1 WB.

bakemono wrote:

My Genoa Turboexpress X4 motherboard is VLB and supports writeback mode for L1 cache. (It doesn't have onboard voltage adjustment, but I use an "upgrade" chip which works in a 5V socket)

I benchmarked DOOM with and without WB, and there was a noticable difference. 4758 realtics with WT, 4122 realtics with WB (at 100MHz). Snooper showed only a small difference, it went from 274 to 282. So it depends on the workload.

PCChips M912 is also a VLB board that supports WB L1.

I didn't quite understand that. For me I get around 4700 realtics with 486dx 33Mhz cpu and I close to 2000 realtics with amd 486dx4 100Mhz cpu and those slower cpus that get around 4700 realtics shouldn't have WB support... Did I miss something?

Maybe I should clarify. VLB is only a problem on writeback L1 boards if you are trying to use cards that do "busmastering". 99% of the time, this means a busmastering SCSI controller like the Adaptec 2942 (not all SCSI controllers do busmastering). Using a VGA controller on the VL bus on a WB board should be no problem at all, as graphics cards don't do bus mastering.

Ah right. Good to know!

Sadly my 4SA W2 motherboard (VLB + ISA + PCI combo) refuses to do WB with my SK096 DX4-100.

Trying to trick it as a P24D with a different voltage results in super-slow performance (around DX-33 speeds), with WB still disabled.

An actual P24D works in WB mode without problems, so I assume it's BIOS code that's not cooperating.

The same MB happily runs an X5-133 at 160 with WB though, with Tseng VLB graphics card and all ISA slots occupied.

Thanks so much for the awesome comments people, it's really appreciated! I am particularly interested in that Asus board gerwin mentions. I did a search on Vogons, and I have tested one. A Rev 2.1 board, with a DX4-120 WB CPU. The link is:

VGA VLB Tseng STB ET4000 W32P video card, quick test

retro games 100 wrote:

I did a search on Vogons, and I have tested one.

Yeah. Ironic when one searches the internet for an answer, only to end up with an old, forgotten answer or description from oneself. Happened to me a while ago and it was kinda strange.

Baoran wrote:

I didn't quite understand that. For me I get around 4700 realtics with 486dx 33Mhz cpu and I close to 2000 realtics with amd 486dx4 100Mhz cpu and those slower cpus that get around 4700 realtics shouldn't have WB support... Did I miss something?

The problem with DOOM as a benchmark is that it depends on the screen size and which demo you run. I've already forgotten the details myself, but without looking it up again it was close to 40fps at full screen. I only wanted to point out that it got faster with the writeback enabled.

bakemono wrote:

Baoran wrote:

I didn't quite understand that. For me I get around 4700 realtics with 486dx 33Mhz cpu and I close to 2000 realtics with amd 486dx4 100Mhz cpu and those slower cpus that get around 4700 realtics shouldn't have WB support... Did I miss something?

The problem with DOOM as a benchmark is that it depends on the screen size and which demo you run. I've already forgotten the details myself, but without looking it up again it was close to 40fps at full screen. I only wanted to point out that it got faster with the writeback enabled.

I always benchmark it with full screen and high details as it has been set in phil's benchmark pack. I thought it was the standard way to test it.

It depends on the BIOS itself as well or the manufacturer of the motherboard. I have a Packard Bell Pack-Mate 28 Plus that has VLB video integrated, as well as the primary IDE port that its own BIOS never supported WB cache that are in the CPUs itself, until I upgraded my BIOS from the OEM Phoenix BIOS 4.03 to Micro Firmware 4.05.10, which not only does it support AMD and Cyrix (VRM required) processors, including the DX5-133 processor and Pentium OverDrive 63/83 CPUs, it also supports WB cache as well. I haven't had a chance to get a DX4-100 CPU with the WB cache since I only have the DX4-100 OverDrive and just soldered on the 3.3V fixed voltage regulator, 2 22uF 6.3V tantalum SMD caps and a bridge for the 0 Ohm resistance recently.

The boards that I had that supported the WB 100MHz CPU were the PCChips M912 v1.7 (genuine L2 cache) and the Shuttle HOT 433 motherboards.

bakemono wrote:

The problem with DOOM as a benchmark is that it depends on the screen size and which demo you run

Not to mention:

• What hard drive you use (Doom loads game assets almost deterministically at the same spot in each demo, so the faster the drive, the shorter the hiccup)
• How many drivers you have loaded in config.sys and autoexec.bat
• Whether you have smartdrive loaded
• Whether you have audio enabled
• ... a whole heap of other factors ...

I use it as a relative benchmark when I change only one aspect of the system, to see what impact that one change has.

Though having no drivers loaded with a stock default.cfg (which resets your screen size and turns off audio) should be an alright comparison between different systems.

I have tried doom benchmark with both a really slow PIO mode 0 120Mb hard drive and fast 7200rpm udma hard drive and the difference in realtics isn't huge. Less than 5% less realtics with the slower drive.