I think the BIOS you seek is available on this page here: http://th2chips.freeservers.com/m915/

EDIT: I also wanted to note that many boards of this era used OTP (one time programmable) PROMS for the BIOS. Just something to keep in mind, since if yours is such a board you'd need to snag and program an EPROM too...

I salvaged several EPROMS from dead socket 7 boards so this isn't a real big deal. Though are they electrically compatible ? My EPROMS chips have 4 more pins than the original PROM but the socket has some extra pins too

I'll try that but I'm not sure if that would work. If I enable the cache in the bios and that it's not populated I suppose that if the bios doesn't fake the bios the board won't start right ?

Did you confirm that the original BIOS will not work properly with respect to L2 when cache is actually installed?

The jumpers on the board are se to 256k of cache and if I enable the cache in the bios, it says "256k cache installed" even though obviously, nothing at all is installed. I don't think it's able to autodetect the cache and enable it, but I might be wrong

Deksor wrote:

The jumpers on the board are se to 256k of cache and if I enable the cache in the bios, it says "256k cache installed" even though obviously, nothing at all is installed. I don't think it's able to autodetect the cache and enable it, but I might be wrong

So does it use the cache if cache is actually installed (which I think is what feipoa is asking)? I mean, if it isn't able to autodetect cache, then it will go by whatever the jumper is set to, even if no cache is actually present in the sockets...

This is the same trick PC Chips did with the M919. I think that is just a silly string they added to the BIOS so that you always think there is real cache installed. On an M919, without the cache module, it says at POST that cache is installed. If you install the cache module, it also says cache is installed. With the cache installed, the cache does work as expected.

I see. Though I didn't do the mod at the moment, like I said, I only removed the fake cache chips. Now I need to salvage the sockets and the chips from another board

You can still buy the DIP sockets new, but you have to piece two together. It is not a big deal and you can simply butt up two DIP-16 sockets (assuming you are needing DIP-32 sockets). Or you can salvage.

After ~2 hours of work of moving sockets to one motherboard to another (I previously removed the fake chips a long time ago), here's the finished product

But the question is ... will it work ?

Well ... here's speedsys's answer :

OH MY GOD ! I never thought it would really work, but apparently it does 😮

Thank you very munch for telling me this ^^

I have finally a 486 board that has works, have VLB and doesn't have flaws anymore (oh maybe one if I want to be picky : the PCB is really thin but if I leave it alone it shouldn't give me any issues)

Well done!

Thin PCBs was just one side-effect of the motherboard manufacturers race to the bottom in an effort to sell them as cheap as possible. It's certainly not unique to your board, and shouldn't pose a problem.

There's just a small flaw to the bios but that's note a huge deal : the write-through/write-back toggle doesn't seem to have any effects unfortunately. I've not tried the write-back setting for the internal cache because I don't own any 486 that supports wb internal cache

Nice work, although I would have used DIP-32 sockets. This way you can upgrade the cache to 512K or 1024K. Remember, the max RAM you can cache when the cache is in write-back mode is 32 MB.

How do you know the write-thru / write-back toggle has no effect? Use CPUMark99 in Windows 9x. Very few programs show much difference with L2 WB/WT, although CPUMark99 is very sensitive to this.

Speedsys made a difference between wt and wb on my Aopen AP43 board and there it didn't change anything, so I don't think that this fonction does anything unfortunately.

I would have put dip-32 sockets if I had some, but I don't ^^

I'm not going to use this board with maxed out settings, I'm going to use my AP43 for that because the SiS chipset seems to be a bit faster than the UMC one. And also, I'm fine with 32 MB of ram, I don't think I'm going to really use more than that anyways

I recall Speedsys showing any speed difference between WB/WT on my UMC-based PC Chips m919 board, but CPUMark99 showed the difference.