VOGONS


First post, by GigAHerZ

User metadata
Rank Oldbie
Rank
Oldbie

Hi!

I have an amazing Socket 3 VLB board Soyo SY-25P:
https://stason.org/TULARC/pc/motherboards/S/S … SY-25M-N-P.html
Need help finding 486 VLB Board information and Manual (Someone's thread with good picture and additional information)

It's a Sis 471 based board: ftp://retronn.de/docs/chipset/SiS%20471.pdf

It has 4xDIP32 + 5xDIP28 sockets for cache. I have 4x IS61C1024 + HA24257AKC for tag, that should give me 512kB of cache.
Now, the problem is, that there is no jumper setting to set it up for 512kB of cache. I've tried 3 jumper selections, each with 3 positions (1-2, 2-3 and open) all combinations, but never got 512kB on my screen. It works well as 256kB total cache with those chips.

Looking at the datasheet, it describes Register 51, where theoretically you could even have 1MB of cache. For both 512kB and 1MB, additional bit is needed to be flipped compared to the lower capacities.
Using modbin, those bits are not changable by software. (I was able to enable proper writeback support a while ago with bios modification. This board otherwise has a bug with writeback)
I assume, that bit is held constantly at 0/low by hardware means.

Is there someone who could help me figure out, what i could maybe do to flip that bit? Where to look, what to test?

Thank you!

"640K ought to be enough for anybody." - And i intend to get every last bit out of it even after loading every damn driver!

Reply 1 of 4, by mpe

User metadata
Rank Oldbie
Rank
Oldbie

Don't have any experience with that particular board.

But most 486 motherboard I've seen had 256kB cache realised as 2 banks of 4x32Kx8 chips in DIP28 sockets + one or two tag/dirty chips. Either one or both of these banks often use 28pin chips in DIP32 socket to allow for upgrade to 512kB (where you use just one bank of 128kx8 chips) or even 1024kB (two banks of f128kx8 chips)

The problem is that the pinout of DIP28 chips is slightly different to those that go to DIP32. They have 2 more address bits, but more importantly the power input moves from pin 28 to pin 32. There is also extra inverted chip enable pin. Moving of jumpers to configure for 128kx8 actually means swapping these pins. Otherwise chips are not powered or have power at CE2 input.

Therefore AFAIK you need to do both software and hardware mod IMHO. Often only the hardware as the board is usually able to find out the cache configuration without changing anything in BIOS. Installing bigger chips and configuring chipset registers isn't enough.

I'd use multimeter to find out the correct jumper positions.

Blog|NexGen 586|S4

Reply 2 of 4, by GigAHerZ

User metadata
Rank Oldbie
Rank
Oldbie

@mpe, as already said, the board works with 4xDIP32 chips. It just never uses more than 256kB.

AFAIK, i need to modify register 51 to start using additional address pin. How should i do it with multimeter?

"640K ought to be enough for anybody." - And i intend to get every last bit out of it even after loading every damn driver!

Reply 3 of 4, by mpe

User metadata
Rank Oldbie
Rank
Oldbie

With the multimeter you can check that:

a) That chips are powered. Should be if they at least somehow works. Do you actually know if the cache actually works (not just being reported during POST)
b) That the extra address bits above A14 are connected between all sockets and to the chipset and so the CE2 input.

You should be able to find out what the jumpers do (where they are connected to).

If the above is confirmed you should also check the tag chip as you need bigger tag/dirty chip for 512k and 1024k compared to 256k.

Blog|NexGen 586|S4

Reply 4 of 4, by GigAHerZ

User metadata
Rank Oldbie
Rank
Oldbie

@mpe,
a) chips are powered and speedsys confirms 256kB of working cache by having a dip of memory speed after 256kB
b) A14 and A15 are all connected between each socket. A16 is not. (This is for 2048 bit chips, right?) On chipset, it's connected to pin 147, 142, 118, 96, 60, 10, 25, 166, 202. (Measured on the U16 socket)

I believe i don't even have a jumper for the first of three bits in register 51, setting the cache size.

Tag chip is fine. 256k is okay for 512kB of cache by datasheet. 256kB would only need 16kB tag, 512kB requires 32kB tag, therefore all is good on tag chip.

"640K ought to be enough for anybody." - And i intend to get every last bit out of it even after loading every damn driver!