A slightly better look at how the motherboard -just- fits here. I have a blank IO shield I'll put a hole in for the keyboard, or I might not and get an ISA slot PS2 connector and wire that onto the board instead.

Today Windows 95 goes on. It's, uh, not exactly spritely but I hope it'll speed up a wee bit with the 486 when I can get it on.

Next thing I was looking at, cache. The board has 64k, but since I have 32Meg on board I reckon 128k is a better start.

I found what I think might be the right chips, the ISSI IS61C256AH. What I can't find quite as easily are the tag and dirty bit chips. I'm going to have to look at that a bit more, unless anyone has any ideas? One needs to be a 16k*1 and the other two 16k*4. 🤔

ChrisXF wrote on 2023-07-13, 17:26:

Next thing I was looking at, cache. The board has 64k, but since I have 32Meg on board I reckon 128k is a better start.

I found what I think might be the right chips, the ISSI IS61C256AH. What I can't find quite as easily are the tag and dirty bit chips. I'm going to have to look at that a bit more, unless anyone has any ideas? One needs to be a 16k*1 and the other two 16k*4. 🤔

Chances are good, that you don't need to replace those 3 chips. What is in there now?
Impossible to see on these pictures.
Do you have a manual for the board? Why do you think you need one 16k x 1 and 2 16k x 4?
I guess you have already 3 times 16k x 4. If my guess is correct, than you just need to replace the 8 data SRAMs with 8 times 32kx8.
But yeah, it is an upgrade from 64 kB to 256kB of course.

Thanks for the post! Ah, the reason I think that is because per the manual:

https://stason.org/TULARC/pc/motherboards/D/D … BACK-CACHE.html

Claimed cache config:

That's what it's telling me I need. That said, eh, the quality of documentation of that era was a bit, uh, you know! 🤣

Here's a better close-up of the cache area:

As near as I can tell, those tag and dirty chips are indeed a perfect match for 64k, and too small for 128/256. Does that look correct? Would be nice to be wrong and I only need the others which are a lot easier and cheaper to find!!

From the photo, I've got 8 HM3-65764-5 (this last digit isn't a speed, as I'd expect!?) and three p4c170-20PC.

I have the cache SRAM on order, just need to source the other three physically smaller chips.

ChrisXF wrote on 2023-07-13, 20:37:

Thanks for the post! Ah, the reason I think that is because per the manual: […]
Show full quote

Thanks for the post! Ah, the reason I think that is because per the manual:

https://stason.org/TULARC/pc/motherboards/D/D … BACK-CACHE.html

Claimed cache config:
Screenshot_20230713_174027_Firefox.jpg

That's what it's telling me I need. That said, eh, the quality of documentation of that era was a bit, uh, you know! 🤣

Here's a better close-up of the cache area:

20230516_165628~2.jpg

As near as I can tell, those tag and dirty chips are indeed a perfect match for 64k, and too small for 128/256. Does that look correct? Would be nice to be wrong and I only need the others which are a lot easier and cheaper to find!!

From the photo, I've got 8 HM3-65764-5 (this last digit isn't a speed, as I'd expect!?) and three p4c170-20PC.

I have the cache SRAM on order, just need to source the other three physically smaller chips.

OK, so then we can assume the description is correct, because the p4c170-20pc are indeed 4k x 4 chips.
I also don't know what the -5 in HM3-65764-5 means. I guess 25ns, but not sure.

OK, so 16kx4 chips I can easily find on ebay. But not 16kx1 (22 dip), that one seems to be problematic. Maybe you can find one. Not sure if it is worth the effort.

Yeah, that's the one particularly that'll be a real pain in the neck to acquire I think. I'll keep an eye, but it might be a non-starter.

ChrisXF wrote on 2023-07-13, 21:03:

Yeah, that's the one particularly that'll be a real pain in the neck to acquire I think. I'll keep an eye, but it might be a non-starter.

Exactly, that will be high effort, better keep the existing 64kiloByte.
I found this:
http://bitsavers.trailing-edge.com/components … M_Data_Book.pdf
There was no 16kx1 with 22 pins from Micron, only with 20pins.

Any chance that manual could be misprinted? Is 16k by 1 even large enough for the dirty bits that 128/256k would require?

Just my luck, 🤣

CoffeeOne wrote on 2023-07-13, 21:01:

OK, so 16kx4 chips I can easily find on ebay. But not 16kx1 (22 dip), that one seems to be problematic. Maybe you can find one. Not sure if it is worth the effort.

I severely doubt that 16k x 1 is correct. If the socket is wired up to accept a 4k x 4 DIP22 SRAM for 32k and 64K (it will only use a single bit in these configuration), it is more likely to accept the common 16k x 4 chips, than a special DIP22 16k x 1 chip.

ChrisXF wrote on 2023-07-13, 21:22:

Any chance that manual could be misprinted? Is 16k by 1 even large enough for the dirty bits that 128/256k would require?

16k x 1 is enough. You require only a single dirty bit per cache line. And 256K consists of 16k cache lines. With the 4K x 4 chips, only one of the 4 bits per line is used. Possibly there were 16k x 1 chips for a short time that were made from defective 16k x 4 chips by just connecting one bit instead all four, and they were cheaper than fully working 16k x 4 chips.

ChrisXF wrote on 2023-07-13, 21:22:

Any chance that manual could be misprinted? Is 16k by 1 even large enough for the dirty bits that 128/256k would require?

Just my luck, 🤣

I just want to edit my post .... 😁

Not sure if 16k x 1 is large enough.

But anyway it is suspicious, 64kB uses 3 times the same time 4kx4, and 256kB 2 times 16kx4 and one time 16kx1? So variant 1 an x4 organization for the dirty ram and variant 2 x1 organization for the dirty ram. Hmm.
Also I know that several 486-EISA boards (both SIS and Intel chipsets) have a cache configuration with 8 times 32kx8 and 3 times 16kx4. I have such a board by the way.
So when you feel brave, you could just try the 3 times 16kx4 variant. But maybe make it dependant on the price you have to pay. Don't know 😉

EDIT: I saw mkarcher's reply too late 😁

Hmmmm, the plot thickens!

Thanks for the help with the maths there mkarcher, that makes sense: it would be perfectly sized for up to 256 and no more.

I wonder if I should set the jumpers and buzz out the socket to see what's connected for that last weird 16*1 chip? That might show something, or actually it might not, I'm not sure 🤣

I do like the idea of just trying the 16*4! That should work OK without any harm coming to the board hey? I mean, it'll just ignore the three other bits?

ChrisXF wrote on 2023-07-13, 22:16:

I wonder if I should set the jumpers and buzz out the socket to see what's connected for that last weird 16*1 chip? That might show something, or actually it might not, I'm not sure 🤣

I do like the idea of just trying the 16*4! That should work OK without any harm coming to the board hey? I mean, it'll just ignore the three other bits?

+5V/GND is on pins 11/22 for all relevant memory chips, so physical damage by trying an incompatible chip is unlikely. You can buzz out whether all address for 16K x 4 chips are interconnected between the address tag and the dirty tag. This should further confirm that the pinout of that socket is compatible with 16K x 4 chips. When you are buzzing anyway, you can also check that none of the four I/O pins of the 16k x 4 pinout is directly connected to +5V or GND. Short circuiting the output pins might degrade a 16k x 4 chip. OTOH, as 4K x 4 (obviously working) and 16K x 4 (not documented) have data pins at the same location, I don't expect any strange stuff to happen at those pins.

Ok, So I found the original databook / datasheets for the opti chipset and sure enough they want a 16*1 chip in that dirty bit socket. I had a look at the socket and I think our idea with a *4 will work, three pins are just without traces 😀

Also, as an experiment I overclocked the 386 to 40mhz with very conservative timings. Nope, sadly not stable with the speed bump... Worth a tickle anyways!

I also observed the PC speaker is completely mute. I'll take a look at that next I have time!

Don't have any advice, but just wanted to say it's always nice to see more 386 builds here! 😁

Grab a cyrix fasmath 40 for it! 😀

ChrisXF wrote on 2023-07-17, 15:24:

I also observed the PC speaker is completely mute. I'll take a look at that next I have time!

Check that the speaker itself has continuity first, or try a different speaker. Speaker coils do break, and checking this first can avoid a lot of mainboard troubleshooting.

Yeah I've tried a loudspeaker and a little piezo speaker: also the switching ground doesn't seem to be doing anything when I hook up a scope to the pins. I gotta pull the board out aof the case again for a proper look!

ChrisXF wrote on 2023-07-17, 21:05:

Yeah I've tried a loudspeaker and a little piezo speaker: also the switching ground doesn't seem to be doing anything when I hook up a scope to the pins. I gotta pull the board out aof the case again for a proper look!

If it's the classic AT circuit, one end of the speaker is tied to 5V directly, and the other end is connected via a small resistor to an NPN transistor pulling that line to ground if required. Even if you got ESD into the speaker header, damage beyond that discrete NPN is very unlikely. If that transistor is integrated into something like the 82c206, replacement will get more difficult. No idea whether the 206 integrates that transistor, though.

Yeah I'm hoping there's some kind of a buffer in between the discrete IC that can be replaced. Or better yet something simple like a dry joint.

I bet it's neither 😂🤣