Your build is very nicely done and has seriously piqued my interest. I'm looking forward to your next update

You're lucky you didn't burn out the cache. If you have a 66MHz DRx2, I highly suggest you not try 80MHz.
For 50MHz FSB, your best bet is to get a TI486SXL2-50. By default those work with a 1X clock, and the cache will work without a problem.

chrismeyer6 wrote:

Your build is very nicely done and has seriously piqued my interest. I'm looking forward to your next update

I will keep things updated here with major developments. My OCAU thread is my actual build log. I put everything in there, when things arrive, problems, tribulations etc.

I now too have found the extensive chipset manual for this chipset and not just the 2 page summary. Hopefully this will help me resolve some of the mysteries of this setup. Although from casual glance it is still pretty abstract and unusual.

You're lucky you didn't burn out the cache. If you have a 66MHz DRx2, I highly suggest you not try 80MHz.
For 50MHz FSB, your best bet is to get a TI486SXL2-50. By default those work with a 1X clock, and the cache will work without a problem.

I won't be trying 40 or 50mhz bus with the DRx2. I will be trying the 25Mhz and perhaps the 33Mhz bus. I expect 50Mhz is probably doable. I think 66Mhz is probably as far as it will go. I will only be using it with active cooling. From what I can see the clock doubling doesn't dramatically improve performance like it did on the 486's, the caches are too small, and you are still trying to suck through an ISA bus. Which even at higher speeds is going to be pretty slow.

Unfortunately doesn't seem to be many Ti486Sxl2-50's around. I have a PGA168 version, but that isn't particularly helpful.

My actual aim of this build is to explore the difference performance between co-processors and 386 and 486DLC. It isn't to build the fastest PGA132 setup. I find the 386 pretty fascinating and the format is quirky. Which is again why I went for a Chips & Tech board, quirky.

Phido wrote:

Some 64kbx4 chips ordered.. I am curious. As I said there are no jumpers on this board. I guess I just pop them in and the chipset does its thing.

I'm not so sure about that. I've always had to set jumpers on 386 boards with different cache sizes and configurations.

Phido wrote:

Wouldn't mind trying a 87.5Mhz crystal and splitting the difference and getting the best of all worlds hopefully.

I have had trouble finding a properly sized crystal oscillator between 80 and 100 MHz. 90 +-3 MHz would be ideal. Anyone know if these exist? Any links?

These exist, but not in the package you want. They normally come in a square can instead of the rectangular one. I have one on a small adapter PCB that was custom made by the Chinese seller I bought it from.

That sounds unique. Do you have a photo of this PCB?

So I am pretty sure I am the only user here with a 386 now with L1, L2 and L3 Cache.. Or with 196kb of cache.
Cachechk doesn't detect my l1 onchip..

So I am now pretty sure it is down to what ever tag chip you stick in. If it doesn't get what it wants, anything is possible.
If only I had that BIOS. It had registers for this.

I am pretty sure I can get 32mb working on this board just need more 4mb 9 pin ram. It doesn't like mixing 4mb and 1mb sims, again, bios issue.

Quirky.

feipoa wrote:

That sounds unique. Do you have a photo of this PCB?

No, but if you give me about 10 days, I can take one.

Further digging has found.
https://bjheins.wordpress.com/2015/11/24/jcis … therboard-1991/

Which was last updated april 2018. Seems to be the origional source of the bios screen shots.

I sent a message. No reply.

How long ago did you send a message?

I think about a week ago.

Hopefully they are just away doing real life things not just ignoring us. It would be an interesting BIOS to have that is for sure.

But I'm not too disappointed, I have 196kbish of cache things could be worse.

Meanwhile I am going further ahead with this build. Now I have 16Mb ram working nicely, and windows is noticeably faster, 3DSMax and photoshop is quite zippy. I will go for 32Mb 9 chip parity ram. Parity 9 chip ram is i believe the key.
I also purchased a 88mhz crystal, in a dip-14 package, hopefully I can run at that speed with the on chip cache enabled.

Phido wrote:

I also purchased a 88mhz crystal, in a dip-14 package...

From where? I'd like to buy one as well.

Based on the last 3 or 4 people who tried 32MB, I suspect it won't be the cakewalk you expect it to be. I'm pretty sure all of us have already tried 9 chip RAM on our boards. I have many sets of 4MB SIMMs.

I only had 1 or 2 boards out of 6 which didn't like 3-chip.

Anonymous Coward wrote:

Based on the last 3 or 4 people who tried 32MB, I suspect it won't be the cakewalk you expect it to be. I'm pretty sure all of us have already tried 9 chip RAM on our boards. I have many sets of 4MB SIMMs.

Would I ever understate how difficult something was to achieve? Fitting a full baby AT motherboard into an mATX case(particularly that matx case) with active cooling the cpu, and fullsized isa card is like fitting an elephant through the eye of a needle using chopsticks. There are many good reasons why people haven't done it before. Clearance in in some aspects mm or less, which is why cooling is also quite difficult.

Getting more than 128kb of cache and 32mb working on a C&T board is also, apparently, quite tricky. I doubt many people would appreciate the difficulty, I would imagine yourself and Feipoa are some of the few. My experiments make me hopeful, we will see.

From where? I'd like to buy one as well.

https://www.ebay.com.au/itm/MF-ELECTRONICS-M1 … a7e7e%7Ciid%3A1

87.5Mhz but, I think that is close enough.

I have also purchased a 110 Mhz crystal. So I should be able to offer 44, 50, 55 or 62 Mhz bus speeds.
I wanna try for say 50 or 55mhz bus with the 486sxl. Should be quite snappy if it all holds together. If I can use the blk2/5 divider at 11Mhz should still be doable for isa as well. 14mhz might be a bit of a leap.
Not sure my 15ns cache chip will work as tag.

Of course, I might just be building a tower of instability. 33mhz board running at 55mhz, with 32mb ram on a chipset it isn't clear anyone has had 16mb or more stable on, with a cpu and fpu not rated for it, with two unequal cache banks, all crammed into a space too small and the wrong shape and design for AT boards. I haven't even looked at using mclk on the video card either. The on top of that running 3DS Max and Photoshop on win95.

It isn't exactly on spec.

I had a C&T board which worked with 128 KB. It also appeared to work with 32 MB of RAM - that is, it passed HIMEM and Windows 3.11 ran fine. However, if I would open up enough programs in Win3.11 such that more than 16 MB was consumed, the system would crash. 256K also had the first appearance of working, until I tested it.

110 MHz crystal osc. That is nuts. Are you trying to get an SXL2-50 running at 1x55?

I grabbed a pair of 87.5 MHz crystal oscillators.

feipoa wrote:

I had a C&T board which worked with 128 KB. It also appeared to work with 32 MB of RAM - that is, it passed HIMEM and Windows 3.11 ran fine. However, if I would open up enough programs in Win3.11 such that more than 16 MB was consumed, the system would crash. 256K also had the first appearance of working, until I tested it.

I can boot windows 95 desktop with 16mb and 32 mb of 3 chip parity ram as well, same result, load it up and it falls over and is pretty flakey. Don't know why, something differently weird. Even better when I put in 20Mb and it detects it as 8mb.. As I have no jumpers on my board and most of the bios options don't seem to have any effect even if they are there (and most of them arent) this started me down the road of performing random and illogical insertions patterns. I now want to know, for sure what it can do.

But, the board and the bios seem to want things to happen in a certain order, it won't allow you just to do anything, it will only work in specific circumstances. It may very well be broke (either bios or chipset level). I would say my chances are <50% that I can get a full 32mb working stable. Then again, that pretty much describes this whole build. I would be surprised if it was broken at a chipset level, C&T had a lot of success with the NEAT chipset (and eventually the SCAT), the PEAK DM was meant to be the same but newer and more advanced, then Intel started tying them up in legal, and lent on everyone else. Opti ended up licencing c&t implementation and integrating it into their own chipsets (the 206 seems particularly popular for example in 82C495XLC implementations). With there own CPU Intel became very worried about C&T owning the AT clone format, chipset, CPU, either through licence or directly.

110 MHz crystal osc. That is nuts. Are you trying to get an SXL2-50 running at 1x55?

That's the idea. My experiments with 50mhz seem to indicate its pretty stable if you have enough cooling (chipset, ram, CPU) and a CPU that is capable. So why not try for 55? If your already mad, why not go a little further.

I grabbed a pair of 87.5 MHz crystal oscillators.

I hope it proves to be a handy speed. They certainly aren't common crystals. An extra 10% boost over the 40mhz should be pleasant and achievable in many setups.