Ok so playing around with cache486 and writing some code to set things directly..
I can now get speedsys to work with cache l1 on. But it doesn't show in the cache chart. But it is working because the cpu scores get a boost from 10.8 to 11.82.

May have to wrangle a few more options. Work out the best combination.

I can also set the motherboard options but I can't initialise it. might need to look into how to do that.

Tired now..

Okay so I can turn the mobo cache off, set the bits.. Having an issue turning it back on..

Any one familiar with how mobo's initalise and activate cache?

Okay, having some issues with my 55mhz crystal.

But my 125Mhz one works!

Which for a 386 results in 62.5Mhz bus!

She boots!

However I had to lax the main memory timings. Swapping over my mobo cache chips to 12ns now. But not sure if I can get the memory stable, it isn't terrible, but it isn't rock solid either. Not sure if I can get the l1 cache working either. Might just be cache related, so we will see if that makes a difference.

I have a speedsys at 63Mhz with no l1 cache, and a cachecheck at 62.5Mhz. With l1 enabled, most things tend to crash before finishing, so its close, but not close enough..

I think 62.5Mhz is just a tad too far.. For memory, for the CPU. If I could get the CPU fine with more sub zero cooling, I think the ram might just not be stable. Or maybe the chipset. A lot of loose ends.

Looking at getting a 115mhz crystal. Giving a CPU and bus clock of 57Mhz, which I think is probably doable in terms of ram/cache/CPU, even without special sub zero cooling, maybe even without backing the timing off.

Still a 15% performance improvement over 50Mhz across the board. (CPU/Memory/L1 cache /L2 cache). And a 43% improvement over 40Mhz.

another 16mb of 60ns 30 pin 9 chip..

32 Mb ram baby.. At 50Mhz. None of the issues of the 3 chip stuff. But will test it.

In fact I will run all my benchmarks at 20/25/33/40/50 with it.

Update..
I can't get the 115Mhz crystals to work either. No boot.
Or the 87.5Mhz crystals. No boot. Not even bad booting noise.
And I am sure I blew my two epson 110 mhz crystals as well. Which I was having problems with (They aren't in cans so in trying to get them working I think I inserted them incorrectly).

Other crystals work fine. Including the 125Mhz half can size crystal, which i adapted using solder offcuts and a DIP socket. Getto but works.

So Now I am back with 50Mhz and 32Mb ram and 128Kb cache, which is still quite fast and runs with just a heatsink. I am going to write a DOS program which will offer the same functionality as the Chips and Technologies bios. I haven't cracked 256Kb cache, working, but feel it is just a matter of experimenting with some options which I can do much more methodically with a whole PEAK DM chipset program which will read and write all registers of the chipset. But to be honest, 50Mhz, 32mb ram and 128Kb cache with a 486sxl is quite fast. I will make a few recording of it in Jazz Jackrabbit II in windows, Sim city2000 for win95, red alert, Doom, Doom II, Photoshop, 3DSMax etc. It is without a doubt the fastest 386 socketed machine I've ever used.

However, I think I am done with having this machine in parts and will put it all back together. I need the deskspace. Plus coding the software I really need an operational machine in the house. Running benchmarks is time consuming and fairly boring, I would much rather try to get that 256kb working anyway. I also want to write my own DOS benchmark software and I might make a general configuration program for older machines that can also ID stuff. I will revisit the performance aspect with co-pros and clocks and cache when I finish the benchmarking software.. I want it to be pre-configured with various (as in all) xt, 286, 386, 486, Pentium etc specs.

I have a Compaq Deskpro 386sx 16mhz for the low end slow. And a 486 PCI build above this one (which can take my 486sxl pga168 socket cpu/486dx40 and my amdx5 133). So this build really covers everything from a 386DX20 right up to a 486DX33-ish. Still probably my favourite machine though. Its been a blast. If you are desperate to get my attention you can find me over at OCAU.

I was able to score 4682 realtics in DOOM using an SXL2-50 in 50x1 mode on an OPTi 495SX board. I think that works out to 15.95FPS. That is an all time record for me.
From what Feipoa has gathered, 55MHz is just about all you're going to get out of a 5V SXL2 chip. I tried 2x30 (60MHz), but it didn't work out so well.
Although the SXL2 can handle 55MHz, i'm kind of skeptical your board can be stable at that speed. You might have a chance on a 386 board from 1993 or 1994 if it uses a 486 chipset. However, even a lot of 486 chipsets didn't properly support 50MHz.

The timings I used for 50MHz on my 495SX board are:

0 DRAM read wait state
2 DRAM write wait states
3-1-1-1 cache read
1 cache write wait state

This was with an AMIBIOS. I tried the MR-BIOS for 495SX chipsets, but it didn't work well. It constantly reports KB error, and won't let me enter the BIOS setup after the first boot. The keyboard works fine otherwise. Also in the BIOS setup my chipset is reported as 495XLC, which makes me suspect this ROM image is not really for the 495SX.

I tested your BIOS image on my FU340, but my system would not post (black screen).

Also, I tried using eight 64kx4 SRAMs rather than eight 32kx8 SRAMs in attempt to get 256kb cache going. The result was also that the system would not turn on.

*edit* I was checking over the various FU3 type boards on TH99 and going over the confusing cache jumper settings again when I noticed that one of the jumpers had not been set correctly. I now have 256kb cache working with 32kx8 SRAMs. I don't know how I missed this before. I'm going to try some benchmarks at 40 and 50MHz if I can figure out how to adjust all the damn settings for memory timings in the BIOS.

I assume you are using the FU340 BIOS? Did you try booting into Win3.11?

I forgot about this thread for some reason. Phido, so the 87.5 MHz crystals are burning up as well? Are thse the ones from MF Electronics? I remember I also had some 72 MHz crystals that would get pretty darn hot. I wouldn't use them. I wonder if the voltage is wrong or the current flow is too high. Did you ever figure out the over heating issue of the cyrstals?

Yes, I'm using the FU340 BIOS. Officially it's the FU3 V8 BIOS. Windows 3.11 seems to load fine.

BTW, here is the oscillator adapter I was talking about earlier.

Does it heat up?

I can probably do that same function with some 20 or 22 gauge solid core wire.

As far as I know, it doesn't heat up. However, I haven't used this one for extended periods of time. It's not a very useful speed. I believe I got this one for attempting to run a Cyrix 5x86 at 133 using the 3x multiplier. This one would do 127.5MHz, an intermediate step between 120 and 133.

Good to here you have the 256kb of cache working!
What are your cache settings? How many bits are your tag ram?

Yeh don't know what is going on with those crystals, they do get hot. Way hot, hotter than they should even if they are 3.3v ones. I might cut one open and see what is inside. Maybe they need some current limiting resistor or something weird.

My small can to big can adapter is just a dip and solder squished in to bridge it.

50Mhz is totally fine, 62.5 Mhz (125mhz crystal) seems fine for the board, but it is too fast for the L1 cache on chip, as soon as its enabled it crashes. I also needed 12ns tag cache ships to boot at 62.5. Also 60ns memory is a must, I couldn't get anything to work with 70ns. If I could get a ~55mhz crystal working I think that would be ok. The 125mhz crystals are cheap, so if you are having good luck at 50mhz and have some 12ns cache and 60ns memory, try the 62.5. Certainly I don't think the chipset is a huge issue at that speed. ISA bus and cpu are likely to be issues.

I will just keep my eyes open for more 110-115mhz crystals.

I will continue working on my Chips and Technology motherboard software so I can easily adjust all the options, even the undocumented ones.

I don't think I will match the doom record you have, but it will probably get quite close. Still a good combo, 50mhz, 128kb cache and 50mhz with a 486sxl. With SD storage and the accelerated windows video, it is quite a usable setup.

AMD and Cyrix could have probably released a 50mhz and even possibly a 60mhz 386 or 486 (91/92/93). But at that point the pga132 socket was dying, the 486 socket makes more sense with its cache control, made more sense. Even at 50 or 60mhz at pga132 is still not as fast as a proper 486, and its FPU is always going to be behind.

Then the early 486 platform gave way to 32bit buses 486, which then gave way to PCI and Pentium. But 386 did find significant install bases in things like laptops, military applications, networking appliances etc. I am quite suprised that they didn't release a faster 386. A 50Mhz bus would have been pretty easy and many types of board would be abe to accommodate that speed. I guess the obvious answer was if you need something faster go 486, which also came in several embedded formats.

In my experience, 55 MHz is max stable for a 5V 486SXL. I'd have liked to see the SXL2 at 80 MHz mainstream as an upgrade path.

Phido wrote:

Good to here you have the 256kb of cache working! What are your cache settings? How many bits are your tag ram? […]
Show full quote

Good to here you have the 256kb of cache working!
What are your cache settings? How many bits are your tag ram?

....

50Mhz is totally fine, 62.5 Mhz (125mhz crystal) seems fine for the board, but it is too fast for the L1 cache on chip, as soon as its enabled it crashes. I also needed 12ns tag cache ships to boot at 62.5. Also 60ns memory is a must, I couldn't get anything to work with 70ns. If I could get a ~55mhz crystal working I think that would be ok. The 125mhz crystals are cheap, so if you are having good luck at 50mhz and have some 12ns cache and 60ns memory, try the 62.5. Certainly I don't think the chipset is a huge issue at that speed. ISA bus and cpu are likely to be issues.

I have two tag chips installed. They are both 64kx4 12ns. The rest of the cache is 15ns. Even with good 60ns DRAM installed, I was unable to make 50MHz stable on my board even with all the BIOS timings set to their slowest states.
I'm not really sure what good my cache jumper settings would do you for, as we have different motherboards. The BIOS settings are just the defaults. I tried adjusting them to speed things up without success. Nothing seemed to affect the benchmark results very much.

I'll have to do better stability testing with the memory fully loaded when I come back home in the Winter. I'm heading back out East today.

So just confirming your tags are 64kx4? Do you have a part number? I couldn't find anything in that at 12ns..

Once I write up the motherboard program you will be able to see what exactly the bios has set your timings to and what options are available. I didn't have much like with your bios on my board.

A 5v 80Mhz Sxl2 would have been a decent upgrade. Certainly a lot faster than something like the cx486DR2 25/50. AMD had a 5v dx2 80 486 didn't they?

I was kind of disappointed with just 50Mhz, as 62.5Mhz seemed in the realm of the possible. If VLB didn't come a long, I think it would have been likely that a 60 or 66Mhz bus speed 486 would have been released. I spent a lot more to go from a mix of 70ns and 60ns ram to all 60ns to make sure I could get 50+mhz reliable.

I also wonder if perhaps they would have come up with an interim Turbo ISA standard. Say 16Mhz. Or maybe EISA would have become more widely adopted in the consumer market. Having the 386sx also complicated things as 32bit buses don't work with it (although IBM had a 16bit MCA bus).

Any I am glad the AC managed to get 256kb working, and hopefully in the next week or so I will have finished the PEAK motherboard software, and see what I can tweak out of the system.

I believe the part number on my 12ns 64kx4 chips is Cypress CY7C194-12PC. However, different types of tag chips are used depending on the size of the cache. For 32kb cache, a single 8kx8 tag. For 64kb cache, two 16kx4 tags. For 128kb a single 32kx8 tag, and for 256k two 64kx4 tags. My board has the most confusing tag RAM configuration I've ever seen.

The Cache configuration is not easy, and I guess cache was a fairly new thing in PC's and usually configured at the integrator level. I still don't understand how they built my Peak board with no jumpers. I've learnt a lot about cache since starting this project, and I realized I was so ignorant of what was going on and how it works. I have never really had to program anything directly to it, or implement anything directly in hardware.

Thanks for the part no.

It would seem the tag chips are absolutely critical to getting the right cache configuration to work.

I would guess your board has no cache jumpers because the board was only designed to work with one cache size: 64kb. (even though the chipset supports more)

Probably. Which means to get 256kb working I will most likely need to modify the board.. But 64kb and 128 kb both work, so maybe not. Maybe some one very clever programs the PAL's and solved this. Which gives me hope that 256kb could work. Not sure at this stage if its software or a hardware issue. But the fact someone has actually got 256kb working means there is hope.
Now I just need to cram everything back into a matx case and get it all working again..