I just discovered that the system will boot fine with the 33MHz CPU if I disable the cache.
I've also received a "cache memory bad, disable cache" (or something like that) message on occasion.
The same cache RAM works fine with the 40MHz CPU though. Puzzled!

naujoks wrote on 2024-06-29, 22:22:

I just discovered that the system will boot fine with the 33MHz CPU if I disable the cache.
I've also received a "cache memory bad, disable cache" (or something like that) message on occasion.
The same cache RAM works fine with the 40MHz CPU though. Puzzled!

Post pics of cpu, maybe it's a fake 33.

Add board model and pics.

thanks

Is it a 33Mhz crystal? Is there a 66Mhz or 80Mhz crystal on the board?

How much external cache is on the board? If you're running a double bank configuration is it possible switch to single bank mode to see if you have any bad chips? Also, did you try turning down the cache timings?

MikeSG asked the right question.
How are you checking that the 386/40 operates at 33MHz and that the motherboard is setup for a 33MHz CPU?

Here's some info on the board:
https://theretroweb.com/motherboards/s/jetway … b4a3-cache#docs

I can't seem to find any other way to jumper the board for different CPU speeds. There is the 33MHz crystal on board. I once tried a 40MHz crystal to get the 40MHz CPU to run at that speed, but the board wouldn't post at all with that crystal.

It's configured for 128KB cache.

The 25/33MHz CPUs both have some weird kind of metallic thermal paste on it, which turned into something like copper. It's proved itself resistent to any attempts to remove it.

naujoks wrote on 2024-07-02, 15:21:

I can't seem to find any other way to jumper the board for different CPU speeds.

There isn't one. This mobo is using that 66MHz as clock (386 CPUs divide the input clock /2, so the actual speed is half of the crystal). And this chipset might not be able to keep up with 40MHz CPU either (if you replace this clock generator with 80MHz variant).
Also, if you are running any CPU tests on it please remove the co-processor if any is installed. I have found this mobo to very picky when it comes to CPU+NPU combinations, long story short it likes to glitch or even hang (sometimes even during BIOS detection phase).

naujoks wrote on 2024-07-02, 15:21:

It's configured for 128KB cache.

The good news is the mobo is actually a decent performer. Not great, this is an OPTi chipset and their memory handling is always slow for some reason, but other than the coprocessor issue it seems to be stable.

naujoks wrote on 2024-07-02, 15:21:

The 25/33MHz CPUs both have some weird kind of metallic thermal paste on it, which turned into something like copper. It's proved itself resistent to any attempts to remove it.

Have you tried soaking it overnight in 99% IPA? This will not damage the CPU but might just soften the glue. If it's a hard, cement type glue I think the only option is to remove it mechanically. It will damage the CPU face and remove the logo but the laser etched writing should survive. Not sure if it's worth the trouble though.

So at present there'a 33MHz crystal on the board. How come the 40Mhz CPU was shown as a 33MHz CPU then and not as 16MHz, if it doubles it? Or am I not understanding this right?
I also still don't undterstand why the 25/33MHz CPUs won't work with the cache enabled.

naujoks wrote on 2024-07-02, 18:07:

So at present there'a 33MHz crystal on the board. How come the 40Mhz CPU was shown as a 33MHz CPU then and not as 16MHz, if it doubles it? Or am I not understanding this right?

Possibly the BIOS has this hard-coded? If you can get the system to boot, use NSSI or HWiNFO to discover the actual CPU clock. NSSI will also run a benchmark so you can easily tell if the result is in line with the reported clock.

naujoks wrote on 2024-07-02, 18:07:

I also still don't undterstand why the 25/33MHz CPUs won't work with the cache enabled.

With 33MHz generator? Or the original one? And what is the exact frequency, 33 or 33.something? Or 32.0000MHz? Could be there is some issue with that particular frequency that upsets the mobo chipset.

naujoks wrote on 2024-07-02, 15:21:

It's configured for 128KB cache.

...not a good idea when in fact only 32KB cache are installed. -> 8K x 8 /8 x4

You keep repeating a 33MHz crystal, this is not what decides the CPU frequency, forget about it. The CPU's 33MHz come from a 33x2MHz crystal, which is the 66MHz crystal. As I wrote earlier, MikeSG asked the right questions that you never answered: If you don't have an 80Mhz crystal, you can't run the CPU at 40MHz.

majestyk wrote on 2024-07-03, 04:16:

...not a good idea when in fact only 32KB cache are installed. -> 8K x 8 /8 x4

Good catch, these chips are indeed 8K x 8

majestyk wrote on 2024-07-03, 04:16:

...not a good idea when in fact only 32KB cache are installed. -> 8K x 8 /8 x4

Good catch. I can' really tell from the photo but that are probably four CY7C185 chips? In which case the mobo needs another 4 of these as 32k is not a supported amount from what I remember. Lowest it can go is 64k.

According to the manual it can do 32kB.
Why would the board work as expected with the DX40 though? I think I did run the cache check program then as well and it didn't throw up any errors.
Identifying those RAM chips is a bit of a dark art for me. How do you guys know from their names what type they are? And what is that mysterious code, like 8K x 8? What does it even mean?

naujoks wrote on 2024-07-03, 10:56:

According to the manual it can do 32kB.

The manual I found only listed 64k and 256k, and I was too lazy to trace connection to all those jumpers. My mobo has 64k anyway so both banks are full and I was not in any hurry to change it for a 33MHz CPU.
You're right though, it can be configured to 32k it seems.

naujoks wrote on 2024-07-03, 10:56:

Why would the board work as expected with the DX40 though? I think I did run the cache check program then as well and it didn't throw up any errors.

What did you run? CACHECHK maybe? Do you remember it actually reporting 128k? I don't think this would work with wrong jumper settings, my best guess is the BIOS actually disabled the cache completly but reported some value being present - for fake chips. It is very interesting, can you repeat that experiment to see what exactly is going on?

naujoks wrote on 2024-07-03, 10:56:

Identifying those RAM chips is a bit of a dark art for me. How do you guys know from their names what type they are? And what is that mysterious code, like 8K x 8? What does it even mean?

Just search for the chip name, find datasheets, figure out the chip capacity. 8K x 8 means 8 kilo times 8 bits, since not all chips are byte wide. In fact this mobo has 3 TAG/WB chips that are 4 bits wide.

I've been playing with the jumpers a bit and eventually got it to work. It's a bit confusing because it's not immediately obvious which is pin 1 on the jumpers.
So now I have the 33MHz CPU running with 32kB cache, and I also successfully tried 128kB. The board is running well.

Do we think that I'm stuck with 33MHz and 40MHz won't be possible?

naujoks wrote on 2024-07-04, 08:39:

Do we think that I'm stuck with 33MHz and 40MHz won't be possible?

I've seen photos of this mobo (it has a few variants or copies, mine is just "386WB" and some have no markings at all) with 80MHz crystal, but as I've said you can't be sure your mobo chipset will handle it. The clock generator should be socketed so it's worth trying if you find a suitable replacement.
Note that cache chips for 40MHz operation should be 20ns, and perhaps even faster like 15ns. It's actually more critical to have the TAG chips be fast enough, and in my corner of the woods it's difficult to find any 4-bit SRAM chips, much less at required speed and decent price. You can always just add waitstates in BIOS settings for the slower chips to work, but then it will greatly reduce performance uplift from going to 40MHz, so it might just not be worth the trouble. And there's the "usual" issues with OPTi chipsets and dual-bank cache...

33MHz is clearly printed on the chipset!

This means:
1. There´s also a 40 MHz version.
2. Whenever (chipset) chips had a MHz rating printed on them back in the days, the chances for overclocking by 20% are next to zero.

You´re not stuck at 33MHz, it´s just the maximum speed of this system.

If you are driving at 320 km/h on the highway in your Porsche 911 you don´t say "I´m stuck at 320!", right?

majestyk wrote on 2024-07-04, 11:48:

Whenever (chipset) chips had a MHz rating printed on them back in the days, the chances for overclocking by 20% are next to zero.

That is somewhat true for parts made in the 80s, but as this board is from 1991 it might fare better.
I have an AMI 386SX board from 1990, and it is stamped with 20MHz (a 25MHz model was also available). I think part of the reason the chipset was stamped as such was simply because a 20MHz 386SX was soldered to the board. I've ran this board at 25MHz without issue. Even at 33MHz it has no issue. Also when this board came out, I don't believe a 386SX-33 had been released yet (the DX had).
If the OP's board came with a 33MHz CPU, it's possible the vendor stamped the chipset at 33MHz just because it was standard procedure.

Forgive me, but where on the board does it say its 33mhz?
Since I tried the DX40 and the 80MHz crystal and it didn't post with that, I suppose there isn't anything else I can do.