DaveDoc1984 wrote on 2022-04-20, 15:25:

Hi Everyone, I have recently been putting together a retro build and initially went with a DX2-66 Intel CPU but have since bought a DX4-100 Intel CPU.

The issue I have got, is whilst it works fine when setup with the DX2-66 and is stable, when I set the jumpers for the DX4-100 (Referred to as P24C on the Motherboard Silkscreen), I will be lucky if the machine posts at all...

......

Too much to read for me 😁

Did you actually try the exact same jumper setting that you had with the DX2-66?
If no, then do it.

CoffeeOne wrote on 2022-04-21, 20:08:

Too much to read for me :D […]
Show full quote

DaveDoc1984 wrote on 2022-04-20, 15:25:

Hi Everyone, I have recently been putting together a retro build and initially went with a DX2-66 Intel CPU but have since bought a DX4-100 Intel CPU.

The issue I have got, is whilst it works fine when setup with the DX2-66 and is stable, when I set the jumpers for the DX4-100 (Referred to as P24C on the Motherboard Silkscreen), I will be lucky if the machine posts at all...

......

Too much to read for me 😁

Did you actually try the exact same jumper setting that you had with the DX2-66?
If no, then do it.

Same. Still hideously unstable if it will even POST at all.

DaveDoc1984 wrote on 2022-04-21, 13:58:

I have got 15ns Cache installed for the "External Cache" which I assume is L2 (the BIOS refers to this as External Cache). The chips are W24257AK-15.

So you've got 9 chips the same right? or 5? If an odd number of chips, the odd one is the tag RAM, this is the one that needs to be the fastest for everything to work right... now while they're all marked 15ns they might range in actual performance from 14.9999 to 12.0001 ns.... or might be just the wrong side of 15 if the operator was half asleep that day of testing, or someone left the side door open and it was cold, or some random event like that which over-rated a chip. Now the only place where this would cause much of a problem would be the tag RAM... so... sometimes swapping chips from one in the main banks of cache to the tag RAM (When all are same type of chips) can cure marginal cache problems... because you might have swapped a 14.9999ns performing chip for a much nicer 13.846 or something performing chip which copes much easier with the higher load/duty of being the tag RAM than the chip that was previously installed there was.

(This does work for the determined overclocker also, get the bus up to 40 or 50 and the cache is crapping, then try swapping the tag around with the other chips until you find the "best" one. In current sitch it seems it got the worst one maybe, so any of the others might be an improvement)

edit: some ppl gonna point out that 15ns should be far and away fast enough for 33mhz bus speeds, true, if it's directly connected to the bus, but 486 motherboard designers had a nasty habit of sticking a buffer or other logic of some sort in the way which introduces a several ns delay of it's own to the whole process, particularly if they didn't use the fastest available logic for this, like a 74F series, but only an LS or something. However, this little bugger if you can ID it, is likely soldered in and not so trivial to replace.

9 that are all exactly the same ones. Make, model etc.

I will try swapping the ones in the first bank around and try again with the DX4-100.

the PowerTech MB457 aka Pine PT-2068.1 has 10/19/94-UMC-498GP-2C4X, the Soyo 025R has UMC-498GP-2C4X, an "unknown" board has 12/08/94-UMC-498GP-2C4X,
all has same main bios string as MD-4DUVC and your sting. I would try the Soyo BIOS rev B1 or rev B2 here: https://www.ultimateretro.net/en/motherboards/4670#downloads if it were me
but would also save current BIOS to see if it has an issue. Maybe the early 1994 BIOS do not support DX4 properly....

DaveDoc1984 wrote on 2022-04-21, 21:54:

9 that are all exactly the same ones. Make, model etc.

I will try swapping the ones in the first bank around and try again with the DX4-100.

I would first remove all 9 chips and test if it is stable. Please don't care about the speed at this point (the 100MHz will perform like the DX-66 with the cache).
So when it is 100%stable then, then continue with testing the SRAMs. If not, there is another - maybe even bigger - problem.

Since the FSB is the same between the DX2-66 and DX4-100, I doubt the cache is the problem. Also, you said even the 25MHz bus is unstable with the DX4.
One thing that differs is the voltage (unless you got a 3-volt DX2), and that's why I'm suspecting the VRM (or something more complex related to the signaling voltage, which I guess is the same as the core voltage for 486 cpus).
I would swap all the capacitors near the socket with new ones. Also, try the 4V option if the board provides it. If it were my hardwares, I would also do a quick test at 5 volts, but this is rather risky so I wouldn't recommend it.

Ok... so to update

1 Removed All Cache from the Motherboard - No Difference. Machine still won't post for several attempts and when it does, it is still unstable
2 Changed the Capacitors near the CPU Socket today - No Difference. The exact same behaviour happens as before.

🙁 I am at a loss as to what else I can do, it will work rock solid with the DX2-66 but not the DX4-100. I am close to giving up on the idea of the DX4-100 despite the board showing jumper settings for it (P24C) and that is the CPU that I have got.

Not surprised to hear that cap swapping did nothing. In my experience, it's *rarely* the electrolytic caps behind faults like these unless they're quite obviously bad or in areas prone to running hotter than the rest of the board. Or they're extremely old.

Is it possible your DX4 is slightly marginal? Do you have another board to try it in?
Intel and AMD DX4s are actually subtly different, enough so that they require their own jumper settings on most boards. Perhaps what's stenciled on the board is for a DX4 make *other* than yours?

I have 2 DX4-100's and they both behave exactly the same way.

Unfortunately I don't have another Socket 3 Board to test.

P24C is what is written on the Silk Screen which is an Intel DX4-100 and that is what I have got.

I have similar problems with a completely different board, an MSI-4132G (SiS471 chipset). A DX2 works without issue, but a DX4 Overdrive has all kinds of problems and instabilities.

However, this only happens with the original AMI BIOS. However, if I slap an Award BIOS from an ASUS VL/I-SV2GX4 (completely different board, but same chipset and similar specs) the Overdrive works perfectly and I even get better performance out of memory and cache timings. Now, I have no proof but this seems to point to a BIOS related problem, and could be the same in your case. Maybe the BIOS sets different register values depending on the CPU ID which cause the instabilities? Just speculating here, but it's definitely worth it trying another BIOS as Horun mentions.

I would need to get a compatible BIOS for my board but wouldn't know where to start 🙁 Someone did mention that earlier in the thread.

Can anyone recommend to me a EEPROM Chip that I could program with another Compatible BIOS? I think I have got a Programmer I can use.

If you already have a programmer, just get a W27C512 EEPROM from eBay or somewhere else. These chips are fully compatible with the 27C512 chips that 486 boards commonly use, but can also be erased electrically for re-programming, no need for UV light or anything.

As for finding a compatible BIOS, just look for similar boards (same chipset, similar layout and so on) and try them out, nothing to lose there. As mentioned above, if the BIOS from another board has the same ID string as yours there's good chance that it will work.

@DaveDoc1984
If you would rather try modding the original 4DUV VER 3.1 BIOS, here is a correct copy of this Award 11/17/94-UMC-498GP-2C4X6A31-00 BIOS.
The BIOS dump from Ultimate Retro appears to be a dump from Shadow RAM and has checksum errors. 🙁

If you want me to enable the hidden BIOS options for you and check this BIOS for proper DX4 support, just let me know. I should have it ready in a couple of days.

Jan

Chkcpu wrote on 2022-04-28, 09:29:

@DaveDoc1984 If you would rather try modding the original 4DUV VER 3.1 BIOS, here is a correct copy of this Award 11/17/94-UMC-4 […]
Show full quote

@DaveDoc1984
If you would rather try modding the original 4DUV VER 3.1 BIOS, here is a correct copy of this Award 11/17/94-UMC-498GP-2C4X6A31-00 BIOS.
The BIOS dump from Ultimate Retro appears to be a dump from Shadow RAM and has checksum errors. 🙁

If you want me to enable the hidden BIOS options for you and check this BIOS for proper DX4 support, just let me know. I should have it ready in a couple of days.

Jan

Thanks Jan, if you could check to see if it supports DX4 (which the motherboard I have suggests it should do) that would be great. I will sort myself out with a Compatible EEPROM so I can burn it to a chip to test.

DaveDoc1984 wrote on 2022-04-28, 10:02:

Thanks Jan, if you could check to see if it supports DX4 (which the motherboard I have suggests it should do) that would be great. I will sort myself out with a Compatible EEPROM so I can burn it to a chip to test.

I did a detailed analysis of the 4DUV VER 3.1 BIOS and I couldn’t find any omissions or bugs in the DX4 support. It is all there, just like in many other late 1994 BIOSes.

Because all L2 cache and DRAM timing options in the CHIPSET FEATURES SETUP menu were hidden and operated on automatic only, I restored them so you now have manual control over these settings.
While I was at it, I fixed the Year 2094 bug, the 2GB HDD size display limit bug, and added Am5x86-133 support as well. 😉

Here is a copy of this “patch J.2” BIOS to play with. I hope this helps with your DX4 problems.

Cheers, Jan

I have made some significant progress with this. I changed out the voltage regulator because when I started pressing that down the machine would post consistently with the DX4-100.

After replacing the Voltage Regulator, the machine now posts consistently 1st time every time. However, I am not out of the woods just yet as the machine is still unreliable.

The 3D Bench test no longer has the corrupted S Logo when I run it so that's a very positive sign and usually that runs through without a hitch.

The Doom Demo Test from Phil's Computer Lab sometimes will run through ok but sometimes will hang during the test. Its quite random so I can't even establish a pattern of when it does this.

Windows 95 is much more stable at times than before where often it wouldn't even boot to the Desktop BUT it will also just randomly Freeze.

So I have an issue somewhere still and looking for suggestions as to where I could start? Even when disabling L2 Cache, Windows 95 would still Freeze at random so not convinced its a Cache issue.

Another small update....

When the machine has been on for a while (say 10-15 minutes) and it "warms up".... its rock solid! I can then run the Doom Demo test repeatedly with no crashes and Windows 95 will also boot and remain stable for a considerable length of time. Not sure what on earth is going on with this machine.

Do you have another power supply to try with it?

Speaking of which, I heard my Enermax ATX12V (with all negative rails present no less) whistling earlier. Probably time to retire the Jen Pans in it.