Have you tried different ALi AGP drivers?

Hunch try different agp cards

Download central tweaking unit and play with settings

Also use the latest jan bios if he has one

http://web.inter.nl.net/hcc/J.Steunebrink/k6plus.htm

DIY Bios Modding guide Jan Steunebrink k6-2+/3+ 128gb

Disable system bios cachable

Here's why I think it's a HARDWARE issue, not software:
I have a complete retro PC built around the exact same motherboard, same revision number, running a 550MHz K6-2+ CPU. The software is identical between the two machines - in fact I cloned the boot CF from that existing PC for testing on this board. Video card is the same type, and BIOS settings are now verified identical as well. Same IDE to CF adapter, same brand and size CF card.

So the only differences between the setup that works and the one that does not are:
K6-2+ vs K6-2 CPU
The additional peripherals that are part of the complete system but absent from the test setup (floppy drives, optical drive, USB card, SATA card connected to SSD with games on it.)
Full system has 512MB RAM, test system has 256MB.

I did try an nVidia GeForce4 MX 440 card in place of the Voodoo and the instability remains.
Disabling 512KB motherboard cache via BIOS did get me through a 3DMark benchmark but the Unreal timedemo still crashes in six cycles or less. It seems to crash faster once the system has been running for a while, but this is anecdotal at this point and there's a good degree of randomness to it.

Can you swap in the k6-2+ into the unstable system, and see if becomes stable? I haven't checked the board you have is voltage compatible. But this is the next step really.

And you should also swap RAM.

The K6-III+ has less power consumption, it would run more stable if the VRMs turn instrable under heavier loads.

Note there´s one revision of the P5A that has a VRM-bug when you select the lowest voltages like 2V. Not sure if this was 1.04 or 1.06.

Let's not forget the issues with the asus p5a (1.05 and 1.06): K6-2/3+ Success on Asus P5A 1.06 - simple mod

A P5A should be stable with a K6-2 (I had one many years ago, but I think mine was revision 1.04), and the VRM problem would prevent it from working at all (the 2.0V setting would totally disable the VRM). I'd also suggest swapping all of the components with the working system (CPU, RAM, power supply, and video card, even though you tried a different video card already).

majestyk wrote on 2022-10-30, 06:49:

And you should also swap RAM.

The K6-III+ has less power consumption, it would run more stable if the VRMs turn instrable under heavier loads.

Note there´s one revision of the P5A that has a VRM-bug when you select the lowest voltages like 2V. Not sure if this was 1.04 or 1.06.

1.05 and 1.06

Those versions require a mod (my tweakers include it) for k6-+

Lots of good thoughts guys!

1. Although I didn't mention it, I've used many different RAM sticks. That was one of the very first things I messed with.
2. The P5A 1.06 DOES have a VRM bug if you select 2.0v on an unmodified board. You'll basically get 0 volts and no POST.
The finished PC I built with the K6-2+ CPU and the same mobo has the exact CPU socket mods linked by Nemo1985, performed by me. I've also soldered a jumper wire across the VRM IC on that working board to access the lower range of the VRM successfully (2.0v and below.) To run a K6-2+ without that second mod, you'd need to choose 2.1v IIRC - which does work but I'd rather not overdrive the CPU like that.
3. The requirement for these mods in order to run a K6-2+ CPU is why I haven't tried dropping one into the troubled board yet.

I've been sick so I haven't tinkered with this stuff since my last post but my current thoughts are that it's either the northbridge part of the chipset that's somehow just unstable at higher frequencies (I really hope not) or maybe an issue with the voltage regulators driving the CPU itself. Their output SEEMS OK with a digital multimeter but that wouldn't likely register quick voltage fluctuations. I can also compare VRM output more closely with that of the machine that works.

I suppose I can order replacement VRM's and just try swapping them; That may be my next step but feel free to post if y'all can think of anything else in the meantime. I keep watching the thread.

I've had this exact same error with the corrupted 3DMark files. I think it was on a Socket A system and had something to do with unstable AGP drivers. Did you try a PCI graphics card? Does it only crash in graphics applications? What happens if you run a CPU stress test such as prime95?

It doesn't ONLY crash in graphics applications, no. I've had installers fail to unpack, claiming the compressed file was corrupted, for example. Also I'm really not thinking it's drivers for two reasons:
1. The same AGP graphics card will loop graphical time demo's for hours on end, rock-solid, with a Pentium 233, even when overclocked to 100x3.
2. As stated, I have the exact same software configuration running stable on a finished machine with the same GPU and a K6-2+ CPU. I cloned that working machine's boot drive for this project, so in my mind that rules out software entirely.

But to answer your questions, no, I haven't actually tried a PCI graphics card nor Prime95. I can do those things for curiosity sake, but at the moment I've shelved this project because I needed my bench space for a really cool system build. I ordered a nice new beige tower from Evercase UK (very expensive shipping to USA!) and I'm building a 1.4GHz Pentium III Tualatin system with a Voodoo 5 GPU! The ultimate Pentium III combined with the latest and greatest 3dfx card! Well except for the Voodoo 5 6000, but that's a pipe dream for another day. I got an acrylic cooler assembly for my 3dfx card too, which replaces the aging stock fans and gives the card a much more three-dimensional look to it, almost like a more modern dual-slot GPU.

Bad ram?

Unpacking a big cab file will find ram issues memtest misses sometimes. (Driver.cab)

Could you try to test quake benchmark and see if it works?
I've had a Asus p5a (rev 1.04 if i'm right) which was unstable with C6 cpu while it was working fine with others, I thought it was because of the higher voltage and power consumption but in the end I used the board with another cpu without issues.