Inspect the cache chips, sometimes when mishandled the pins get bent and short, or lift from pads. Sometimes even damp dirt between pins can cause issues so if the mobo is dirty just give it a good bath in clean water (plus maybe some dish washing detergent) with a soft brush. Let it dry for at least a few days in a dry place, there will be water under BGA chips that does not evaporate easily so it needs time.

Deunan wrote on 2022-04-07, 21:05:

Inspect the cache chips, sometimes when mishandled the pins get bent and short, or lift from pads. Sometimes even damp dirt between pins can cause issues so if the mobo is dirty just give it a good bath in clean water (plus maybe some dish washing detergent) with a soft brush. Let it dry for at least a few days in a dry place, there will be water under BGA chips that does not evaporate easily so it needs time.

I've already bathed it after initial testing, like I do all my motherboards, expansion cards, cpus, etc:

A couple minutes in a detergent + distilled water mix, not long enough for stickers to float off
Fine brushes to clean everything, being careful of any paper stickers
Rinse thoroughly in sink/tap water
Rinse for a few minutes in distilled water tub
Blow dry/under chips until no more water visible, with datavac blower
2 days of forced-air dry standing on end with 16in fan underneath blowing upward, box fan on side blowing across

Overkill a bit, but I've never had a board or card (CPUs too) get worse by doing this. Some parts have started working after cleaning. Occasionally you get a part that has a lot of residual flux from the factory though, and that gets all cloudy with the detergent - isopropyl can only get those so clean and most boards have stickers that I don't want lost in an ultrasonic cleaner.

Chip and all traces that I've seen around the board look great, and no pins move/disconnected from pads when pushing on each one with tweezers

That's too bad, but if there's only one cache chip that means that it only has 512k L2? The performance hit to a K6-3+ would be pretty small in that case. I agree with your conclusion that it's likely the cache chip itself that's faulty, which could be replaced with the right tools; but now that it's been discovered that you can turn K6-2+ CPUs into K6-3+ with a relatively minor modification, it seems much less important.

But it all depends on how much that percentage of performance means to you. You could turn it into a project to see if you could not only replace the bad chip, but also add another to have 1024k L2.

Repo Man11 wrote on 2022-04-07, 22:15:

That's too bad, but if there's only one cache chip that means that it only has 512k L2? The performance hit to a K6-3+ would be pretty small in that case. I agree with your conclusion that it's likely the cache chip itself that's faulty, which could be replaced with the right tools; but now that it's been discovered that you can turn K6-2+ CPUs into K6-3+ with a relatively minor modification, it seems much less important.

But it all depends on how much that percentage of performance means to you. You could turn it into a project to see if you could not only replace the bad chip, but also add another to have 1024k L2.

Correct - 512KB of L2 on this board.

I think I have another one of these same boards somewhere, but not sure. There's always the possibility of swapping components (I'd want more practice on lesser-importance pcbs first), but nothing is really narrowed down much yet, other than "disabling external cache allows boot." My only ATX MVP3 board(s) - I still want to find the EPoX MVP3G2 model that I owned decades ago, or the G5...without the $200-400 premium.

I was surprised how well (dos) games still seem to run without any L2 cache - I imagined the hit would be larger. P233MMX quake timedemo at 320x240 still returns a decent 45fps, doom at 85 or so. I just imagine Windows and their apps is where I'd notice the big difference, like the Covington Celerons that only had L1. Games like Age of Empires? I imagine the performance hit would be noticeable with late-game large maps, more than that already was on a 233Mhz cpu. Still far better than the 486-dx4 I started that game with 😉

Chip pins looks fine so it might be faulty after all. BTW you said the mobo works with L2 disabled, but what about timings? Did you try the slowest cache settings possible, still no-go?

Deunan wrote on 2022-04-08, 13:37:

Chip pins looks fine so it might be faulty after all. BTW you said the mobo works with L2 disabled, but what about timings? Did you try the slowest cache settings possible, still no-go?

My only options are "fast" and "fastest", or something like that, and nothing else. It defaults to the slower settings already. Everything in the BIOS appears to default to the slowest/safest settings, so I didn't see there being anything for me to do in there 🙁

Try the "fastest" option then, nothing to loose 😀 Also, there might be some other options there (for example some L2 chips can offer either synchronous or asynchronous operation), so do flip some settings around. Might be something in there that is a default for this mobo model but perhaps not well suited to your particular HW configuration.