Reply 20 of 21, by kool kitty89
I wouldn't underestimate the K6-2's ability to overheat.
With typical PII/PIII era ATX cases, a case fan or in-era ATX power supply with sufficient case airflow (preferably both), and a decent sized heatsink/fan (S370/SS7 oriented ... or late Socket 7 oriented with hot-running K6-233s) you probably won't have problems, but with a more typical AT case, or an ATX case with poor (or no) airflow it could be a problem.
I'm pretty sure this is what happened to several attempts at a K6-2 build in a 1999 vintage low-profile (but still full-height internally) horizontal ATX case Apolloboy and I were using 5-8 years ago. No case fan (original had been gutted) and SFX power supply with relatively poor airflow. (the case originally had a 60 or 75 mm intake and 40 mm exhaust fan, plus a beefier/noisier, slightly deeper than normal 200W SFX form factor power supply with inline rather than side mounted fan)
It worked fine with his HX board an Pentium 133, then even did fine with an overvolted (3.3V board limit) 200 MHz PMMX, though the 40 GB hard drive got pretty toasty. (it was actually that drive that led to the K6-2 build: an attempt to improve drive bay cooling that managed to fry the IDE interface and drive's controller board ... no idea what happened, maybe something weird with the fan wiring piggybacking on the same MOLEX connector or a fluke of the PSU or something)
We tried an FIC VA-503A after that, that seemed OK aside from bulged caps (which he replaced), but it kept throwing errors and randomly crashing when he was using it some time later (after going OK with the initial install), and particularly stability issues with 2 different USB cards we tried in there. (never got a header for onboard USB)
He didn't try the MMX-200 as I recall, but that might have fared better. (also didn't try underclocking and undervolting the CPU, but 300 MHz at 2.0V might have been a good idea)
Thought most of that he'd stuck with a dinky orange anodized aluminum heatsink/fan that looks a bit like some late gen 486 coolers, but we tried a coppermine PIII era 370 cooler as well (with the annoying screwdriver-lever style mounting clip ... also way too tight for socket 7, so needed some bending; I later swapped in a different clip that suits S7 much better) but that didn't totally solve things.
We'd had a 2.1V rated K6-2 400 embedded in there initially at 500 MHz (as I'd had good luck with that on my P5A-B), but eventually tried 4x100 at 2.2V then he gave up and stuck his old Pentium 133 in there and still had problems and he eventually got rid of that board without giving me warning first, so I never got to tinker with it more, but given more recent issues I've run into with my P5A-B in a (probably mid 90s) era horizontal AT case, and weird stability issues that didn't manifest with the case open, I realized how sensitive it was to floor/desk/wall proximity.
OK, on top of that I'd had an AGP S3 Savage card installed that ran rather warm and butted right up to the CPU socket, so that didn't help either. (and the RAM is close enough to get rather toasty off stagnant air as well)
And in this instence, only very, very cool running CPUs managed to maintain stability, plus running at lower bus clocks also helped, and given how warm the RAM was getting that makes sense. (also possibly the chipset itself or board level cache ... they don't generate much heat at all, even if overclocked and with raised I/O voltage, but they're right next to the CPU socket and in-line with heatsink exhaust air)
Once the case is opened, things cool off so fast that it's really hard to tell what the problem is, but it really seems to be heat.
So be warned for K6-2s upwards of 400 MHz (and IIIs and 2/3+ chips) they need decent cooling, but also awareness of cramped conditions and other components overheating.
And in Apolloboy's and my case, we'd also been 'spoiled' by doing a lot of initial testing and builds in a bulky, somewhat gaudy ex-Alienware case that his dad had had (or the same style of case Alienware sometimes used; in any case it had been in a Presscot Pentium4 mid-range gaming rig, probably with the 2.8 GHz CPU overclocked). It had all the case fans removed, but was still huge, had side vent holes, and allowed tons of convection cooling (plus had a fairly hefty PSU fan).
He gave me that case after migrating the old PIII build into the low profile ATX case, so I did all my early P5A-B Socket 7 dabbling in that big thing, which gave me a somewhat skewed view of cooling. (well, that and I had a funky overclocking set-up with one of those S462 thermaltake copper finned fan heatsinks and a 90 mm fan propped up in the case)
And all that is probably exaggerated by likely partially dried up 20+ year old capacitors and elevated temperatures impacting those as well. (though this all on a board that's proven to perform quite well with overclocked CPUs in open-case and test bench set-ups, and also tolerate undervolting a good deal on certain CPUs, especially Pentium MMX and Cyrix MII ... or 250 nm Cyrix chips in general, albeit it's never liked running at 120 MHz FSB, and 105, 110 are usually OK, and 115 with some special cases, but usually not worth the trouble and scandisk wait times)
And the VA-503A is an oddball with the 686A southbridge ... and we might not have used the right drivers (not sure if the standard MVP3 drivers were right for that ... but I'm pretty sure we disabled the legacy audio and onboard USB in the BIOS, and didn't know VIA's legacy audio was actually pretty decent with its OPL3 clone), and he was convinced something else was wrong (like leaky capacitor corrosion damage), but I don't think so. Heat and FSB speed and RAM timing would be the big considerations. (he had the Pentium 133 down to 66 MHz, though so that should've been fine ... maybe dirty SDRAM slots: and minor dust can get worse at high temps: resistance going up with heat and all, plus thermal expansion shifting contacts)
Oh, and Voodoo 3s tend to run very hot, but Voodoo 1s, 2s, and Banshees are usually OK. (I have a 3 that came used with discolored/burnt looking tinned pad area on the back side of the PCB opposite of the GPU, and after running it it became obvious why: you need tons of airflow to cool those dinky stock heatsinks ... I eventually screwed on a little fan and made do with it taking up 2 slots: annoying when in that P5A-B given it's a PCI Voodoo and that board is rather poor for expansion slots ... nice for Voodoo2SLI and full length PCI and ISA cards given the CPU placement, but bad for other things and problematic for most AGP cards in terms of airflow ... so an AGP voodoo 3 would be even worse anyway 😜 )
So anyway, if stability problems do arise, don't rule out heat. (floppy disk x-wing was the main warning for my machine ... that thing is all sorts of finicky ... sensitive to I/O recovery time settings, too)