sheath wrote on 2019-02-18, 22:56:
Thanks, yes it booted back into Windows after swapping in the K6-2+ again and setting voltage to 2.2v. What I'm actually after with this chip is the slower speeds, I was hoping to test older graphics cards' benefiting Pentium 1 era processors without actually putting a P1 or K6 in the system. Should I uninstall/reinstall any Device manager items when I switch out a K6-2 with a plus?
Also, I've been using a Startech (ugh) heatsink/fan and non-expired arctic silver 5 compound. The plus has been running about ten degrees cooler than the regular K6-2, so I think heat isn't the issue. I'll reinstall Ali Integrated driver 2.13 again and see how things go after I do a couple of stress tests. Maybe I'll try setmul instead of CTU next time I go for lowering the multiplier in windows.
Any basic white goo (zinc or aluminum oxide) type paste should work fine for these and the tacky kind has worked well for me (actually some very old Techspray silicon-free stuff, but also the rather nice modern Arctic Silver Ceramique 2, which tested to have nearly identical performance to Arctic Silver 5, potentially kept better in storage, and was definitely cheaper and easier to clean up; I got one of the larger 25g syringes around 8 years ago, I think). The less viscous, oily/creamy zinc oxide pastes are way easier to apply to large heat spreaders and ceramic-top CPUs, though.
One thing to also consider with K6s and K6-2s is many, if not all, have some sort of thermal glue/epoxy or dry, waxy (or hardening over time) thermal material applied to the die and few if any are soldered on (and the tops are aluminum, I think, which complicates solder anyway). Whatever it is, at least some cases are bonded pretty well to the heatspreader and delidding can break the die while lifting (that or people just apply to much pressure and end up bumping the die, but I see lots of examples of delidding gone bad with K6s and K62s).
like this:
https://i.ebayimg.com/images/g/wqcAAOSwCV5caEMh/s-l1600.jpg
Though I can't vouch for the story behind that (compared to forum posts of actual experiences) and for some reason it looks like all the K6-III examples there had their dies come off entirely.
The skinny rectangles should be K6-2s, those larger dies with glue all around the edge are 350 nm K6s, I think, and the ones with asymmetric resistor mounting should be IIIs. (one or two III+ dies in there, too, I think that square die on the far left is a 2+ or 3+ given the resistor configuration and die size/shape) I'm not sure what the K6 250 nm die looks like, but it should be slightly smaller than the K6-2 die.
I snagged an intact, delidded K6-III, and the dimensions look just like the missing-die examples I mentioned, and with glue at the corners like with K6-2s.
In any case, I'd think the die to heatspreader contact might be the main limiting factor for overheating and quality thermal paste might not do much on cases with poorer TIM applications.
I can't recommend delidding though ... and if that glue is epoxy based, I don't think heating it would help reduce the force needed to pry it off. (though the gap between the glue points looks wide enough to get a good sided prying tool in there without putting too much pressure on the ceramic)
If you did delid a K6-2 or 2+, I'd also still use the heatspreader and just replace the TIM with quality thermal compound. (given the intermittent and random issues with your K6-2+, heat might not be a factor anyway, or not a consistent factor at least)
Yep, it's still a no go. After running 3dmark 2000 and 2001 and Dethkarz all night the system just started crashing randomly. I left it off all night and this morning it returned to the same mix of Windows Protection errors after it seemed like the video output crashed, tons of garbled ASCII characters and flashing symbols over a quarter of the screen. I popped back in my K6-2 500 AFX and everything is back to normal. Maybe I just got what I paid for with this plus and it is just unstable.
I've had that garbled BIOS before, but somtimes it happens on the first post after a crash/unstable configuration, but not after another hard reset. (plus I've had odd cases of unstable configurations that sometimes post and sometimes won't even when cold, though none of those will boot to windows ... though this might have only happened with 6x86 or MII overclock attempts, not sure)
With my (Rev 1.004 I think, maybe 003) P5A-B, I've had relatively little luck with overclocking 2+ and 3+ chips to 600 MHz and also not great luck with >100 MHz or even 100 MHz when set to CL2 timing (which should work for 7 ns PC-133 RAM, but doesn't seem to be happy, especially at the base 3.5 VIO) and for some reason 110 MHz is sometimes more stable than 105 MHz.
My K6-3+ 400 (1.6V) and K6-2 550 (2.0V) both max out at about 570-575 MHz, and happier at lower FSB (95x6 or 105x5.5) and raising the voltage over 2.0 didn't help a whole lot if at all. (and refused to post if I tried 2.4V or so, and/or gave me motherboard beeping error codes of some sort that I didn't work out ... maybe complaining about no CPU or no RAM inserted)
I've had some similar problems with K6-2s and K6s in my boards, where upping the voltage causes them to fail to post, and the threshold depends on the individual example. I've even overclocked some on stock and reduced voltage with good results, but the same chip fails to post when overvolted at the same multiplier and bus speeds. (usually over 2.6V, but I think one 250 nm K6 classic wouldn't even do 2.5) Though I do have one K6-2/500 that isn't happy below 2.3V, so these CPUs seem weird and not related to heat.
On that note, I've had very good luck undervolting some K6-2s and I think all the 250 nm 2.2V K6 classics I've tried in the P5A along with most Pentium MMXs, especially 233s, and (maybe surprisingly) a lot of Cyrix and IBM 6x86Ls and MX/MII chips (even some of the single-rail 6x86s, though I don't think the chipset or RAM itself likes going much below 3.0V, so the latter case is more limited) and the cyrix chips all seem to overclock to some extent and some exceptionally well, and some just surprising for overclocking at all. (including 3.5V rated parts, both 650 and 350 nm, going by CPUshack's identification, seem to do at least one speed grade higher, and most PR166s will turn into PR200s at 3.5V 2x75) All my 250 nm (NS and IBM both) Cyrix chips seem to do fine at 2.2V, sometimes lower, at the stock speed settings (and 250 MHz rated PR-366 and 333 parts seem to do 3x100 at 2.9V or slightly higher and tend to post up to 333 MHz, but don't seem very stable ... so sort of like 350 nm P55C chips) 350 nm 6x86MX chips don't overclock much on stock 2.9V, but do OK maxed out at 3.5 (and I wonder how they'd do at 4.0V like the 350 nm Nx586 ... or AMD's 350 nm x5 in a 486 board on 4.oV)
I'd initially thought the VIO selection jumpers would work for the 3.3V rail of Socket 5 chips, but apparently those jumpers get bypassed in Socket 5 mode (presumably socket-pin-detected by the chipset) and the Vcore jumpers select the CPU+IO voltage together from 2 to 3.5V. (which on mine seems to read about 3.54V in Sandra 99, which seems more consistent than the BIOS voltage readings for some reason)
Given the poor reputation for overclocking 6x86s have, I wonder if my board is unusually happy with them or something odd like that. I'll need to try other boards at some point. And even overclocked (including 3x66 or 3x68/69 MHz 6x86Ls at 3.5V) they don't seem to run all that hot, especially for the not-excellent airflow of my baby AT horizontal/desktop case (though I guess the desktop orientation at least allows better convection around the CPU socket and the P5A-B's socket placement is better inline with the PSU exhaust flow and also quite nice for the typical exhaust fan location of ATX cases, so that's nice on top of not blocking any of the expansion slots, though long AGP cards can block cpu heatsink airflow)
That K6-III that was delidded appears to be an embedded model or just the rare 2.2V rated desktop model. It's definitely not a III+ and the motherboard actually identifies it as a K6-3 (3+ and 2+ usually show up as 'MMX processor' or some such) and appears to actually run stable at 500 MHz all the way down to 2.0V, which is really neat ... and cool. 😀
It doesn't overclock well much above 500 MHz even at higher voltages and does the weird thing where it won't post somewhere above 2.4V (I forget exactly where, maybe 2.6), but it'll probably do something in the 520 MHz range at some setting. (I was pretty happy with the 5x100 2.0V set-up though)
I almost got a lot of 3 K6 series chips back in 2010 or so with a typical K6-III 450 2.4V at something like $20 for the lot I think, but decided to pick up one of those 3+ 400 embedded 1.6V desolders instead and ended up regretting it (it ran OK, did 550, but for the price at the time, not so great and less compatible with other motherboards ... plus I ended up wanting to see how an old K6-III ran), then the IIIs got scarce and expensive, aside from the 333 that would pop up once in a while (and I missed several times) and unreasonably high BIN listings. (I think there's a 2.2V 450 on there at over $100 ... wait no it's $150 along with a 2.4V one at the same price, but at least the 2.2V ones are actually rare and unusual) III+ chips don't seem to be that much more expensive or unusual these days, though 2+s aren't quite as cheap or common. (a relative abundance of 550 2+ chips, though)
Also, the only K6 series chips that have actually done 600 MHz with any kind of stability are some 500 and 550s that tolerate 2.5-2.8V reasonably well. (I don't think I got any to post at 2.9 even at lower speeds, but might not have tried that much, so among other things, can't say the 250 nm IBM/NS process on Cyrix parts was especially 2.9+ volt friendly)
On that note, though I do somewhat suspect they went for 2.9V across the board to just save cost on finer speed/voltage rating and testing and to be compatible with older and less expensive Socket 7 motherboards, also possibly why it took so long to release 100 MHz FSB parts: my P5A-B seems happy with every 6x86MX chip I've tried at 100 MHz, 350 and 250 nm both, at stock voltage and often below, at least for all the 200 MHz or higher rated parts ... some of the 166-188 MHz parts might have also overclocked happily to 2x100, but the 3x66, 2.5x83, 3x75 and higher parts all do 2x100 or 2.5x100 if not better, so all the PR266/300/333/366 parts plus some 233s and maybe 200s)
I've gotten my 6x86L PR200+ to do 2x100, but not without some signs of instability, and similar with some other 6x86s (including 650 nm parts) at 1x100 even under-clocked, so the I/O portion of the chip seems to be a possible limiting factor there, but 2x83 and 3x66 are fine, even 3x68 at 3.5V.
Oh and Sheath, I assume you wanted a 2+ or 3+ due to the ability to use the PowerNow! feature with a software utility to change the multiplier, right? Otherwise, earlier model K6-2s go all the way down to 2x multiplier (so 120 MHz in the P5A-B with the 60 MHz FSB setting), but if you wanted the flexibility to do multiple trials without a reset and jumper change, the + chips are convenient. Though I'm also not sure why you'd need to overclock it if you wanted to use slower speed settings. (6x66 for 400 MHz clocking down to 133 with PowerNow! should be useful with PCI/AGP clocks at stock 33/66 MHz, though 75 MHz could be relevant too, even though AMD never explicitly supported it ... I imagine it was a popular option for running K6-2 300/66 chips and 266s that wouldn't do 100 MHz FSB and/or just on cheaper motherboards and even ones without 4x multiplier settings, though that'd go up to 75x6 for late model K6-2s)
Meanwhile, 6x86s are convenient for dropping down to Pentium 66 or 486 territory with the 1x multiplier, but that's even better on boards that go down to 50 MHz. (and a fun side note: Sandra 99 defined my 6x86s at 60 MHz as PR-79.8, 66-75 MHz as PR-90, and 83 MHz as PR-100 ... I forget what it thought of 1x95 or 1x100, but the BIOS just calls 1x100 PR120 and 60-83 are all ID'd as PR-90)
