It shows both the 20-pin ATX connector and the old style AT connector - some motherboards from that time did that (I have one). My current K6-2 system was originally on an AT case with and AT PSU when I got it, and I moved it to an ATX case+PSU... AGP cards worked fine with either PSU.

alexanrs wrote:

It shows both the 20-pin ATX connector and the old style AT connector

Yes. I noted that in my last post.
Although I did edit several times before I was done.
Maybe I distorted time and space and 'phased' us.
.

Oh yes, back when I replied I did not see the updated post. Anyway... most boards with linear regulators had issues supplying the power the AGP slot needs... Though, to be fair, this one uses a fairly big heatsink so its probably not that bad. Anyway, things WERE working when he first installed the card, and now both the V3 and the TNT2 are having issues.

For the OP: IMHO the best thing you can do to troubleshoot things now (except buying a P3 ATX system for testing) is to go for a clean Windows 98 install and avoid installing the AGP drivers. If Windows 98 has built in drivers for that try Windows 95, just to prevent the system from using AGP-specific features. If it works fine that way you can at least rule out that the TNT2 itself is bad.

Might it not be easier to uninstall the AGP drivers/device and clean them out of Windows 98 driver store so it can't reinstate them, then install the graphics card drivers? Would that give me the same result?

Probably. Its worth a try anyway.

So I've had a go at installing cleanly without AGP drivers, and it made no difference - same behaviour for the TNT2 (i.e. works fine in low res mode, fails to get into Windows once drivers are installed + higher resolutions/colour depths are selected).

Also I took some photos -

Board overview:
https://dl.dropboxusercontent.com/u/683235/DSC00617.JPG

Close up of the cluster of caps near the CPU:
https://dl.dropboxusercontent.com/u/683235/DSC00609.JPG

The cap next to the AGP port:
https://dl.dropboxusercontent.com/u/683235/DSC00615.JPG

If it is of interest, here's the voltage regulator with the big heatsink:
https://dl.dropboxusercontent.com/u/683235/DSC00613.JPG

I had a close look at all the caps in case there were any of the more subtle signs of failure - leaks, or bulging undersides - but all look in really good condition to be honest. Still, next step is going to be replacing that AGP cap just in case.

AaronAsh wrote:

So I've had a go at installing cleanly without AGP drivers, and it made no difference - same behaviour for the TNT2 (i.e. works […]
Show full quote

So I've had a go at installing cleanly without AGP drivers, and it made no difference - same behaviour for the TNT2 (i.e. works fine in low res mode, fails to get into Windows once drivers are installed + higher resolutions/colour depths are selected).

Also I took some photos -

Board overview:
https://dl.dropboxusercontent.com/u/683235/DSC00617.JPG

Close up of the cluster of caps near the CPU:
https://dl.dropboxusercontent.com/u/683235/DSC00609.JPG

The cap next to the AGP port:
https://dl.dropboxusercontent.com/u/683235/DSC00615.JPG

If it is of interest, here's the voltage regulator with the big heatsink:
https://dl.dropboxusercontent.com/u/683235/DSC00613.JPG

I had a close look at all the caps in case there were any of the more subtle signs of failure - leaks, or bulging undersides - but all look in really good condition to be honest. Still, next step is going to be replacing that AGP cap just in case.

Like PCBONEZ mentioned, a good looking cap is no guarantee it's performance will also be good.

Btw, is it me or does the cap next to the AGP slot look a bit funny? As if it's bend a bit. If you have the skills to replace it, then that might be a good consideration.

Another thing: I did read back this thread a bit so I might've missed it, but have you had the chance to take a look inside the PSU? (be careful though and don't touch anything inside it as it could electrocute you. I'm a bit paranoid about this though, but better safe than sorry 😉 )
If the PSU is faulty, then no amount of switching hardware around will fix the problems your rig is having.

If the PSU is good (or if you switched it around with one that is working) then my next suggestion would be to (for now) keep the S3 card in the AGP slot till you figure out what's wrong. If your AGP slot is killing all your graphics cards, it might get immensely frustrating for you...no fun in that!
You could add a V2 so your rig is at least somewhat usable. Or just do what I would typically do: Just build another rig and set this one aside for the time being till you figure out what's wrong with it (just my 2 cents here)

edit:
Oh, very very uber-important question for you here: How did you take those lovely photographs? My camera has gone into retirement and I need to replace it with something better than that crummy old mobile of mine 😁

I've had an ali board (ms-5169) that simply hated any nvidia card. I tried tnt, tnt2, gf2 mx, gf2 ultra and geforce 3. Some games started up fine but crashed, some didn't work at all.
The board worked fine with banshee, voodoo3, matrox g200, matrox g400 and matrox g450. Savage4 worked fine as well. Changing the bios settings for agp to off and the speed to 1x agp made things a bit better but still a lot of bugs with Nvidia cards.

I have several other boards that work just fine with MOST of my cards but some cards give troubles. For example: my Asus p5a works fine with a geforce 2 mx and Geforce 2 gts but not the geforce 2 ti. Another board (jetway) has no problems with the geforce 2 ti but doesn't like the mx..... Geforce 3 and geforce 4 works fine in both.

It is a hit and miss with these boards and agp cards, especially Nvidia imho. i never went into this too deep so there might be a solution, just keep digging.

The best agp driver imho is the ali 1.72 driver and with the right registry and bios settings you might get it working. (if your board is not faulty)

If you want to make your life easier use a pci card.

AaronAsh wrote:

So I've had a go at installing cleanly without AGP drivers, and it made no difference - same behaviour for the TNT2 (i.e. works fine in low res mode, fails to get into Windows once drivers are installed + higher resolutions/colour depths are selected).

I had a close look at all the caps in case there were any of the more subtle signs of failure - leaks, or bulging undersides - but all look in really good condition to be honest. Still, next step is going to be replacing that AGP cap just in case.

Thanks for the photos. It's harder to work on things you can't see.

Visually the only cap I can see that is questionable is the one by the AGP slot. It's leaning off to one side.
Sometimes that's just sloppy installation or it 'got handled' at some point in it's life but it also can be a sign the bung (the rubber plug} it trying to push out the bottom of the cap.
I agree. At least replace the cap by the AGP slot as a just in case.

On the other caps. Bad visually means bad. Good visually doesn't tell you anything for certain.
Only some bad caps show visible signs.
That was true 40+ years before the so called capacitor plague and it still is.
The capacitor plague (and the resulting News media misinformation 'program') seems to have made people think that all bad caps show visible signs.
That has never been true.

Even with premium brand caps any equipment that is acting flaky for no apparent reason, the caps should be suspected.
More so the older it is and especially anything over 15 years old or that has been run in a high heat environment.

The life span / failure rate curve for caps is the famous bathtub curve.
At about 15 years old the failure rate starts going up more or less exponentially with time.

Not meant to say that at 15 years *poof* there are bad caps everywhere. It starts much slower than that.
But if you are still into retro machines 5 or 10+ years from now needing to replace caps is going to become more and more normal.
It already is for people that restore old 70's-80's radios/stereos but the thing is motherboard caps usually work much harder than caps in old general electronics so motherboard caps will tend to have shorter lives.

That is exactly what happened with the old barrel type CMOS batteries. They have a very similar bathtub curve.
A few years ago people knew they -might- have to replace them.
At this point most people automatically assume they -will- have to replace them.

Aluminum Electrolytic Capacitors ARE going to go the same way eventually. It's just hard to say exactly when.
Hard to say when because the caps used (and their uses/jobs/stress) are different from the 486 era to the 586 era to the 686 era and so on.
It may even turn out that 686 era parts become a problem before before 486 era parts.

So. All of that said.
Were that board mine and acting like it is and if I planned on keeping it a while I would refresh the caps on the whole board.
That is me though. I've been doing that sort of thing for years so to me it isn't a big deal anymore.
When I first started soldering on motherboards, that was different. I was convinced I would ruin every one I touched.
I totally get why people that haven't done it much (or at all) have reservations about recapping motherboards.

(General: To no one in particular.)
Recapping motherboards is different than recapping most other circuit boards.
Motherboards are multilayer and thick. Things just don't work/behave the same.
If you have never recapped a motherboard before I recommend practicing on a crap motherboard that you don't care about before working on something you do care about.
.

Just to add to and summarize what BONEZ is saying, the "bottom" of the bathtub curve is more completely represented as a stress function of which age, ambient temperature, cooling, and circuit operation are components. If you want to guesstimate whether you have failed caps, you should take all these factors into account.

Would you like to know more?

An update on this.

Finally got all the caps I needed and had some spare time last weekend, so I went ahead and recapped the motherboard (after practising on a junk one first). That went smoothly, and actually only took a few hours. Firstly I just replaced the cap next the AGP port, but the system still failed in the same way with Voodoo3 and TNT2Pro, then I recapped the whole board and tried again but, unfortunately, it still failed to work in exactly the same manner as it did before I recapped it.

A shame, but the process was an enjoyable little project - it's pretty neat to turn the PC on and have it still work (as well as it was before, anyway) after having replaced literally every capacitor on the motherboard.

I have one last possibility I might try, which is recapping one of the Voodoo3 cards in case the caps on all of them happen to have become faulty. They only have few caps, but they are surface mount so look a little more fiddly to remove - does anyone here have any experience doing this?

If even that doesn't work, I think I may be in the market for a new motherboard. There are a few on eBay at the moment, but the ones that are reasonably priced for me are actually the same model as the one I already have - and I don't really want to risk getting another the same in case it is just a basic compatibility problem. Does anyone have recommendations for other good models of Super Socket 7 board, with an AGP slot? Are there any particularly good/bad chipsets? Is the "AGP Pro" chipset any good?

@Tetrium: Sorry I didn't notice your question earlier, but the camera I use is a Sony RX100 (first model); it has makes significantly nicer pictures than a phone camera, but is also handy enough to keep with me more than a bulky DSLR.

AaronAsh wrote:

They only have few caps, but they are surface mount so look a little more fiddly to remove - does anyone here have any experience doing this?

Overall they are usually easier than through hole.
There are different ways.
If you are going to try the 'rock-loose' method make sure you are not rocking lead to lead. Should be rocking 90 degrees from that.
If you don't have some then get some solder braid for cleanup.
When you put them back clean then pre-tin the pads (small mound of fresh solder) then just sort of melt them into the solder.
https://www.youtube.com/watch?v=Y1l7qSezFDQ
https://www.youtube.com/watch?v=5Ts6lPkYDsU
https://www.youtube.com/watch?v=3bdMS0SsHnQ
.

I'd try measuring some voltages with the system running (under a heavy load would be best). Easiest would be the 5V and 12V pins on a loose molex connector. Don't trust the voltage readings in the BIOS, sometimes they're way off.
It would be very informative to measure the onboard 3.3v rail, but doing that is more tricky. You'd need to find a safe point to measure it from.

If I understand correctly, the original PSU is ATX, and not just the modern one that you temporarily swapped. So nothing you've been doing ever involved the AT connector?
As others mentioned, these type of boards will have an onboard 3.3v regulator which might be inadequate or possibly damaged if it was overstressed. However, when it's running with ATX, I think that onboard regulator would most likely be bypassed. Actually, on a Tyan board that I looked at, the onboard regulator was in parallel with ATX 3.3v. Seemed kind of odd to bridge them like that, but that's how it was.

PCBONEZ wrote:

Max amps for 3.3v on a 1x/2x AGP slot is 6 amps.
It's just under 5 watts for two sticks of SDRAM (I don't have data for modules smaller than 256Mb though), so ~1.5 amps or less.
Unless there are other 3.3v loads the regulator would only have to handle 7.5 amps.

Part of the CPU power also comes from 3.3v. According to my old notes, for the K6-3 450Mhz, max Icc3 is 0.66A.
Single voltage 3.3v CPUs are much worse - some are in the 4-7A range. However, depending on design (and jumper configuration) they might get helped out by the Vcore regulator. Anyway, he's not using one of those.

============
@AaronAsh
Although it might not solve the immediate problem, I suggest trying to get hold of a quality ATX PSU to replace the generic one. Cheap power supplies can be problematic in general, and can have noisy output voltages that can be stressful/damaging to components that are exposed to them. That said, choosing a quality PSU for that era of machine is a bit of a controversy in itself. Modern PSUs aren't optimized for the kind of loads that appear on old systems (mostly 5V / 3.3v with little 12V), and old PSUs (even the good ones) often need to be recapped.

The Windows startup to a black screen sounds like an AGP problem I had on a Tyan VIA MVP3 board. In my case, I think the black screen portion of the problem was solved by enabling onboard USB in the BIOS (seriously). However, your chipset is an ALI which I have no experience with. The non-Intel AGP chipsets from back then are quirky.
It's interesting that you once had it boot to the GUI and show a corrupted desktop. I don't know if that represents something more serious.

As far as alternate motherboard suggestions:
AGP with super socket 7 is inherently frustrating territory. At that time, the only solid implementation of AGP was on Intel chipsets, which of course don't appear on Super 7 boards. PCI cards can avoid frustration, otherwise you may have to fight with settings and drivers to get AGP working on these boards. I'd definitely leave AGP set to 1X. 2X accomplishes nothing in this speed range and just reduces the chances of success.
If you want AGP to just work, a slot-1 440BX board is a great way to go. They don't have the same underclocking potential as super 7 though.
Intel slot-1 boards are great quality and common (a fair number are even out there as NOS), but you have to be careful that the board is truly ATX. A lot of them are Dell versions which at the time had their own PSU pinout. Intel boards also won't let you underclock.
Asus slot-1 BX boards had good caps and will allow tweaking, but they tend to be expensive. Every other brand I can think of usually have bad caps so expect to need to replace them.

Thanks for the replies.

@shamino: No I've never used the AT connector, and I did only use the high quality PSU for a short bit of testing; perhaps I should swap over to that again for a bit more troubleshooting. Is there any reason I couldn't use a mini-ATX power supply with this machine, if I were to buy a new one?

I did spend the better part of a day playing with different AGP drivers, graphics drivers and BIOS/registry settings trying to get the TNT2 to play nice with the motherboard, but maybe I'll take another look this weekend and see if I missed something.

For now I'm going to stick with super socket 7, as I'd like to be able to do the under clocking trick for older games; I unfortunately just don't have the space to setup multiple retro gaming machines.

So I worked on some hardware troubleshooting a bit more just now.

Firstly I recapped one of the Voodoo3s - did not fix the problem or alter the behaviour of the card from before the recapping; a shame.

Thern, I took some voltage readings. The 5 volt line on the motherboard runs at a very stable 5.05 volts and the 12 volts runs at a stable 12.38 volts - is that too high?

Perhaps more interestingly, I got an AGP pinout diagram and tested one of the 3.3 volt supply pins on the AGP slot while a card was inserted (fortunately the pins are easy to access with a multimeter from the top of the slot). So when using the TNT2 card the 3.3v supply sits nice and steadily on 3.3 volts. However, when using the Voodoo3 the same 3.3v pin shows a very different reading - after initial bootup it is just sits at 0v for maybe 10-20 seconds, but then starts to jump about erratically in the range of 2.4v to 3.3v volts, but sometimes dropping back as low as 0v and then shooting back up. Very strange - does that shed any light on the issue to anyone here?

The best way to diagnose it would be a fully-fledged osciloscope, but that is not exactly a cheap piece of equipment.
Btw, with the motherboard off, does the AGP slot's 3.3V contact pass a continuity test with the ATX connector's 3.3V pin? If not, then the motherboard is probably the one responsible for providing the 3.3V rail and the problems could be in the motherboard itself. It could also be that something fried on the Voodoo3 card, but that would not explain the TnT2 misbehaving.

He said the 3.3v is stable with the TNT2 and unstable with the Voodoo.
.
(O'scopes)
The Tektronix TDS1002_ and TDS2002_ (and similar) are required equipment for some electronics schools.
When school is over a lot of students dump them either for cash or to trade up.
If you catch it just right there are little short bursts where they are plentiful and not too expensive.
They don't have many hours on them and they usually have all the original documentation.
These are small portable digital storage scopes. No CRT. Not much bigger than two ATX PSUs set end to end.
.

AaronAsh wrote:

Is there any reason I couldn't use a mini-ATX power supply with this machine, if I were to buy a new one?

I don't know much about those smaller form factor PSUs, but I'd be worried about whether they have much 5v and 3.3v output, which older boards primarily use more than the 12v.
The only experience I ever had with smaller sized PSUs was with a relative's microATX PC. He needed a new PSU, and the replacement only lasted a few weeks. The presumably better quality units were too expensive, so we ended up rigging up a full size ATX PSU after that.

AaronAsh wrote:

The 5 volt line on the motherboard runs at a very stable 5.05 volts and the 12 volts runs at a stable 12.38 volts - is that too high?

The tolerance on those rails is +/- 5%, so it's not too high. The 12v rail is pushing closer to the upper end of the range but the multimeter reading is within spec.
With cheaply made power supplies there's a worry that they might have a lot of ripple and the voltage might go out of spec during short transients, but it would take an oscilloscope to see that.

The 3.3v behavior with the Voodoo3 is certainly strange. Not sure what to make of that.

alexanrs wrote:

Btw, with the motherboard off, does the AGP slot's 3.3V contact pass a continuity test with the ATX connector's 3.3V pin? If not, then the motherboard is probably the one responsible for providing the 3.3V rail and the problems could be in the motherboard itself. It could also be that something fried on the Voodoo3 card, but that would not explain the TnT2 misbehaving.

Whether the slot is powered directly from the PSU via the ATX connector would be interesting to know.

I suspect the problem with the TNT2 could be a result of chipset/BIOS/driver quirkiness with AGP. I know AaronAsh has already been exploring that, but It sounds like how my first Geforce2 MX behaved back when I had a VIA MVP3 board. I had just about given up when I found an obscure mention of a BIOS setting in an FAQ on the mobo manufacturer's web site. It was a setting that seemed totally irrelevant to AGP, but it apparently had to be turned on for AGP to work. Until that resolution, it could run Windows in a non-accelerated VGA mode, but after drivers, it would boot to a black screen.
I also had some driver problems, but I don't remember if they ever caused the black screen symptom. Those might have just caused graphical glitches in 3D.

It is strange behaviour indeed... I forgot to do a continuity test to see where the 3.3v was coming from though, I'll check next time I open it up.

Haven't had a fiddle with AGP settings with the TNT2 yet, though it is promising to hear you were able to resolve it; maybe this weekend (though XCOM 2 may ruin that plan...)

I don't mean to bump this thread unnecessarily, but I want to give a little update for people seeing a similar problem who might find this thread in future.

So I a few weeks ago I secured another SS7 motherboard (a PC Chips M577 with an "AGPPro" chipset) to do some comparison testing with my Acorp 5ali61, and as expected none of the Voodoo3s worked in the new board either but also it turns out that Elsa Erazor III (TNT2) Pro card also displayed the exact same behaviour with the new motherboard too (black screen on Windows login with after drivers were installed); as the new motherboard uses a different chipset this suggests to me it was just a problem with the TNT2 itself. I confirmed this by picking up a cheap Diamond Viper V770 and that actually works fine in both motherboards!

Moral of the story is, sometimes retro hardware is just broken and no amount of tweaking can fix it; the other moral is buy more retro hardware! Tell them its for comparison tests!