I expect what chatgpt is referring to is resetting the PC to work in real mode rather than protected mode. I don't see how that could kill it.

Have you tried cleaning the contacts on the voodoo agp connection pads? Maybe there's some residue on it?

Re-summarized using ChatGPT (sorry, I typed a lot of stuff out yesterday while I investigated this with my meager home tools):
=======================

Update after second-system testing – failure mode narrowed down

I wanted to follow up now that I’ve had time to do controlled testing on a second BX system and gather more evidence. The picture is a lot clearer than it was when this first surfaced.

Original system: Intel SE440BX-2
Secondary test system: Gigabyte GA-6BXE (rev 1.9, used strictly as a diagnostic bench)

Summary of new observations

• On the original SE440BX-2, the Voodoo3 3000 AGP now presents as a black screen with video POST beeps, while the system otherwise continues booting blindly. Swapping in a Rage card immediately restores normal operation.
  
  On the GA-6BXE, the card does sometimes initialize and produce video output, but the output is severely corrupted:
• Jumbled or incorrect BIOS text
• Snow/static-like artifacts
• Repeating block patterns and corrupted glyphs
• Wildly corrupted Windows 98 splash and desktop graphics
  
  Importantly, the corruption persists even when Windows falls back to Standard VGA. This rules out driver issues entirely.

Firmware ruled out

To eliminate the BIOS angle, I dumped the video BIOS directly from the card using DumpBIOS:

• Dump size: 64 KB (with a valid 32 KB option ROM header)
• BIOS ID string:
  “Voodoo3 3000 BIOS – Version 2.13-MB BETA”
• The dump is structurally valid, readable, and was hashed (SHA-256) for reference.

This strongly suggests the video BIOS itself is intact and not the root cause.

What the symptoms point to

Based on the combination of:

• Pre-OS corruption (BIOS screens),
• Corruption in multiple video modes,
• Persistence under Standard VGA,
• Intermittent ability to POST depending on board/environment,

…the most likely failure mode is hardware-level VRAM path failure, not firmware or drivers.

The leading candidates are:

• One or more failing SDRAM (VRAM) chips
• A marginal VRAM address/data line or solder joint
• A fault in upstream power/steering logic feeding the VRAM (regulator or bus/logic IC)

[*]This lines up with known Voodoo3 failure cases where VRAM corruption presents as readable but scrambled text, snow, and block artifacts, sometimes worsening after a reset or mode transition.

Why it briefly worked before failing

The card ran Quake II for ~40 minutes prior to the failure event. The current thinking (funny, ChatGPT...) is that the hardware was already marginal, and a reset / mode transition (Win98 → DOS → Windows) pushed it past tolerance. After that point, the failure became persistent.

Current status

• The card is not dead-dead, but it is not reliable or usable in its current state.
• I’ve stopped power-cycling it to avoid further degradation.
• It’s being preserved as a candidate for proper bench diagnosis rather than DIY reflow or “baking.”

I’ll be posting a few photos next that show the corruption patterns on the second system, since they illustrate the issue better than words.

If anyone here has experience specifically with Voodoo3 VRAM rail or bus-logic failures (as opposed to simple ROM or driver issues), I’m very interested in hearing how those cases were confirmed and handled.

(end of re-summary) So yeah, I have pulled the card. The contacts look good to me. All the components on my card look good. Although I'm squeamish about trying to remove the heatsink. My GPU is definitely glued together with epoxy. In any case, I appreciate everyone's time and I'm considering hiring some skilled person to do some board work...I guess. I have a cousin who says he can do it if I buy the chips, but having watchined Bits und Bolts targeting a switching BUS chip prior to replacing memory chips because it can be a 'silent killer,' I'm just not sure about how to proceed. Seems like it could be a lot of things. I just don't want to make it worse before someone more skilled than I can make it better.

The actual card, to my untrained eye (although I've been watching these rescue Youtube PCB videos), is that my card is essentially a good-looking card. What is so bizarre to me is that on the GA-6BXE board, it shows the board's BIOS in perfect English, and the Energy Star logo, all perfect, and when Windows boots, hitting start brings up the start menu...looks perfect. But it's own wrapper is jumbled, the BIOS screen where the enumerated IRQ assignments run right before handoff to the OS...all jumbled. And then you can see what the state of Windows looks like now. This is after the first extended use of acceleration I did - the forty minute Quake 2 session. I'm getting a better understanding of the failure types after watching twenty Youtube videos, and the weird purple bit attached was more of a booting into Windows before Windows could reset to Standard VGA than anything else. I can't think of anything else to add. I lost a bid on a new Voodoo 3 card on eBay and now I'm extra sad.

Defective RAM. Even VMTCE would confirm this.
Look thru vswitchzero's clips on Voodoo 3, he replaced the RAM on a few Voodoo3 like yours - some of the cards proved to have defective RAM chips as well, and I can also confirm this as I've had to replace RAM chips on a few Voodoo 3s myself too.

Especially with the way that DOS corruption presents, it's clearly RAM at fault. Those 6, 5.5 and 5 ns chips used on those started developing an tendency to fail - quite expected from hardware that's 27 years old as of this year.

VMTCE and vswitchzero, okey doke. I will try looking into/using that. 'ppreciate you.

I just won a bid on a second card on eBay, so I can cruise and fix this thing at my leisure. "You hear that, buddy! You're gonna have a little brother!"

First post, second picture -> C93 (at end of AGP slot) is severely damaged/missing and what's that "scratch" over the "3" of "U13"?
Fourth picture: C11's looking suspicious (top left of picture). Could also be random compression noise...
Fixing these will not likely fix your card but read this as "there may be more damaged or missing components".

Also the AGP contacts can look great but still don't work. A good clean with some IPA ("cleaning alcohol") won't hurt but can rule this one out.

Good luck!

ChrisK wrote on 2026-01-09, 07:45:

"C93...is severely damaged/missing"

DUDE! Eagle-eyes! That sent me on a solid tangent. I took some extreme closeups and then compared them against my new board that I won from an auction battle (oof, the cost, tho). And I'm learning so much about ceramic capacitors and other thing, 🤣. Please see the two C93 images and the other card's good C93. ChatGPT swears replacing this first might (I acknowledge it only as possibility) save me in replacing the memory chips with my cousin (I have sourced 25 chips from Digikey just to have them on-hand in case I ever need them) by potentially fixing the current behavior, which was Quake II for 40 minutes...then garbled text/posts okay...and then now, doesn't even post at all.

ChrisK wrote on 2026-01-09, 07:45:

"and what's that "scratch" over the "3" of "U13"?"

I think this is an anomaly. I took additional images and the seeming scratch doesn't reappear. I included a closeup of the U13 region on Card A (the busted card).

ChrisK wrote on 2026-01-09, 07:45:

"Fourth picture: C11's looking suspicious (top left of picture). Could also be random compression noise..."

Attached is a closeup different angle of the C11-C14 and some other incidentals. My eyes don't see anything. Which is great, if good! Thanks for pointing at them.

ChrisK wrote on 2026-01-09, 07:45:

"Fixing these will not likely fix your card but read this as "there may be more damaged or missing components"."

No, no. I appreciate it. And also, ChatGPT - who I refer to as my SLOPBRAIN, seems to think that my specific - as I already mentioned - symptoms might get resolved if I target this fix first.

ChrisK wrote on 2026-01-09, 07:45:

"Also the AGP contacts can look great but still don't work.

For sure. I have some like 96% around here...but I'll invest in some pure like 99.99999% stuff and give a good going over. 'ppreciate the input.

ChrisK wrote on 2026-01-09, 07:45:

Good luck!

YESSIR! This is kinda fun. Wish I didn't work 70 hours a week so I could do it more, 🤣.

Offered merely as a further reference, some other component images I want to keep in mind. Youtuber Bits und Bolts has a video about a Silent Killer that quietly kills memory modules: to wit, the BUS CHIP. Mine has some kind of dark pattern that I'm not sure is dust yet, or scorch or burn. I'm going to source some 99.9% alcohol or whatever and so some component massaging with some Q-tips. But the voltage regular also looks a little weirdly dirty on one side. And then the SST BIOS underneath the card. That doesn't look weird, but flashing the BIOS has been brought as a fix in other Youtuber videos, like Bits und Bolts, and Necroware (whose memory tester I just tried the other day for the first time - FUN! - doesn't work on my bad card though, because current behavior is it no longer posts at all), and uh...'vswitchzero.'

OMG YOU GUYS

On a lark, I used Youtuber 'vswitchzero's Windows 2000 dual-card VGA cable swap trick AND IT WORKED!

I thought it was a good candidate for this next step because I had proven to ChatGPT that it enumerated fine under a dual-card setup on my 2nd Slot1 test bench Gigabyte 6BXE board, has its driver installed, and resources assigned and the OS reports no conflicts...and sure enough. Setting the OS to extend the desktop and then switching the VGA cable (carefully) and walla! Perfect output, no garbled data or display artifacts. Nothing bad. I could even set it as the main display and turn extend on an off, switching the OS's main menu taskbar to it and basically turn the S3 Trio64V+ 2D card in the PCI slot off.

The only thing I can't prove right now is an AGP load like running Quake II. And that's because my 6BXE is Rev 1.9, which has inferior MOSFETs (TO-220) that the Voodoo3 3000 AGP card absolutely turns into skin melting madness. But I'm encouraged now to just put the S3 Trio64V+ in a PCI slot, and the bad Voodoo3 3000 AGP card in the AGP slot of my "main" (now) Slot 1, an SE440BX-2, and try re-running it. But that's a problem because that's my Windows 98SE OS and it won't let me use my Compact Flash as a hard-drive base. It just won't play nice with it. So, in order to use Windows 2000 on it, I'd have to make a new SCSI image and stick it on the ZuluSCSI. That's the only way to can try to put a load on this sucker to see if it'll handle it.

Basically, I'm trying to find out 1) are the memory modules really the problem after all, in any way, 2) replacing the broken C93 capacitor...does it fix All The Things?, and 3) does a re-flash of the BIOS fix All The Things without even touching the C93 capacitor at all?

I'd do the following:

1. Clean the card with distilled water + soft brush, then rinse with IPA and dry completely.
2. Replace C93 or at least remove its remnants (card will most surely work without it and should at least POST).
3. Check the bus switch: Is it just dirty or actually discolored? Does it get hot (should not)?
4. Check C41 (looks a bit leaning to the side, but could also just be a perspective thingy).
5. Look out for more damaged parts or scratches.
6. Reflash the BIOS with the exact same version or, as a first step, just read it out and compare it to a known good image in a hex editor. I think I have seen this youtube vid with the Voodoo as 2nd card in W2K, but not sure anymore about the conclusion/fix.

I wouldn't suspect the RAM chips since the desktop looks good. But who knows. I'm not an expert in this regard unfortunately.

ChrisK wrote on 2026-01-15, 10:19:

1. Clean the card with distilled water + soft brush, then rinse with IPA and dry completely.

I am now officially on my first vacation in like a decade, so I have plenty of time to mess around with this.
Distilled Water and some anti-static brushes/soft brushes shipping to me. I even splurged on a digital microscope.
I'm really not trying to become Mr. Solder Guy, but it might be unavoidable.

ChrisK wrote on 2026-01-15, 10:19:

2. Replace C93 or at least remove its remnants (card will most surely work without it and should at least POST).

Working with ChatGPT over the last week, and fighting with my available tools, has yielded some additional
clarifying information about the nature of the card's failures, but C93 is the primary target, with a BIOS re-flash
right behind it, if repairing C93 doesn't wind up working. It was super crazy testing the cards in two
different ways and learning that the card is actually not dead-dead.

After some trial and error, was able to put Windows 2000 Professional on a real IDE hard drive and use
dual-card trickery to swap the VGA cable and Extend Desktop, and lo and behold, the failing Voodoo3 card
was handling the 2D desktop, without artifacts, like a champ. But, for reasons, I wasn't able to test a Quake II load
on it. So, overwrote that HDD with Windows 98 again, settling on a test of my new Voodoo3 3000 card,
updating the OS, and patching Quake II, and getting the drivers installed for the card that came with
the card when I purchased it from a reportedly 'only one owner over the life of the card fella on eBay. Ran the new
card for a half hour doing Quake II demos, and let it warm itself up, blowing air across it in an ad hoc fashion (spare fan).

After letting it cool, however, and swapping in the previous Voodoo3 3000 card, it was the BIOS (SE440BX-2) once again
emitting the Long Beep and Two Short Beeps and then the system shutting itself off before AGP -> VGA. I mean, I'm
super relieved it doesn't appear to be memory module related failure, but it's also more confusing to me because it's
like 'board voltage weirdness' and Intel AGP hard-lines of intolerance that's happening. Guessing from what
The GPT be telling me.

ChrisK wrote on 2026-01-15, 10:19:

3. Check the bus switch: Is it just dirty or actually discolored? Does it get hot (should not)?

The Bus Switch: I tried numerous times with 99% alcohol to get the dark soot looking bit to clear, to no avail. I could perhaps stick the card into the 6BXE (my other Slot 1 board)
and finger test it for hotness, but I don't know. Since it won't post on the SE440BX-2, all I can say is it look suspect to me and it was featured
in a Necroware video about it being a Silent Killer that can age memory modules into the grave.

ChrisK wrote on 2026-01-15, 10:19:

4. Check C41 (looks a bit leaning to the side, but could also just be a perspective thingy).

I put my finger on this little barrel fella and added perhaps a little more force than I intended and it did kind of wiggle into a straightened position. I tried to capture
some still images of what the solder points look like, but my phone camera isn't fantastic and I'm untrained, so...it goes on the list of things I might have broke.

ChrisK wrote on 2026-01-15, 10:19:

5. Look out for more damaged parts or scratches. […]
Show full quote

5. Look out for more damaged parts or scratches.

I'm going to use my new toy to see if I can't get some better, extreme closeups of the tiny bits, and maybe see under the heatsink. I want to replace this thing
so bad, but I'm also keen to not mess with parts of the card that aren't broken, and I think the 2-card swap trick in Window 2000 has convinced me that the actual GPU
and its little solder balls are probably fine. Provisionally 'fine.' One of these days, I'll get the gumption to un-rosin the heatsink, carefully, and stick something
better on there that will do the heatsink work but allow me to re-paste form time to time.

ChrisK wrote on 2026-01-15, 10:19:

6. Reflash the BIOS with the exact same version or, as a first step, just read it out and compare it to a known good image in a hex editor. I think I have seen this youtube vid with the Voodoo as 2nd card in W2K, but not sure anymore about the conclusion/fix.

I did see a video about the flashing utility and the different version of the BIOS, and the (I think) German site that is hosting all the different versions,
and after C93 gets fixed, this is the very first thing I'm going to do.

ChrisK wrote on 2026-01-15, 10:19:

I wouldn't suspect the RAM chips since the desktop looks good. But who knows. I'm not an expert in this regard unfortunately.

Yeah, the thing that made me wonder about the memory modules was the way it went from "working fine in Quake II for forty minutes" to "garbled text in DOS and Windows" and finally just not posting at all...

You're organized and talk like an expert, which is better than where I'm at. I appreciate your input.

The beep code of your motherboard seems to be VGA related. So nothing that's not already known.

C41 looks to be ok from that perspective in your latests pic. A little wiggling in the axis of its solder pads is ok. If there was really something wrong you would have felt it.

Nothing else regarding damaged components to spot on the rest of your pics.

Having watched the video from Necroware now I suggest you to concentrate on that bus switch. Measure resistance from 5V to GND, that is, between its pins 12 and 24 (see https://www.alldatasheet.com/html-pdf/82611/I … 5/2/QS3861.html) or alternatively between voltage regulator pin 4 (EZ1582CM) and GND. For GND you can simply use the slot bracket. Just like in the video. It's very fortunate that you have another card to compare to.
If you're lucky it is just that switch that makes you trouble. If not there'll be more work to do. Let's hope for the best.

As far as I understand it (and that's the point where I'd differ from the video which links that switch to RAM function -> would need to make some measurements first to confirm this assumption) the bus switch connects some of the address lines from the flash chip to the feature connector which would make kinda sense in respect of the problems you desribed. With the switch beeing defective and impairing the communication with the flash chip by corrupting things on the address lines the system won't be able to boot due to not beeing able to access the cards BIOS. I don't know if such a card could be functional as a secondary device so I can't comment on that.
What I'd exclude is some interfunctional relation between that bus switch and the RAM chips itself since I don't see any technical reason for that. The 3dfx chip and the RAMs just don't need any kind of switch between them. They are designed to directly work with each other so this wouldn't make any sense at all.
On the other hand, switching address lines also connected to the flash chip to the feature connector after boot but not before POST may have some technical use case.

So my next steps would be:
- measure and compare resistances as stated above
- if there's found something conspicuous remove the bus switch and measure again
- if both cards show similar values now test the defective card on your known to work installation (make sure there are no shorts on the solder pads of the bus switch's footprint), in theory the card should also be good without that switch
- if it POSTs and works it's up to you if you replace the switch or not (provided all theory about it's function is correct)
- if it POSTs but has graphical errors there may additionally be some defective RAM chip(s) which would require further diagnosis

Again all of this is based on the theory of the bus switch connecting address lines also used for the flash chip to other connections.
If you haven't done work like this before it may be good advice to get help from someone with experience in SMD soldering work (and also the right equipment).

Edit: If something from the above holds true it won't make sense to reflash the BIOS in system since it's obvious that it won't work if something's wrong with address lines.

Edit 2: Here the pictures show very clearly that this switch doesn't connect to anything RAM related. Different card but the principle is very likely the same.