I don't have that computer with me right now, but here's a drawing:

DO NOT TRUST THIS PICTURE BLINDLY, PLEASE RECHECK THE PINOUT BEFORE DOING ANYTHING. It's 11.30 at night, I'm tired and my eyes were crossing while colouring the pads. It should be right, but please check again before.

Basically you get the pinouts of the PCI slot and check if the pins supposed to carry 3V3 are left NC. If so then you can wire them to a 3V3 source of your choice. If you are using one of those stepdown converters as in my beautiful art piece be mindful that their output is quite noisy and might give you interference on the picture (or, at least, this happened to me). To "fix" it i replaced the output cap on the DC module (originally 220uF) with a bigger one (genuine fake sanyo 1500uF from china, but good enough for this). In a ideal world, you might want to think of building a proper filter, with chokes and whatnot. I CBA and it still works.

Another thing to be mindful of is that by doing this you will only provide power to the slot you wire, not all of them.

andrea wrote on 2020-11-26, 13:32:

I'd guess the problem is not PCI spec but rather the lack of 3.3v supply on older boards (Remember AT does not have a native 3.3V rail). Many newer graphics work off (at least partly) the 3v3 rail, I know for a fact the ubiquitous ebay Rage XL does.
I have a 430TX board whose PCI slot only supply 5 and 12V, to fix it i just wired a DC-DC stepdown converter going from 5 to 3V3 to the PCI slot pins that are supposed to carry 3.3V.

Hmm. The PCI spec of 5 vs 3.3 voltages refer to logic levels, though. I don't think it has anything to do with the actual powering of the cards.

At any rate, I've decided to just stick with the system's onboard s3 virge. There is little benefit to having this card in the system. ultimately a 3dfx card utilizing glide optimized games would probably work better for this hardware... opengl is still slow (15-20 fps or so), i imagine its a combination of the CPU bottleneck and/or other driver/sw limitations of the era.

drosse1meyer wrote on 2020-11-28, 14:03:

andrea wrote on 2020-11-26, 13:32:

I'd guess the problem is not PCI spec but rather the lack of 3.3v supply on older boards (Remember AT does not have a native 3.3V rail). Many newer graphics work off (at least partly) the 3v3 rail, I know for a fact the ubiquitous ebay Rage XL does.
I have a 430TX board whose PCI slot only supply 5 and 12V, to fix it i just wired a DC-DC stepdown converter going from 5 to 3V3 to the PCI slot pins that are supposed to carry 3.3V.

Hmm. The PCI spec of 5 vs 3.3 voltages refer to logic levels, though. I don't think it has anything to do with the actual powering of the cards.

At any rate, I've decided to just stick with the system's onboard s3 virge. There is little benefit to having this card in the system. ultimately a 3dfx card utilizing glide optimized games would probably work better for this hardware... opengl is still slow (15-20 fps or so), i imagine its a combination of the CPU bottleneck and/or other driver/sw limitations of the era.

I'm not talking logic levels, but actual supply rails. Even a (modern) 5V signaling PCI is supposed to provide 3.3, 5 and 12 volts. On a AT board the only way to supply 3.3V is by adding a dedicated regulator. Like a dog biting it's own tail, period cards did not require 3.3V, so board manufacturer didn't bother, so cards did not use 3.3V...
3.3V started being used after ATX become commonplace, and AGP required it.

As I said, on my Rage XL both the regulator for the core and the SDRAM are wired straight to the 3V3 supply pins (A21, A27, A33, A39, A45, A53, B25, B31, B36, B41, B43, B54). If there is nothing there then obviously the car doesn't get power and doesn't work.

Would it be possible to feed the voltage in through a card in another PCI slot?

douglar wrote on 2020-11-28, 22:08:

Would it be possible to feed the voltage in through a card in another PCI slot?

In my case the each slot had the 3.3v pins going nowhere, so no. If you don't want to touch the board I guess you could mod the card, injecting 3.3V on the appropriate point on the card itself.

Maybe I'll mode one of these once I finish my storage project--

In my instance, I'm not sure if it's a PCI power problem per se, since the vid card actually works, albeit only with a specific card in the other slot. My PC also utilizes a riser board. shrug

I found my PCI riser card. Small correction to your diagram.

All of the orange points are connected in this riser, red pin is not 3.3V.

I'm using an ATX --> AT power adapter. I should be able to pull the 3.3v orange wire from ATX and feed that in to one of the orange pins here, and it should power a PCI card, yes?

I checked my available AT boards, and none of them have the 3.3V pins connected to anything, except for a Gateway BAT4IP3e board that does.
It supports 3.3V CPUs, but the 3.3v PCI pins are not connected to the CPU voltage regulator. The 3.3v PCI pins are connected to the three inside pins on this supplemental power connector.
The other three pins are ground.

Which may explain this connector on this ATX powersupply that I never used that is just label AUX. The one red wire worries me though. I should probably hit that with a meter.

More research and I found a Full Yes 430TX board (2A59IF3BC-00) that is baby AT and has 3.3v going to the PCI. It has AT and ATX power connectors, but curiously, the PCI 3.3v doesn't come from the ATX power supply, it comes from the socket 7 voltage regulator. Lol!

Thanks to this thread I was able to get TNT2 M64 PCI working in my socket 7 system (P/I-P55T2P4 Rev 3.10). It was the same problem (no 3.3V on the PCI slot), so instead of trying to add 3.3V to the slot, I added 3.3V to the card with a small TPS62827 buck converter module.