Creative's own flasher won't allow flashing if the card doesn't already have a Creative BIOS image, but flashing a CT6750 AGP SGRAM BIOS from 3dfxzone.it using the 3dfx flasher gets the Creative flasher to run. But it comes out with the error "AGP SDRAM memory not supported in this BIOS!" The board has SGRAM chips, so something's confused here. Generic Banshee BIOSes for SGRAM and SDRAM produce the same artefacting in any case.

May be that "this BIOS" in the error means the BIOS on the card rather than the BIOS being flashed with and that the flasher is for an SDRAM version. The file description doesn't say, but the D in the filename BNF1032D.EXE might suggest it. But I wasn't aware of there being an SDRAM AGP version of the Creative Banshee.

Got the Creative flasher working, though no change in the artefacting. Claims the card has 1 MB of VRAM, which is obviously a problem. I don't have hot air to reflow the graphics chip, the card is going in the bin if that's where the issue is.

Considering how collectable 3Dfx cards are, don't just bin it. Collectors are even buying non working examples.

Check all the passive components with a multimeter. Cracks in surface mount components can be invisible even under magnification. You might get lucky and discover something you missed before.
Unfortunately bad solder joints under the BGA is likely possible. If force was ever used to remove the heat-sink, it is almost certainly the problem.

The voltage regulator looked ok with 5 volts in and 3.5 volts out. No idea what voltage the Banshee is expecting.

All resistors measured good except the 33-ohm array RP10 next to the heatsink where one resistor had unusual readings, in the 20-50 range. Out of circuit it measures dozens of ks and doesn't settle, the other resistors in the array still measure fine. Looks fine under a microscope. I don't have time to replace it right now, place your mental bet that this failure causes these symptoms. Keeping in mind the card was not only artefacting but also unable to flash the type of BIOS chip it originally came with.

Replacing RP10 fixed the artefacting, at least in DOS:

I think the failed resistor went into a bank select pin on half the memory chips. The card needs a heatsink before testing in Windows, but looks good on that front.

As a reminder, the card was sold as tested and working, yet had no output without replacing the BIOS chip and then needed this resistor array replacement to fix artefacting. The magic of Voodoo on eBay. It's possible these aren't the only things that need fixing either.

vvbee wrote on 2025-04-30, 11:01:

As a reminder, the card was sold as tested and working, yet had no output without replacing the BIOS chip and then needed this resistor array replacement to fix artefacting. The magic of Voodoo on eBay. It's possible these aren't the only things that need fixing either.

Did the seller at least provide pictures/screenshots in the eBay listing? Apart from some cases, I usually tend to go with listings like that. It's a shame what goes on there sometimes.

Good find, an open or intermittent memory address signal would definitely cause the problems you were experiencing. The damage could have easily occurred during shipment.

While not common, EEPROM chips are known to fail on 3Dfx cards.

It is entirely possible the card actually worked when the seller last tested it. Considering most of the desirable parts on ebay are now recovered from scrap, an unscrupulous seller is still just as likely.

No screenshots, but the previous Voodoo I bought as tested and working had screenshots but simply not of the parts that were artefacting. Well-rated typical sellers.

Just a flaky BIOS that works in one motherboard but not another would be somewhat plausible, but this card doesn't look too likely to have been working when sold. I think at best the seller was confused about which of their cards were working.

The 3dfx flasher has a bug with SST chips where the program gets stuck in an infinite loop if the chip doesn't read back the correct data after writing. I modified the code to do restarts after failed writes, which happened occasionally and normally would just do the infinite loop, but even when all bytes were written and verified the flashing wouldn't produce a working BIOS. It's possible the issue was in reading data from the chip rather than in writing to it, and the one replacement chip of the same make looked to have the same issue. To me this says maybe the voltages the card was now feeding were too low for these chips and the other working models had higher tolerances, but no idea. I did put in a socket for the BIOS but I don't have a proper chip puller right now so don't want to mess about with it testing the old chips again.

Works. Although for current prices it should work ten times better.

Voodoo is now better to use for games with Glide instructions. And for DirectX games you can use any GeForce.

The Voodoo hardware is temperamental and expensive, but the GeForces have their issues as well. I only know of Matrox and 3dfx that render Need for Speed 5 in this particular soft way, ATI to some extent maybe. Newer cards have an overly sharp and sterile look to their texture filtering, I haven't seen any driver settings that perfectly replicate the softer look in them. It should be possible but haven't seen it.

There's some factors:
- 3dfx has two dither modes and optional dither subtraction (that can create certain artifacts). Different APIs and cards' drivers change these defaults, so V2 could have 4x4+DitherSub where a VB/V3 would just 2x2 for one example. UnrealEngine1 games always use 2x2 without dithersub.
- 3dfx cards all have a filter. Banshee has options for the Voodoo1 4x1 and the "Voodoo3 22-bit" 2x2.
- 3dfx has a higher than default gamma correction value, usually 1.3 (1.7 for V1)
- 3dfx also has a slightly differing lod bias.
- 3dfx has interference. 😀

Voodoo 3, G400, GeForce 256, GeForce 2:

Baseline texture resolution is the same for all except somewhat worse for the Voodoo. Looks kinda like a game bug of incorrect mipmap levels for the newer cards. I wasn't able to find driver settings for a Radeon 9600 either that would've had the mipmapping look correct rather than too sharp. Almost a guarantee the way the G400 renders it is how it should look.