3Dfx was doomed right after beginning on PC market

It's pretty obvious that nVidia had an unfair advantage in that they had infinite resources to produce, test, alter, and sell their silicon.

Somewhere along the line, nVidia became a client of the state and had unmitigated access to TSMC.

All 3DFX has in the end was their proprietary API, and Microsoft made that obsolete by making Direct3D good enough.

In retrospect, this turned out to be a terrible turn of events, as they have now cornered the market.

I will sit back and laugh as I observe the sycophants of the technocratic class eat the $1000 GPU prices. I have no use for the modern stuff, I am a classicist.

douglar wrote on 2026-02-23, 01:36:

Not much would have changed unless 3dfx managed to release the voodoo4 6-12 months earlier and if 3dfx had a change of design philosophy with the rampage chip. Those were the things that doomed the company to follow Trident, S3 and Cirrus. The STB acquisition made the cash flow worse, but didn’t change the general trend.

Without the STB distraction they likely would have had pressure to speed up the release schedule and struggled on a few more years.

This is similar to what if Matrox did better and got good ?
or what if Intel had released AGP alongside the Starfighter gpu on the ppro in 1996 when the technology would have been relevant as opposed to later when the core part of agp technology which shared system memory became completely irrelevant and just a faster pci bus due to cheap ram.

Most of the graphics cards on agp would have done just as well on a mythical 64bit PCI bus running at an appropriate clock as the whole purpose of agp side band didn’t matter as ram got cheaper, faster and larger.
It was hard to know The hyper priced ram shortage environment of late 95/96 was going to abruptly end

3dfx paid for STB with stock so it didn't cost them anything. I don't know if STB had any cash reserves or debt at the time. The reason for buying STB was for 3dfx to enter the OEM market which is where S3/Matrox/ATI was making all their money. The OEM market probably made up for losing retail (more money made on a card then just a chip, more volume). The problem with buying STB was that managers spent too much time and money getting that purchased straightened out instead of getting new chips designed. Nvidia made a few missteps in the beginning but their ability to get chips designed fast made sure they got into the lead in features and speed which they have never been dethroned (just a few hiccups here and there). 3dfx was never going to win the 3d wars and we are lucky ATI/AMD is still around.

rmay635703 wrote on 2026-03-03, 00:29:

Without the STB distraction they likely would have had pressure to speed up the release schedule and struggled on a few more years.

I mean it sounds plausible right? Until you think about it. If the company was still only the 12 people who were on the payroll in 1995, then yeah, it would have been a distraction. But there were a lot of employees in the company by the time STB was acquired. If the engineers doing silicon design were hands on in the STB acquisition and card design, the company had bigger problems than we know about.

The 3dfx roadmap was going to be rocky with or without the STB acquisition. Maybe they could have held off the creditors long enough to get some value from the GigaPixel purchase if they hadn't purchased STB. Still seems like a long shot that they would have lasted long enough to get there. 3dfx flubbed two generations in a row with Napalm and Rampage. That was fatal and both were already on the drawing board when STB was brought in.

Here's a list of the Voodoo products and code names:

Released:

• Voodoo Graphics SST-1 Ground breaking chip
• Voodoo Rush SST-96 A ho-hum 3dfx 3D core and a third-party VGA chip.
• Voodoo2 SST-2 Expanded the lead
• Voodoo Banshee "Banshee" The first single-chip 2D/3D solution from 3dfx. Almost OK.
• Voodoo3 "Avenger" chip that we know and love. Smoother frame rates than the competition and a great ramdac
• VSA-100 "Napalm" Late and slow and the multi-chip solutions gave the engineers fits (fail)

Unreleased:

• "Fear" Supposed to be the first DirectX 8 card with programable shaders and hardware T&L but was superseded by "Rampage" before silicon (fail)
• Spectre "Rampage" Intended to compete with the Geforce 3. The 20-30 prototypes that were built were falling way short (failing)
• "Sage" Dedicated hardware T&L chip to be paired with high end "Rampage" boards.
• "Mojo" A planned future architecture intended to follow Rampage, focused on tile-based deferred rendering; never left the design stage.

So that's why 3dfx folded. They failed on two designs in a row and the next design in the pipeline was not promising.

douglar wrote on 2026-03-03, 02:52:

rmay635703 wrote on 2026-03-03, 00:29:

Without the STB distraction they likely would have had pressure to speed up the release schedule and struggled on a few more years.

. But there were a lot of employees in the company by the time STB was acquired. If the engineers doing silicon design were hands on in the STB acquisition and card design, the company had bigger problems than we know about.

. Don’t underestimate the motivating pressure of board partners wanting a product to sell.
Many companies that acted more as a supplier to other companies b to b after trying to vertically integrate have failed spectacularly because the business dynamics are very different.

Possibly apocalyphal but 3DFX decision makers were distracted by finding a board maker to purchase long before they actually bought STB, and the lack of proper leadership (or result of leadership) leading up to stb stifled and confused design language and roadmaps for possibly a year before stb.

I’ve been looking for an interview with 3dfx designers for a while similar in form to the c64 engineering whimsical tales of designing the c128 but have never found much about the culture and freedom to innovate in the organization.

I got a sense that 3dfx had the apple computers early 80’s problem by having a hit then floundering around not long after throwing tons of warm bodies into the org.

rmay635703 wrote on 2026-03-03, 21:17:

Don’t underestimate the motivating pressure of board partners wanting a product to sell.
Many companies that acted more as a supplier to other companies b to b after trying to vertically integrate have failed spectacularly because the business dynamics are very different.

3dfx ended up in the same boat as Cirrus, Matrox, s3, Tseng, and Trident.

Some had third party partners, some didn't, but the one thing they have in common is that none had competitive product in the retail space in the year 2000.

i'd think Brian Hook leaving had to do with the decline, but how much I don't know. He went to id (to iron out idtech3 until May) and then Verant in 1999

There was also that time 3dfx contracted a STD InterAct for the Hammerhead Fx.

Don't forget the TV marketing blitz they've had in May 99, that must've been a lot. They even had sponsored a few episodes of WCW Monday Nitro

douglar wrote on 2026-03-03, 21:58:

rmay635703 wrote on 2026-03-03, 21:17:

Don’t underestimate the motivating pressure of board partners wanting a product to sell.
Many companies that acted more as a supplier to other companies b to b after trying to vertically integrate have failed spectacularly because the business dynamics are very different.

3dfx ended up in the same boat as Cirrus, Matrox, s3, Tseng, Oak, SIS, Intel and Trident.

Some had third party partners, some didn't, but the one thing they have in common is that none had competitive product in the retail space in the year 2000.

I would argue some of them never had a competitive product, but consumer requirements changed and raw price advantage stopped mattering as did reputation, power draw, resolution and refresh rates.

And I can still buy newly manufactured Matrox cards , so they remain the strange outlier.

Matrox was making money in their video division so gaming chips didn't matter much to them. Cirrus and S3 never had a 3D chip for gaming but did well with OEM's and budget 2D office buyers.

Unknown_K wrote on 2026-03-03, 22:17:

Matrox was making money in their video division so gaming chips didn't matter much to them. Cirrus and S3 never had a 3D chip for gaming but did well with OEM's and budget 2D office buyers.

If Cirrus and S3 never had 3d chips for gaming, then what are these?

https://vintage3d.org/cirrus.php#sthash.L5SqukEa.dpbs

https://vintage3d.org/savage3d.php#sthash.4HvTxkCM.dpbs

douglar wrote on 2026-03-04, 00:33:

If Cirrus and S3 never had 3d chips for gaming, then what are these? […]
Show full quote

Unknown_K wrote on 2026-03-03, 22:17:

Matrox was making money in their video division so gaming chips didn't matter much to them. Cirrus and S3 never had a 3D chip for gaming but did well with OEM's and budget 2D office buyers.

If Cirrus and S3 never had 3d chips for gaming, then what are these?

https://vintage3d.org/cirrus.php#sthash.L5SqukEa.dpbs

https://vintage3d.org/savage3d.php#sthash.4HvTxkCM.dpbs

Cards nobody purchased and were obsolete when they came out? Last gasp and then nothing.

Unknown_K wrote on 2026-03-04, 02:03:

Cards nobody purchased and were obsolete when they came out? Last gasp and then nothing.

yeah, that s3tc tech introduced from Savage3D sure meant nothing and certainly never influenced the consumer 3d graphics industry. /s

Unknown_K wrote on 2026-03-04, 02:03:

Cards nobody purchased and were obsolete when they came out? Last gasp and then nothing.

Same could be said for the Voodoo 4500. 3dfx was in the same boat as those companies.

douglar wrote on 2026-03-04, 02:24:

Unknown_K wrote on 2026-03-04, 02:03:

Cards nobody purchased and were obsolete when they came out? Last gasp and then nothing.

Same could be said for the Voodoo 4500. 3dfx was in the same boat as those companies.

Except 3dfx had a long history as a 3D card leader. The Voodoo 4500 was a major flop, and I would like to know how many were made and sold (I have a PCI model I found in a gifted freecycle PC).

douglar wrote on 2026-03-03, 21:58:

3dfx ended up in the same boat as Cirrus, Matrox, s3, Tseng, and Trident.

Some had third party partners, some didn't, but the one thing they have in common is that none had competitive product in the retail space in the year 2000.

Most of them did have competitive products, just not in a niche you care about

• Cirrus would be Audio Codecs. Graphics was just a small side hustle between 1992-96. You probably owned more than one product with their chips inside in 2000 and didnt even know it 😀
• Matrox was making bank. Only game in town for dual VGA output and just entered NLE market.
• S3 in 2000 would be for example Diamond Multimedia Viper V770 Ultra (TNT2) or one of Diamond Rio mp3 players. Then in 2001 Diamond Multimedia STEALTH S60 Radeon 7000 64MB DDR 😉 Yes they also made fail Savage3d/2000, but hey, at least they gave us texture compression. Fail Savage found its niche in ThinkPad 23/30 series.
• Tseng was ATI in 2000 so that would make their flagship ATI Radeon DDR (Rage 6)
• Trident low power graphics in laptops. ThinkPad/Compaq/Toshiba/Sony, they owned the low end laptop discrete 3D space.

V3 dominated retail sales in 1999
https://web.archive.org/web/20231113205402/ht … hottest-selling

industry research group PC Data:
>Voodoo3 products have topped the PC Data charts since their introduction in
>April of this year. Capturing the top five positions in the market points to
>the continued strength of the Voodoo3 product line.
> According to the most recent PC Data report on August sales figures, the
>Voodoo3 2000 PCI was the best selling graphics card in terms of unit shipments
>and revenues. PC Data measured retail sales of the entire video card market
>and in addition to the top position, the Voodoo3 family scored the second,
>third, fourth and fifth spots including the Voodoo3 3000 AGP, Voodoo3 2000 AGP
>and Voodoo3 3500TV.

Voodoo3 was still competitive in the mid range up to June 2000 ~$100 GeForce2 MX launch. Voodoo4 was always trash fire with no redeeming qualities, too expensive for low end, too slow for mid end.

douglar wrote on 2026-03-04, 00:33:

If Cirrus and S3 never had 3d chips for gaming, then what are these?

https://vintage3d.org/cirrus.php#sthash.L5SqukEa.dpbs
https://vintage3d.org/savage3d.php#sthash.4HvTxkCM.dpbs

Broken jokes. Later Savage 2000 also turned defective, but at least this time it was only TnL unit they could disable in software so chips were usable in the low end market.