You may remember a company that practically dominated the VGA landscape back when ATi was a smart-mouthed kid and Nvidia was still practically a toddler. 3DFX was a heavy-hitter in the past, smashing performance records year-on-year and introducing some new tech that we still use today. While their memory lives on with Nvidia, it was a company that pretty much kicked ass back in the day, but was let down by mismanagement, terrible planning and not being able to keep up with the rate of development of their competitors.
The roots of 3DFX actually take place within the now-defunct Silicon Graphics Incorporated (SGI), another GPU manufacturer that had a healthy market share back in the day. Three engineers at the company, Ross Smith, Gary Tarolli and Scott Sellers, left SGI in early 1994. They received financial backing from Gordie Campbell’s venture company, TechFarm. If you’ve never heard of TechFarm, don’t worry – they back the start-up companies with ideas that later turn into long-term investments for TechFarm, often with significant paybacks to the company in the long-term. That’s what venture capital companies do – they loan you money if you’re a start-up and later get their money back with no interest repayments. However, most venture deals aren’t publicly detailed and some may even include inflation when you’re repaying them. They’re not in it to make a profit, but the various shareholders in the company are. Anyway, 3DFX’s founders got wads of cash to start up their own party.
In the beginning, 3DFX focused on development of their products and acted as an OEM for partner companies. 3DFX produced the chips and reference designs but didn’t actually get into the consumer channel, nor did they involve themselves in distribution – a practice that other manufacturers later adopted and a rather important part of 3DFX’s future. In many ways, you could consider their method as similar to how Nvidia and AMD operate today.
Their main lineup was called Voodoo and the chips were first seen in the wild in 1996, in an arcade machine featuring a new game called ICE Home Run Derby. It was a baseball game that included a bat with motion sensors, allowing you to swing the controller in real-time and see the player’s arms and torso aim in the same way. For its time, it was an advanced machine and it was the first home for the Voodoo GPU. Some of these machines are still running in arcades in America and Japan.
What 3DFX first created was a drop-in interface module for modern JAMMA systems (Japanese Amusement Machine Manufacturer’s Association) that were popular at the time. Many older readers might remember them and they were favoured by arcade managers because they were robust and featured network ports to allow the machines to be linked together for multi-player fun. This board drove the sensors for the bat and plugged into a PCI slot on the mainboard. Also in the same system was the company’s first card produced by one of their partners, Quantum3D.
This was the Obsidian SB50-2200. It featured a single texture unit, one frame buffer processor and a DAC manufactured either by AT&T, ICS, or similar companies operating at the time. The card had just 4MB of EDO RAM and 50MHz core and memory clocks. On its own it wasn’t particularly pretty and your Nokia 3110C today is easily a hundred times more powerful. It was also only capable of 3D and required a separate card that handled 2D interfaces in the same system, joined with a VGA cable. However, the Voodoo chipset brought with it a huge breakthrough in 3D graphics and was more powerful than any competing design. Towards the end of 1996, the price of EDO RAM dropped significantly and allowed 3DFX to transition into the desktop environment. At this point in time, DOOM and Quake were at the pinnacle of 3D design but both were hamstrung by contemporary graphics solutions. Once the Voodoo chips landed into the hands of gamers, the company saw a massive surge in sales thanks to the speed-ups the Voodoo chipset brought to these two AAA titles.
At the same time, the biggest threats to the Voodoo were PowerVR and a company called Rendition. Both were in the process of producing graphics cards that could handle both 2D and 3D conversions without relying on daughter cards, making modern systems less complex while also reducing ambient heat and power consumption. These cards had 3D chips that were considered merely adequate, whilst separate hardware designs designated for either purpose were much faster. Despite the performance drop, they were preferred because they were cheaper. Companies vying for market share in this new low-cost market were Matrox, S3 and ATi – all three companies which are still in business today.
During their early success, the company developed a API for their Voodoo chipsets called Glide. Compared to OpenGL, Direct3D and Quickdraw3D it was much quicker because it didn’t have to fight for resources. Unlike a modern API like DirectX, Glide only was as capable as the card powering it. Later revisions to the API made it significantly more powerful but it was ultimately destined for use in arcade games. Glide didn’t use a hardware abstraction layer and was much faster than other solutions because it didn’t have any RAM overhead. It was abandoned in 2000, with OpenGL and Direct3D receiving much more support as GPU manufacturers added more and more RAM to their designs.
A year later in 1997, 3DFX returned with the Voodoo Rush, their first card designed primarily for the consumer market. They contracted other third-party companies to include their 2D chips in their design and this allowed 3DFX to get to market faster, owing to the fact that they didn’t have to design a 2D chip of their own. On paper it looked like a good idea, but in reality it was just as meager and mediocre as the cards that 3DFX were competing with a year prior. The EDO RAM was shared between the 2D and 3D chips and that meant that the RAM and PCI bus were overloaded often, especially when gaming in windowed mode. It was a disaster for the company and they had to discontinue the Voodoo within ten months of its release. Third-party designs later alleviated the issues that plagued Voodoo, although it was too little, too late at this point.
To add salt into its wounds, a deal struck to include Voodoo into the Sega Dreamcast fell through halfway through the year, with both companies ending up in court over intellectual property disputes and reasons as to why the deal fell through. It dragged on through the whole year, with the matter finally being settled out of court in late 1997. Ironically, this was also possibly the nail in the coffin for Sega, as the Dreamcast became a paperweight the moment Sony’s Playstation 2 hit the market.
1998 brought with it a renewed hope that 3DFX would earn back market share and the Voodoo 2 pretty much drove that home. It was back to a single 3D chip, compared to rivals like ATi’s Rage Pro and Nvidia’s 128, both of which had already incorporated 2D and 3D chips in a single die. Voodoo 2 also had its own quirks, only offering 800 x 600 resolution and 16-bit colour support. But that didn’t matter, because it was much faster at 3D, once again, than anything on the market. It had two texture units and would pretty much do double the work of competing solutions. Months later, Nvidia’s TNT would present little challenge to the Voodoo 2 and many regard it as the company’s finest achievement.
Voodoo 2 also brought with it the world’s first working example of SLI. Scan-Line Interleave, as it was then called, allowed consumers to link up their Voodoo cards with a SLI ribbon cable. The image rendered on the screen was divided in half with both cards doing their share of the work and a software component fitting it all together in the end. It wasn’t as effective as claimed, mostly because it was CPU-bottlenecked. The true gains using 3DFX SLI were only realised in 2004, when a patch released for Doom 3 (yes, Doom 3) allowed Voodoo 2 owners to actually see just how fast it really was. The way Nvidia renders the frames today is through each frame alternately rendered by the GPU, as opposed to each one working on half a frame.
AFR or Alternate Frame Rendering is still the achilles heel of the multi-GPU platform today.
After the Voodoo 2, though, cracks began to show. At the end of 1998, the new Voodoo Banshee traded 3D dominance for 2D performance, making the card more suitable for desktop use than gaming, although it could match the Voodoo 2 in certain situations. Their main rival, the new Nvidia RIVA TNT, stormed the party and was faster, in addition to supporting 32-bit colour support. The decision to acquire STB Technologies also was a poor one, as neither company could integrate with the other and 3DFX’s hopes of entering into the consumer channel as its own brand were dashed. As 3DFX hoped to enter into the market as both a manufacturer, distributor and a name brand, Nvidia left it to become an OEM supplier, leaving the market in the hands of partner companies that would go on to release their own products based on the Geforce chipset and reference designs distributed by Nvidia.
In another twist of irony, STB was actually a customer of Nvidia’s and supplied a good deal of chips to OEM markets and was one of Nvidia’s primary resources for distributing their products. Once STB was bought out, its customers left and formed their own retail channel companies, supplied with chips and reference designs by none other than Nvidia. There’s a very well-known phrase for that kind of situation that 3DFX led themselves into, but I can’t mention it here.
The Voodoo 3 (pictured above) was the company’s comeback in 1999, although it was pretty terrible. I own a working one and keep it in storage for the day that I can mount it on a wall. Its a tribute to how well people can screw up a good thing. The core was based on the already-terrible Banshee and it was not as powerful or as capable as its competitors. It was somewhat in line with the RIVA TNT2, but 3DFX and the rest of the world was later blindsided by the Geforce 256. It still lacked 32-bit colour support and couldn’t address many textures, falling very short of the acceptable performance standards. By the time 3DFX could start working on its successor, the Geforce 256 took the performance crown and ran away with it, ATi chasing a close second.
By the time the Voodoo 4 and 5 arrived in 2000, it was powerful enough to counter the Geforce 256, but Nvidia and ATi were already miles away with the Geforce 2 and Radeon DDR. The Voodoo 4 was SLI-capable and was competitive with Geforce 2 when in SLI, but the price arguably offset any performance benefits to be had. The last product from 3DFX to the consumer market was the Voodoo 5-5500, which had better anti-aliasing capabilities, but suffered because its dual-GPU design was too power-hungry and couldn’t keep up with the competition.
The swansong of 3DFX was the quad-core, 128MB Voodoo 5-6000. It was never released to the public and every single one of the thousand-plus examples out there are engineering units in various stages of completion. The Voodoo 5-6000 was found by reviewers to be much faster than the Geforce 2 Ultra and the Radeon 7500. It was even able to trail the Geforce 3 cards by a small margin and had this been released at an affordable price, it would have brought the company back into fame and possibly a longer future. Unfortunately, it was rather expensive to produce and used more power than was necessary. Despite its large muscles, it also didn’t fully support DirectX 7, let along 8. It was a project that should have been shelved within the first week of development.
In late 2000, 3DFX’s creditors launched a bankruptcy application against the company. Feeling a little generous and tickled by the irony, Nvidia bought up the company and only completed the acquisition proper in 2009. All of 3DFX’s assets, employees and intellectual property are now owned by Nvidia and several of its ideas have been used in products we see and use today – SLI and multi-GPU designs are just a few of the things that Nvidia has launched that began as 3DFX technologies. For all intents and purposes, they’re now one and the same company.
So what happened to the founders? Gary Tarolli and Alex Lupp, one of the chief hardware designers, are now both employed as engineers at Nvidia and Gary in particular has been involved in the development and demonstration for a few of the new technologies the company has been working on in the last few years. Ross Smith resigned from 3DFX before the sale began and began working at Quantum3D. Whether he’s still there, or still alive, is unknown.
Scott Sellers was initially outspoken against the sale and went into early retirement. He later resurfaced and began another company called Azul Systems, using the Java runtime on the Linux platform to develop stock trading systems that are now used by Wall Street. He also writes a few blurbs for ComputerWorld, although he mostly deals with money more than computers.
As for 3DFX, it remains well-dug into the hearts of enthusiasts who still dedicate time and money to get the older systems working again. Some of the company’s fans transferred to Nvidia, others have gone to AMD/ATi. In spite of it being gone, it was probably the one company that had the biggest impact in the GPU market, pushing its competitors to do better 3D acceleration and even perfecting 2D with the Voodoo 3. We owe things as they are today to 3DFX’s input and that’s why it’s today’s Oldie But Goodie.
Discuss this in the forums: Linky