A short story of 3dfx

Each Founder sees his project as the best and his actions as the best for the development of the company. This combination leads to the creation of great companies and the creation of personalities admired by mere mortals; these are cases that are role models. What tends to be forgotten is the second group, a much larger one, which, through their blindness, hubris and misguided decisions, led themselves to collapse. Today, I will briefly describe the story of 3dfx, a manufacturer of graphics cards and accelerators that, in the space of a few years, went from being the creator of a revolutionary solution that changed the world, to bankruptcy.

PC GPU Market share

Graphics cards are a much more important component than most think. These devices are crucial for good gaming, cryptocurrency mining, graphics rendering and many different activities. Increasing demands mean that more and more users are starting to use dedicated solutions instead of those integrated with the CPU. This is a kind of return to the revolution sparked by the protagonist of this article.

According to Statista, in IQ 2022 the GPU market is dominated by three companies: NVIDIA, AMD and Intel. Intel is the largest GPU manufacturer, with around 60 per cent market share. Intel's high share of the GPU market is due to the fact that iGPUs (integrated graphics cards) are included in the statistics. Intel's market share fell by 8 percentage points over the year. AMD and NVIDIA each hold around 20% market share. The discrete GPU market is a different story, with NVIDIA retaining 78% market share, AMD having 17% share and Intel 4%. Intel is attempting a stronger entry into the standalone GPU market with the introduction of Intel Arc chips at the end of Q1 2022. These processors are also limited to certain markets, so will not have a significant market share until the end of 2022. The apparent lack of competition is highly deceptive, as each GPU manufacturer now sells its solutions in several lines tailored to specific types of customers. In the late 1990s, there were a number of GPU manufacturers, but each made one product, so their success depended on that particular model.

3dfx Interactive

Founded in 1994 and based in San Jose, California, 3dfx Interactive specialised in the development of 3D graphics processors and, later, graphics cards. It was a pioneer in this field from the late 1990s until 2000. The name 3Dfx comes from the abbreviation of the words "three dimensional effects", which, as it were, says everything about the idea behind the venture, which brought completely new standards to the showrooms, being completely out of reach of the competition for a long time. It is worth mentioning that, 3Dfx was originally only spelled with a capital 'D', but in 1999, after the acquisition of STB, there was a logo refresh and a cosmetic change to 3dfx.

In 1996, the company released the first device designed solely for rendering 3D graphics on a computer screen: a 3D accelerator. The product was called Voodoo Graphics, 3Dfxlogo, and consisted of a PCI card equipped with a 3D graphics processor (now known as a GPU) running at 50 MHz and 4 MB of DRAM. The company also created a dedicated API called Glide that developers could use to interact with the card and exploit its capabilities. Glide was originally developed as a subset of the industry-standard 3D OpenGL graphics library, focusing specifically on the functionality required for game development. Another key difference between Glide and other 3D graphics APIs was that the functions exposed in Glide were implemented directly using native GPU instructions in Voodoo Graphics. In other words, while OpenGL, and later Microsoft's Direct3D, provided a common set of APIs regardless of the specific graphics hardware that actually processed the instructions, Glide only provided GPU-supported functions.

This approach gave all 3Dfx cards superior performance in graphics processing, a key advantage that persisted for many years, even when competing solutions from Matrox, ATI and Nvidia entered the market. It also resulted in some serious limitations in image quality, such as a maximum resolution of 640×480 (later increased to 800×600 with cards called Voodoo 2) and support for 16-bit images. Being the first consumer hardware with dedicated 3D graphics on the market, 3Dfx completely revolutionised the PC gaming space, setting a new standard for how 3D games could and should look. All new games developed from the mid-1990s through to 2000 were optimised to run on the Glide, allowing 3Dfx owners to enjoy stunning and smooth 3D graphics at a level unattainable by the competition.

The market leadership position was very short-lived - after the launch of the Voodoo graphics card and several successors such as Voodoo2 (1998), Voodoo Banshee (1998) and Voodoo3 (1999), 3Dfx was overtaken by two powerful competitors: Nvidia's GeForce and ATI's Radeon. A series of bad strategic decisions led to 3Dfx's loss of market share and eventually to bankruptcy in late 2000. The simplest way to put it is that the demise of 3dfx is the consequence of a number of factors, but the key ones were hubris, short-sightedness, not understanding the customer perspective, falling in love with their own technology and betting everything on one card. Below I will describe the elements that led to 3dfx's downfall.

Reason 1: Failed expansion into new verticals

The first major failure in 3dfx's history came in 1997 with its unsuccessful collaboration with SEGA to develop the Dreamcast console to compete with Sony's PlayStation 2.  In the first quarter of 1997, SEGA proposed to 3Dfx to develop a graphics chip for the Dreamcast console, while also collaborating on a chip called 'Katana', developed in Japan using NEC and VideoLogic technologies.

On 22 July 1997, Sega announced that it was terminating its contract with 3dfx , and chose NEC's PowerVR chipset for its games console. Sega informed 3Dfx that it was not terminating the contract due to the performance of the chipset and did not give reasons for its decision to terminate the contract, additionally SEGA did not want to give up the know-how used in the development process.  SEGA's decision to choose NEC's solution may have been influenced by a number of factors, including NEC's proven experience in supplying chipsets for the Nintendo 64 and its demonstrated ability to cope with a large influx of production capacity should the company decide to increase production significantly. 3Dfx sued SEGA for breach of contract, accusing SEGA of initiating a bad faith transaction to acquire 3Dfx's technology. The case was resolved through mediation.

Reason 2: Overconfidence in creating market development

Another of the elements that led to the demise of 3dfx was the arrogance and belief in being the paradigm setter for development. 1998 was the most important year in the industry's history, with ATI releasing the Rage 128 and Nvidia's Riva TNT chip, which could already compete with 3dfx. The rivalry with Nvidia was particularly fierce, with 3dfx suing Nvidia for copyright infringement of its multitexturing patents. The lawsuits, after a flood of counterclaims from each other, were not resolved until the end of 3dfx's existence.

3dfx announced two further products in late 1998: Voodoo 3 and Voodoo Banshee. The design used a tried-and-tested patent - a combination of a new 2D core and the slightly higher-clocked Voodoo 2, which, however, had no texturing unit. This made the Voodoo Banshee a card that was significantly slower than the Voodoo 2 in games that supported multitexturing, but offered slightly better performance thanks to its higher clock frequency. The 2D graphics rendering part was much improved over the Voodoo Rush and was on a par with the best cards from Matrox, ATI or Nvidia. Despite this, the Voodoo Banshee became 3dfx's least popular product (1 million units sold by January 1999).

Voodoo Banshee's poor sales were due to the lack of support for a 32-bit colour palette at the time of the RIVA TNT from Nvidia. Well, 3dfx believed that users would not use 32-bit graphics because it would cause the frame rate (FPS) to drop below 60, while Nvidia's overarching goal was to overcome this limitation. The 3dfx cards were primarily add-on cards for 3D. Their solution was provided by SLI (Scan Line Interleave). This allowed one VooDoo to render 2D graphics and the other 3D, which gave really good performance. The second reason for 3dfx's growing problems, was the move away from rendering games in Glide (3dfx's proprietary API), to the use of DirectX and OpenGL by developers over time.

Reason 3: Unsuccessful acquisitions

Mergers and acquisitions are used, in theory, to increase the value of a company by entering new markets, acquiring important technologies, reducing costs, etc. It's very simple in theory, but sometimes M&A leads to a lot of havoc. At the time, Nvidia had yet to launch a solution that sold as well as 3dfx's Voodoo line. Nvidia's RIVA TNT was a highly integrated product that had two main advantages: faster 3D speed and 32-bit 3D colour support. 3dfx, on the other hand, had very limited OEM sales, as Banshee was only adopted by OEMs in small numbers.

Just before the release of Voodoo3, 3dfx made a major change in strategy, buying STB Systems on 14 December 1998 for $141 million. STB Systems was one of the larger graphics card manufacturers at the time; 3dfx's intention was to start manufacturing, marketing and selling its own graphics cards, rather than operating solely as an OEM supplier. Nvidia has decided not to acquire STB and continue to support multiple brands of graphics card manufacturers. The decision to go into chip production plunged not only 3dfx but also other players in the market (Diamond, Creative, Quantum) paving Nvidia's place at the top. The decision to acquire STB had a tragic impact on 3dfx for several reasons:

1. STB's manufacturing facility in Juarez, Mexico, was unable to compete on cost and quality with the ODM (Original Design Manufacturer) and CEM (Contract Electronic Manufacturer) manufacturers that supplied solutions to Nvidia in Asia.
2. STB's poor reputation caused some of 3dfx's OEMs to warn the company that any STB products did not meet quality standards and to stop working with the company.
3. Chip development by 3dfx after the acquisition cut off an important source of revenue for the company. Existing OEM customers became competitors and began sourcing graphics chips from Nvidia.
4. A requirement in the OEM industry was the ability to consistently produce new products within the six-month cycle required by PC manufacturers; 3dfx did not have the methodology or mindset to operate in this business model.

Reason 4: Protracted R&D processes

The turn of the century was a time when each generation of graphics cards was a revolution in the market. The problems associated with the STB acquisition took a toll on 3dfx's product development pace, while the development of the first card in collaboration with STB (the Voodoo 3) went smoothly, the development of its successor was an ordeal.

The Napalm card was supposed to be Nvidia's answer to the RIVA TNT 2, but the protracted R&D process meant that Nvidia had already introduced a next-generation card offering better 3D graphics performance using Microsoft's new Direct3D 7 library interface. 3dfx felt that Napalm, in its current form, was inferior to Nvidia's new chip, so redesigned it to enable multi-processor cards (VSA, Voodoo Scrabble Architecture). The resulting graphics chip was called the VSA 100 and was supposed to be able to outperform the GeForce 256, 3dfx at the time was not selling graphics chips to other manufacturers, which forced them to solve this problem. Hoping to increase production capacity and access new markets, 3dfx bought out Gigapixel in March 2000. 3dfx's next product (the Voodoo 5) was already significantly inferior to the competition in both quality and price.

The second acquisition was related to a desperate attempt to accelerate the development of the 'Rampage' card by acquiring GigaPixel for $186 million at the end of March 2000. The move left the company stripped of its financial reserves and bankrupt six months later, surrounded by the collapse of the dotcom bubble.

Reason 5: Establishing a market benchmark

Often a fast-growing market requires a degree of standardisation, favouring certain players in the market - such was the case with 3dfx. In 1999, the 3DMark 2000 graphics card benchmark was published, which made fairly heavy use of the Transform & Lighting module for evaluation, the absence of which made the Voodoo 5 card appear weaker than it actually was. The T&L solution itself was not yet supported in games at the time. The poor performance in 3dfx benchmarks was attempted to make up for by a feature offered in the Voodoo drivers called Geometry Assist. It was intended to enable the Transform & Lighting module in the software. At the time it was at a fairly early stage and caused a lot of compatibility issues, so it was disabled by default. Some of the games that worked with Geometry Assist ran faster on the Voodoo 5 5500 cards than on the GeForce 2. For gamers the intricacies of the 3DMark system didn't matter, all that mattered was the score, which in the case of the 3dfx was significantly lower than competing solutions.

Summary:

At the end of 2000, shortly after the release of Voodoo 4, several of 3dfx's creditors decided to initiate bankruptcy proceedings. 3dfx, as a whole, had virtually no chance of successfully contesting these proceedings and decided to sell its assets to Nvidia for $70 million, effectively ending its operations. The 3dfx story shows that even gaining market leadership does not guarantee a sense of security even in the medium term. Company managers need to be constantly aware of risks, mitigate them and update their risk profiles with new threats that emerge during the course of business. Without a continuous risk management process, even those who take the emerging market by storm are likely to fall quickly.

^Sources:

^{https://tedium.co/2018/02/14/3dfx-history-failure/
www.statista.com
https://www.liquisearch.com/3dfx_interactive/dreamcast
https://megakemp.com/2009/01/27/the-rise-and-fall-of-3dfx/
https://www.sharkyforums.com/showthread.php?72576-An-excellent-commentary-on-what-happened-to-3Dfx
https://wccftech.com/nvidia-retained-80-discrete-gpu-market-share-amd-20-in-q2-2022-despite-gaming-revenue-losses/
http://www.x86-secret.com/articles/divers/v5-6000/letter_to_customers.htm}