Age of Nvidia

Keep the gamers happy, and the world is yours: How one 3-D graphics company shrugged off a recession and vanquished every foe.

Published May 15, 2002 7:30PM (EDT)

The players at Next Game rarely look back. As they man the computers at this game center in San Francisco's Japantown, they're too busy concentrating on what's ahead. Thunderous percussions of grenades and automatic-weapons fire jostle out of speakers as the gamers sit before oversize monitors, hands flying over keyboards and mice, playing round after round of Counter-Strike or Medal of Honor.

What's ahead could be an enemy popping into sight on-screen or the newest game of the moment or a tweaked hardware component that might squeeze out a crucial performance enhancement -- like a precious extra frame per second (fps) of computer video, enough to give the player a virtual reaction-time boost that could be the difference between life and death.

These young men are only renting the Next Game machines, but they represent a larger community of hardcore gamers whose passions drive an entire industry. Even as major PC manufacturers such as Gateway and HP struggle with lowered earnings expectations, razor-thin profit margins and sluggish growth, one sector of sales remains healthy. The high-end gaming PC is in such steady demand that it alone keeps specialized manufacturers such as Alienware, Falcon Northwest, Voodoo Computers and others healthy. The rigs produced by these companies, costing upward of $6,000, are tricked out with fast processors, big monitors, artistic cases and the latest "3-D accelerator" technology -- these days a GeForce 4 chip from Nvidia or a Radeon 8500 from ATI.

They're the fast and furious of the PC world. And their specs are a moving target. A true gamer might replace part or all of his setup each year simply to play the newest games, games that each season further push the limits of what a computer can do. Just try running this year's leading-edge entertainment on a computer from a year or two ago: In addition to getting suddenly sluggish frame rates, you'll also miss realistically rippling water effects, the glint off an opponent's helmet, or smooth, jag-free edges. To the true believers, yesterday's tech is less than useless. It's an insult.

On the other hand -- to the dismay of the PC industry -- the vast majority of productive users have been realizing for several years that they don't really need a faster, newer computer just to run Microsoft Office. Though both Apple and Microsoft are scrambling to build new graphics features directly into their operating systems, such as translucent widgets and drop shadows that will drag current computers to a crawl, PC manufacturers are facing a hideous fact: Most buyers find that their two-year-old machine runs just fine, thank you.

Couple that with a shaky economy prompting businesses to hold off on capital improvements, and you're looking at trouble for the consumer PC industry. But not where gamers are concerned. The market for 3-D accelerator cards alone -- chips specifically designed to make computer graphics snap, crackle and pop -- is worth almost $10 billion a year in sales to consumers and computer manufacturers.

These chips aren't meant solely for testosterone- and Mountain Dew-fueled boys or lonely men spending hours in their custom-designed flight chairs, mastering the latest jet-fighter sim. Computer gaming pervades modern culture, threatening to unseat or overwhelm even Hollywood as an arbiter of cool. As a nexus of recreation, entertainment and computers, gaming has become a focal point for social concerns and entrepreneurial energy. The companies that make 3-D accelerators are feeding the dreams of a generation that seeks virtual reality entertainment on an ever more immersive, ever more addictive scale.

It's a big market, and it's getting bigger all the time, but it is also absurdly fast-moving and turbulent. And though the companies that make graphics chips increasingly serve a mass audience, their own profitability is pinned to keeping hardcore gamers satisfied. And that is no easy task. Gamers are fickle -- their loyalty is not to brand but to performance. Indeed, one company, 3Dfx, which played a pivotal role in building this mighty market, was itself fragged barely two years after dominating the industry.

In 1996 3Dfx Interactive virtually created the demand for consumer-level 3-D acceleration hardware. Within two years it was king of the hill, with a capitalization of over $200 million. Yet by the end of 2000, the company was bankrupt and sold for parts. It's buyer, Nvidia, meanwhile went from nowheresville to defining the legitimacy of the market. In late 2001 Nvidia was named to the S&P 500, replacing former high-flier Enron. While all around it the mightiest tech companies were falling victim to the dot-com bust, the telecom market collapse and the ensuing recession, Nvidia surged from strength to strength. In late April, when as a result of an SEC investigation Nvidia had to restate several years of earnings, the company ended up reporting that it had earned more profits than previously indicated.

Does that mean catering to gamers is a foolproof business model? Not necessarily, or 3Dfx wouldn't have stumbled while Nvidia leaped forward. What, then, explains Nvidia's success?

Cynics look at Nvidia's close relationship with Microsoft; the company was one of the first graphics chip makers to endorse Microsoft's software protocols for technical gaming support, and it was also chosen to manufacture the Xbox graphics processor chip. But Nvidia's success is as much a tale of 3Dfx mistakes as it is of canny Nvidia prowess. Nvidia proved able to move with the agility necessary to keep up in a market keyed to hardcore gamer needs. 3Dfx, in contrast, promised too much and then failed to deliver.

Together, the contrasting fortunes of Nvidia and 3Dfx help fill in the background to an explosive period during which computer gaming rose steadily in cultural and economic prominence. The story of Nvidia's rise, and rise, demonstrates how great the rewards can be for feeding the voracious maw of gamers who are always desperate for what's better and faster and flat-out cooler. But the moral of 3Dfx's decline, and fall, is that in this world of bleeding-edge technology the mighty can self-immolate at any moment. All it takes is another frame per second for another itchy trigger finger.

Back in the early 1990s, when many of those now making games were in their teens, 3-D was mainly the provenance of specialized arcade machines and high-end workstations. It wasn't uncommon for home-computer gamers to be treated to lush 3-D scenes for things like introductory movies and then get booted back to Donkey Kong-like graphics for the actual gameplay. In 1993, in fact, 15-bit "high" color was state of the art, and the hardware focus was on video cards that could pump out enough individual pixels -- the dots that are the building blocks of 2-D graphics (think Lite-Brite) -- simply to bring color to that flat world. To be fair, this wasn't just a concern of gamers; burgeoning applications such as desktop publishing and computer graphics saw more colors and higher resolutions as a boon, too.

But true-blue gamers wanted to bring home what they saw (and lost innumerable quarters to) in the arcades. Those games at first presented the 3-D world as a construct of green-on-black vectors, as in the original Star Wars and Battlezone uprights. Then came the color Virtua Racing and Virtua Fighter from Sega, both of which presented an exciting hint that programmers could attain their dream of creating lively interactive worlds that might actually look real. The machines that drove those worlds, however, were custom-designed and expensive. Unless people were willing to shell out a few thousand dollars for a machine that could run just one game, a different solution would be needed for home gaming.

There was absolutely no consensus among 3-D technologists at the time; the goal of creating more realistic games was simple but no one quite knew how to achieve it. Call it anarchic or innovative, but during this period a flurry of 3-D techniques with Lewis Carroll-like names -- voxels and surfels, ray-tracing and tiling -- were all claimed as the holy grail at one point or another.

This experimentation (or confusion) was exemplified by the different tacks taken by two companies that sprang up, one in 1993 (Nvidia) and one in 1994 (3Dfx). The latter was founded by the arcade fan, engineer and amateur game programmer Scott Sellers along with Gary Tarolli and Ross Smith, all of whom had developed graphics hardware for the soon defunct company MediaVision. Nvidia's founders likewise had engineering backgrounds -- Jen-Hsun Huang had worked at LSI Logic's "system on a chip" project; Chris Malachowsky had been a senior engineer at Sun Microsystems; and Curtis Priem, who also came from Sun, had been instrumental in creating the first graphics chips for the original IBM PC.

Some longtime industry observers say, in retrospect, that Nvidia's first offering in the 3-D acceleration market was ahead of its time. As usual, that roughly translates to "It was a flop." Or, as Brian Hook, a 3Dfx lead programmer in 1996 who now runs his own game company puts it, "Nvidia nearly died with their first product."

For that first product, the NV1 processor, Nvidia took a leap of faith and chose to rely on "quadratic texture maps" as a way of describing 3-D objects. Instead of breaking a 3-D object into straight lines and building them up from triangles (polygons), "quads" define an object by breaking it into curved surfaces. It is, theoretically, a more advanced method that could provide a smoother look than polygons.

In contrast, 3Dfx didn't go about reinventing the wheel. It stuck with what seemed to be the most popular, if not the most powerful, solution -- polygons. But 3Dfx did contribute something important of its own. In the mid '90s, PC gaming lacked a standard set of 3-D APIs (application programming interfaces). An API is a specific set of commands that enable a software program to work with an operating system -- in this case, the 3-D APIs might tell the computer to draw a line somewhere, or to move it vertically. 3Dfx wrote its own APIs, reasoning that this would give all programmers a standardized set of ways to describe the images they wanted to build on-screen. They called these proprietary APIs Glide; it was both what led 3Dfx to the top and what played the biggest role in its downfall.

Glide allowed game developers to write accelerated versions of their games for computers sporting 3-D cards equipped with 3Dfx's new-for-1996 Voodoo chip set. In so doing, it set the stage for a great gaming leap forward.

"Glide was necessary early on because -- and this is something that many people forget -- [Microsoft's own API] Direct3D didn't exist when Glide was released," says Hook. "It wasn't until after Glide was released that Direct3D finally arrived -- and the first iteration was abysmal."

But Glide's superiority over Microsoft's offering was far from a deal clincher. When 3Dfx first began courting game developers, asking them to write games that would take advantage of 3Dfx chips, the company met with resistance. The developers not only saw more work in coding an extra version of their games for the as yet nonexistent owners of as yet nonexistent 3Dfx cards, but they also felt the card's projected price of about $400 would severely limit the market.

The high price also caused problems when 3Dfx tried to get third parties to manufacture cards based on 3Dfx's generic, or "reference," design. (This handing-off to a manufacturer was standard practice among most video chip makers; later, when deviating from this paradigm, 3Dfx would make a costly gamble.) For consumers, the expense would be even greater, since computers equipped with a 3Dfx product would still need a 2-D graphics card -- and this limitation remained until very late in the game for 3Dfx, long after competitors had released successful 2-D/3-D cards.

Hook says that at the time 3Dfx estimated it would sell about a thousand units a month if it was lucky. The rest of the company's revenue, 3Dfx thought, was going to have to come from selling its chips to coin-op manufacturers, who could then make easy versions of its games for personal computers.

Then came 3Dfx's lucky break. The price for memory chips -- RAM -- fell through the floor in 1996. The expected retail price for a Voodoo card dropped by almost $100, smashing through a psychological barrier. Manufacturers lined up to build cards with Voodoo chips, and game developers such as Core Designs turned out Glide-specific versions of games that stunned the gaming press from the fan boys to the jaded old coots.

Core's Tomb Raider, with the curvaceous Lara Croft, became a worldwide model of the killer app for the new 3-D tech. PC owners without acceleration could still run the game, but they would see flat and opaque ponds rather than the Glide version's translucent, shimmering water; their scenes would not have colored lighting; and, not least, the lower polygon count in unaccelerated versions meant that Ms. Croft's greatest assets appeared a bit ... pointy.

Another coup came when id Software's game god, John Carmack, built a software compatibility layer that enabled 3Dfx cards to accelerate a version of id's blockbuster game Quake. Thanks to Carmack's reputation and the previous success of Doom and Doom II, Quake was a must-have, and it set a new standard for how a game should look; it alone prompted unknown numbers of gamers to shell out for a 3Dfx card. The company was on a roll.

As Thomas Monk, a writer for the Web site Accelenation, puts it: "In 1996 3Dfx produced a great product with fantastic image quality and performance. The transformation of Tomb Raider was evident for all to see and the Voodoo chip set was the most eagerly anticipated that year." Other game developers jumped on the Glide wagon, writing effects-heavy games that created a demand for the Voodoo card, which, in turn, created a demand for more and more advanced-looking games. The cycle had begun.

While 3Dfx was revolutionizing the field, Nvidia was learning how to play nice with Microsoft.

"In 1995 Microsoft decided to roll its own 3-D APIs," recalls Alex St. John, a former Microsoft developer who was one of the leaders of the effort to make the company a player in gaming graphics.

In true Microsoft fashion, the company "rolled its own" by purchasing another company -- in this case, the British firm Rendermorphics, which had developed a set of graphics tools based on the polygon approach. Then Microsoft began approaching graphics hardware manufacturers, including Nvidia.

At the time, Nvidia had just made a deal with Sega (which would later fall through) for providing graphics acceleration chips for what would become the Dreamcast console. Since Sega and Nvidia were both exploring quadratic textures, it seemed like a good match.

But Microsoft, which had just purchased a polygon-based technology, respectfully disagreed.

"When the first [quadratic] products went out," St. John says, "it was well known that Microsoft didn't support them." And so, he says, Microsoft put its considerable marketing power behind other companies, such as Creative, that were building accelerator cards using 3Dfx chips. Even though 3Dfx's Glide was nominally competing with the Microsoft Direct3D API, the more high-end and accelerated games that showed up for Windows, the more desirable a new Windows-based PC would be.

St. John says he remembers making the point bluntly to Nvidia representatives, telling them, "You guys need to learn triangles [polygons] like everyone else."

Nvidia took the advice to heart: "They got really religious," St. John says. Nvidia began working on polygons.

Though having Microsoft on your side has rarely hurt a company, this was far from an anointing for Nvidia. The company still didn't have a commercially successful hardware product; Microsoft's own 3-D technology was clunky at best and drove developers away with buggy and difficult code. 3Dfx and Glide were still the darlings of a cabal composed of developers, hardcore gamers and the press.

It was a step, to be sure, and one that helped set the stage for the friendship that would eventually lead to Nvidia's becoming the graphics-chip supplier for the Xbox. But Nvidia needed more than a helping hand from Redmond to start making its ascent.

And of course 3Dfx and Nvidia weren't even the only players in the market -- that is, if Nvidia could even be counted as a competitor at that point. By early 1997, companies such as Matrox and Diamond had their own products, some of which were even competitive with the Voodoo. The Voodoo name, plus the established Glide API, still held a solid lead in the field. But ATI's Rage Pro 3-D card, which was catching on thanks to its ability to accelerate both 2-D and 3-D graphics, was a strong second.

But by 1997, Microsoft's influence with developers was already beginning to make its own Direct3D API (soon to be rebranded "DirectX") a serious challenge to Glide. And Nvidia wasn't going to make the same mistake twice.

According to Monk, "Nvidia had learned from their initial mistake regarding Direct3D and decided to adopt it wholeheartedly ... This was not so critical while D3D was feeble, but eventually Microsoft would get their act together, and once developers had gained a greater confidence with Direct3D, they no longer felt it necessary" to support Glide versions of their games.

For gamers, Microsoft's influence was increasingly noticeable. Purchasers of a new game would install it, learn that their system required an update of Microsoft's gaming APIs, and then wait while the upgrade was installed.

And then ooh and ah at the wonders of new technology. The late '90s were heady years, not just for stock market speculators, start-up entrepreneurs and a world discovering the joys of online life, but also for a surging crowd of PC gamers. Games like 1997's Total Annihilation and Diablo, 1998's Starcraft and Quake II, and a host of others improved technically by leaps and bounds over what had been released the year before. The startling pace of change for the software APIs, the hardware chips and the games themselves was just one more proof that time, in the Internet era, ran faster than ever before.

Later, at the turn of the millennium, time slowed again for most of the tech industry. The economy began to stutter and air exploded from the dot-com balloon. But for gamers and the gaming biz there was no break. The pace remained breakneck, and only the nimblest would be able to keep up.

First of two parts. Read Part 2.


By Daniel Drew Turner

Writer and editor Daniel Drew Turner covers technology, design and politics.

MORE FROM Daniel Drew Turner


Related Topics ------------------------------------------

Gaming