NVIDIA CEO, President, and co-founder Jen-Hsun Huang sips a Diet Mountain Dew (the "drink of CEOs of the 21st century") after spending a good part of the day overseeing company presentations to industry analysts. We are catching him between an interview for The Wall Street Journal and a nice, quiet dinner with about 50 international technology journalists flown in for the next day's briefing on NVIDIA's newest graphics chip, NV40. Complicating his image as a tenacious competitor, he comes across as relaxed, occasionally even goofy, with a quiet but infectious enthusiasm. It quickly becomes obvious that he loves his job.

"It has never occurred to me that I shouldn't be the CEO of the company. It occurs to me often that I wish I could do a better job for the company, but I also believe deep in my heart that I'm the best person for the job and I tell myself constantly that I need to keep reaching out and keep learning and keep expanding and keep growing as a CEO or someday I wouldn't be the right person for this job. It is really just a learning process for all CEOs. Nobody was trained to be a CEO. Nobody deserves to be a CEO. Nobody, quite frankly, went to school to be a CEO. CEO is a lot [about] skills but it's a lot [about] leadership, talent, instinct, gut feel, courage; [it's] a lot [about] loneliness. I don't think that there is anything that would train you for it. I think you just have to keep growing as a person."

With only a branching shader for modesty, NVIDIA's Nalu character floats effortlessly while various lighting techniques play with her delicately rendered hair.

It's hard to pinpoint when an individual's personality fully develops, but Taiwanese-born Huang says he was pretty far along when he and his brother arrived at Oneida Baptist Institute, the Kentucky boarding school his aunt and uncle had mistaken for a prep school. "As a kid, I was very focused, very determined, very interested in academics and sports. When I went to Kentucky, in a way I was too young to know that it was a reform school, although I wasn't confused about the fact that the kids there were really tough." He was 10 years old. As the school's youngest student, things could have gotten ugly, but Huang's 17-year-old roommate, newly released from custody and sporting a few stab wounds, watched out for him. In turn, Huang taught him to read. "It never occurred to me that that was difficult. It occurred to me that it was exciting and interesting and people treated me well . . . I learned a lot and it taught me a lot of independence. My older brother and I were away from our parents for almost two years." To stay in touch, Huang's parents would send audiotaped messages, which the brothers would record over and send back. Perhaps that is the origin of Huang's interest in technology and its potential to facilitate visceral, emotional experiences. Wherever it came from, he firmly believes that orientation has always set NVIDIA apart from other semiconductor companies.

"When we started our company, the majority of the world's semiconductor companies were founded by semiconductor industry veterans. They built factories and they manufactured chips for the industry. We had the perspective that we were a unique systems and computer company in the sense that the type of products we build deliver an experience; that content, art, and applications are at the core of it and that our technology is really a facilitator of it. We always had the perspective that delivering the experience was our ultimate mission."

Yet, NVIDIA's road to success was not necessarily a predictable one. "I never thought of myself as a person who was going to start a lot of businesses. I know I'm definitely not a serial entrepreneur." Huang admits, "There's not an ounce of me that wants to start another company. Not because it's too hard, not because I don't think there are other opportunities, but because that's just not my style. My style is to believe in something, give it everything I have until it doesn't want me any more, for whatever reason. If it weren't for a couple of people I really valued and really believed in, [NVIDIA co-founders] Chris Malachowsky and Curtis [Priem], and the circumstances that surrounded 1993, I don't think I would have started a company. I would have stayed at LSI Logic. I would have worked there until I was running the company. I'm an entrepreneur in the sense that I always have ideas. That's the part of me that is always thinking . . . about ways to do something better, ways to kick it up a notch. I'm always looking for new ideas and I'm never satisfied." This goes beyond grace under pressure and appears to permeate the entire company. "I always believe that the best ideas come out during the worst times, that the most inspired ideas and the most inspired work come during the worst adversity. It is just a fundamental belief of mine. I think best under pressure. I work best under pressure. I am most fired up when we are flat [on] our back, and I always see a silver lining. I really pride myself on being balanced. Things aren't horrible; things aren't perfect. But if you're not focused and if you don't give it everything you have, it's not going to just come your way naturally."

Huang's easy confidence comes off without egotism. "I surround myself with really smart people, making sure that I don't become so hung up on my own ideas that I don't give them the opportunity to show me the right answers, and making sure that [I] keep my guard down so that they're willing to come and tell you when things aren't going well. If you just practice those few skills, and it's counter-CEO, because you're supposed to be the super-smart guy and you're supposed to be the know-it-all guy, but if you can fight that off, you can learn amazing things and people help you, they protect you, they protect the company, and they make you successful.

"There's no secret sauce for having great ideas; the secret sauce for having great ideas is trying many of them. That's the secret sauce."

In The Beginning . . .

It speaks volumes about the man that in co-founder and Vice President of Hardware Engineering Chris Malachowsky's official corporate portrait, he is the only NVIDIA executive not wearing a necktie; it didn't occur to him to wear one to the photo shoot, and when they asked if he wanted one, he declined. Extremely well-liked by NVIDIA support staff, he is notorious for his Halloween costumes. (Last year he donned black leather chaps and spiked hair as the ne'r-do-well brother of NVIDIA icons Dawn and Dusk. A snapshot of him in full regalia is still on his official NVIDIA ID pass.) "My objective in life had nothing to do with business. It was just to do interesting stuff and build products people liked." Lest you think he is not a serious person, his desktop motto is this: Success is not so difficult—just bite off more than you can chew and then go out and do it.

According to Malachowsky, the three NVIDIA founders developed a relationship back when he and Curtis Priem (who retired from NVIDIA in 2003) were working at Sun Microsystems, and Jen-Hsun Huang was their silicon vendor at LSI, facilitating production and manufacturing. As the two Sun engineers were moving up in that organization, Huang was transitioning into a business role at LSI, licensing technology and cutting business deals. The original idea for the company came out of a cancelled Sun project. Priem and Malachowsky wanted to design a graphics controller to demonstrate the use of a new DRAM and license it back to the DRAM manufacturer, Samsung.

"Now that sounded like a good plan except that neither of us knew diddly about licensing or dealing with an international company from a business point of view, and we were kind of worried about that. Jen-Hsun was now in this business of licensing technology, and
the business end of the relationship grew from there," Malachowsky remembers. "So from day one Jen-Hsun was going to be the business guy. He was the only one of us with any business background, and it was clear he was the CEO of the operation. I was going to be the engineering guy and Curtis was going to
do architecture."

This fish-eye lens shot distorts NVIDIA's Santa Clara, Calif., headquarters considerably. The logo really isn't bigger than the moon.

Malachowsky sites the mutual respect for each founder's abilities and the clear delineation of their respective roles for the lack of infighting. "It wasn't that we didn't argue a lot about what were the right things to do and the right things to be focusing on, but in the end we would decide. If it was architecture, Curtis would decide. If it was business or strategy, Jen-Hsun would decide. If it was implementation, I would decide and we would go on about our lives. We would trust, even if we didn't agree."

Cutting-Edge Tools For Bleeding-Edge Development

What worked early on for NVIDIA is no longer a viable way to make semiconductors. As Malachowsky puts it, "The days of designing chips just based on sweat and effort are gone." One crucial element of groundbreaking innovation now is being able to pioneer the use of new tools and methodologies, many of which are being developed and tested on the fly. "The chips I've worked on have always been the largest in the industry at the time I was working on them. Therefore the tools and approaches that were stable, well tested, and battle proven weren't appropriate. They didn't have the size, capacity, or performance, or they didn't address whatever the new technology was. After we get done plowing through a large design on a new technology, in subsequent years that's the technology that everyone will make use of. But at the time we're doing it, there are only a few companies out there on the bleeding edge." Thus, a mutually supportive partnership with independent tool developers becomes essential and Malachowsky is not shy about speaking out on behalf of these guys. About half of his NVIDIA-related Google search results are for speeches or endorsements he has made for various chip design/verification tool companies. "We want to be supportive of their efforts so that there is a financial gain for them in trying to keep up with us. Ultimately there is nothing terribly proprietary about these tools. Somebody has to do the effort. It makes no sense for us to develop them ourselves, even if we knew how."

But Malachowsky does not credit technology alone with NVIDIA's rise to dominance. "What I didn't appreciate prior to NVIDIA was the value of all the strategic stuff: the business and sales and marketing stuff." He always thought they had good technology. "Our engineering team was second to none. But if you hang your hat on just one thing, and the typical thing is the product, [that] is not sufficient in this market for business success."

". . . we can be one of the most influential technology companies in the world."

Core Business & Corporate Survival

Huang's passion for graphics only enhances his view of the larger picture, "Our core business started out being graphics, and 3D graphics was very important to me and to the company, but at the core of it, I've always seen NVIDIA in a very simplistic way: I want us to be and I believe we can be one of the most influential technology companies in the world. If you say that when you are zero billion dollars large and three people large, it sounds kind of silly . . . but when you are 2,000 people large, it's starting to sound possible." This kind of talk may come across as corny (or worse) on the printed page, but having spoken with long-term NVIDIA employees who used to roll their eyes at this type of company cheerleading but now are seeing the vision too, it can be pretty compelling in person. More than once, the image of Bill Gates, spinning his vision for the PC industry back in the early 1980s, came to mind. "Being able to add value to the industry and being able to change people's lives has always been at the core of what we believed. Now, is it central around graphics? In a way, I hope so, because it is such a visceral, such an emotional technology, and actually makes a connection with you. So from that perspective, it's exciting. But I don't need it to be, because I think that is one of the reasons we called our company ‘NVIDIA' vs. ‘3-D-F-X.' I think NVIDIA can do a lot more than just graphics and we do a lot more than graphics today."

VP and co-founder Chris Malachowsky has a pilot's license and a cabana with multiple plumbing fixtures. Google does not lie.

Malachowsky's observations on corporate survival are even more to the point: "No one has been fired for having a product be late [or] misfunction. We ask everybody to work as hard as you can, use your best judgment, decide what you think is the right thing, and run with it. Make sure everybody knows where we're going and if it doesn't work out, be the first one to say, ‘I need help.' One of the things that we've enabled here is that every group is chartered with, is expected to, and is waiting to carry the day. We are considered a hardware company, but if the chip is late, well shoot, software better write a better driver, it better look fresh. If software can't eke out more performance, then maybe operations needs to get a chip build that's faster. If they can't do it, marketing better make up some feature and tout the crud out of it such that it seems fresh. If that doesn't work maybe sales has to come up with a new pricing strategy. . . . The tactics, the strategies, [of] how you relate to your customers, how you deal when the schedule is late, what do you do when the product costs too much, when it doesn't yield right, when competitive threats come from here and there, when the press is friendly or not friendly? Throwing up your hands is not the right answer, and one of the strengths we've always had is, the larger the crisis, the crisper the thinking. In general, most things can be turned around either with effort or insight or clever thinking. The way we've responded to calamities and challenges actually defined us and helped ensure that we survived another day."

This may never have been more evident than in the past 18 months, when the very late arrival and somewhat disappointing performance of NVIDIA's NV3x family of graphics chips kept engineers fixing, marketing folks spinning, and customers waiting. Huang's 2004 Editor's Day presentation focused primarily on the many lessons learned from NV3x, what has been carried forward into NV4x, and how NVIDIA has been transformed by the fallout.

Created to demonstrate high dynamic range lighting effects using 16-bit floating-point math, Timbury is NVIDIA's "first fantasy character." Apparently, pixies and mermaids are "real."

What Went Wrong

With NV3x behind him, Huang is more candid than he could afford to be during the last few product cycles. When NV30 didn't appear for Christmas 2002, despite assurances from Huang and others, it cost NVIDIA a lot more than revenue; it cost credibility. "We took a lot of flack for NV30. We made some pretty big mistakes, but we also did some really great things." When asked why analysts should believe his claims regarding the current NV40 chip, Huang simply replied that the current product is not eight months late.

The short list of NV3x technical problems isn't actually very short. Some are manufacturing-related, such as opting for a largely untested (for graphics) 0.13-micron process, which certainly played a large role in delayed time to market. Others stem from inappropriate technology choices, such as the big math machine VLIW architecture, which was a nifty idea but so unworkable in practice it was scrapped altogether for NV4x in favor of "wider is better" parallelism.

Many items, such as FP32 and the run-time compiler development, appear to have been mistimed. FP32 is definitely the way to go eventually, and even though Huang asserts strongly that it was the right choice for NV3x, it is a hard sell in light of how expensive it was (Huang says it cost NVIDIA a quarter of a billion dollars in operating profit last year) and how little developers could make use of it within a reasonable time frame. Had early communication with Microsoft been completely clear regarding DirectX 9 floating-point precision requirements, NVIDIA may have chosen the lower 24-bit precision ATI continues to employ. The run-time compiler, quite useful for a programmable part, was not developed simultaneously with the architecture, perhaps was not staffed with seasoned compiler programmers early enough, and was ultimately diverted into fixing specific programmability shortcomings of the architecture rather than focused on overall performance and ease of use.

On top of the technology challenges, there seems to have been a fundamental flaw in NVIDIA's project management method. Though not directly responsible for NV3x or 4x, as a veteran of earlier NVIDIA chip engineering cycles, Malachowsky explained that from the company's inception, there was never much in the way of hierarchy. "Whoever was responsible for the chip handled all aspects of the design and the verification and the implementation. He coordinated with the software guys, the board guys, and the systems guys. . . . There was one guy, and everything was under him." Over time, as projects became more complex, this approach apparently became difficult, even unworkable. As he observed in a presentation in June 2002, "Project scope has just about outstretched a single person's ability to understand it all and be enough of an expert to lead it appropriately." By then it would have been too late to completely revamp the management of project NV30.

Lessons Learned

The rightness of specific technology choices will fluctuate from project to project, and Huang has a long list of the good and bad outcomes, benefits, pitfalls, and hard-won lessons from the NV3x mistakes. When faced with lemons, he definitely comes across as a "make lemonade" kind of leader: unafraid to scrutinize errors, determined not to repeat them, and always looking for that silver lining. He says, "Two or three years ago, NVIDIA had the vision that programmable shading was going to change real-time graphics." Until that time, Huang describes, the technology of 3D graphics had been mostly about the evolution of military and flight simulators with everything based on geometry and textures, which, even if painted by a wonderfully talented artist, are created at a time before the game is being played. "Texture mapping, fundamentally, is a sophisticated decal, and there is only so much you can do with a decal. Programmable shading is the right direction because video games are at the intersection of science, technology, and art. It's a complicated thing. It's hard for an engineer to think like an artist, for an artist to think like a physicist. Yet in order to build a great video game, you need all of those skills. So what we've really done with programmable shading and programmability is put the artistic pallet in the hands of the content developers instead of capturing it on our chip. We're engineers. We can draw the polygons and [apply] texture maps, but beyond that, how would we know what [is] the next beautiful effect or the next artistic look?"

Huang's goal of having NV30 take 3D graphics to the next level fell somewhat short. "NV30 was designed to be programmable, but it was actually not a very programmable chip. It was great at legacy applications, which was kind of a weird thing. [Developers] just loved DX9, they went to it right away and that was good and [NVIDIA's high-level shading language] Cg pushed that along, but talk about irony. We told everybody, ‘This is the future and we ought to move to programmability.' And when the time came and we built our own chip, we went ‘Um! This didn't turn out the way we thought.' "

Yet Huang sites the unbalanced and non-orthogonal nature of the NV3x architecture and the difficulty developers had programming for it as the impetus for NVIDIA to write the real-time optimizing compiler. "Compilability will always imply an elegant architecture to program to but you need to invest deeply in compiler technology. Give me an example of a great microprocessor company who doesn't believe investing in compiler technology is necessary."

His emphasis on including forward-looking features is also quite emphatic, "If we want to continue to push content development to the next level, we need to make sure that we implement features that run at full limits, full rates, even though no games access them today. I have no idea how many times we were asked the question, ‘Sure, you've got these great features, but how many games are available today?' I now have found the perfect answer for that. I'm perfectly comfortable saying it, and I said it yesterday to a bunch of financial analysts: None! So sue me! If we don't implement the features, the games will never use them. It's back to the chicken or the egg issue again, and that's OK. If you want to be an innovator, if you want to push the market forward, you're going to have to put features in your chip that nobody is using today. If you're working closely with developers and the industry, those features will be used."

The Dustbuster. "Nobody likes to hear harsh criticism. But we have to hear it." With high chip temps a very real problem, NV30's variable speed cooling solution was crowned "The Dustbuster" by some reviewers. Huang jokes about his personal contribution. "I made a mistake during the launch of the first GeForce FX. I had really pushed us for a very powerful fan. I didn't actually ask [for] it to be loud; I asked for it to be very powerful. It turns out that powerful typically equals loud on fans. So here was my idea: It's called ‘Silent Running,' ironically, and depending on what applications you are running and depending on the workload, we would control the fan electronically and adjust the RPM, and it would dynamically apply the amount of thermal management on the chip depending on the workload. Well, it turned out that the fan was really loud. So when 3D turned on when you were playing a game, it got super loud, it was sufficiently annoying. Of course, everyone knew it was my idea, and not a single person would let me off the hook on it. They know that I don't mind when they tease me so they made a whole marketing video. It's hysterical." (Download the video at users.pandora.be/darkt/NVIDIAFlowFX_video.zip.)

The lemonade part quickly follows, "Well, look where that technology went. It turns out that on Apple G5, all the fans are electronically adjusted so that when they use demanding applications, the fan turns on more, and when they're just browsing the Web, it [goes] silent. More and more people are using this technology. We're using it now all over the place, in laptops, in some of our other desktop products. The point is that even good ideas can be taken to an extreme, but if you learn something from it and you adjust those ideas and refine them, [they] can become great ideas. So we don't mind laughing at ourselves, and it hasn't discouraged me from taking calculated risks and offering more ideas to the marketing guys, but sometimes one of them will say, ‘This sounds like one of those fan ideas from Jen-Hsun.'"

Huang summarizes, "So, we had a speed bump. But NV40, NV50, and NV60 will show very clearly we pointed the industry in the right direction. We didn't capitalize on it ourselves right away, but that's OK. We're going to capitalize on it now."

Moving Forward

Overcoming inertia is a challenge, change is hard work, and success breeds complacency, even for engineers. Malachowsky may have felt earlier than most that all was not well in NVIDIA-land. "When you are on top of the world as we'd been prior to the 3x generation, it's hard to make fundamental changes and improvements. You can tell someone to do better and they say, ‘Sure. Stocks are going up here, products are selling out over here, and everybody loves us. I'm not really that bad off. I'll deal with improving later.' But as soon as you feel challenged, then you can say, ‘Let's be introspective.' We're not perfect. Even the things we thought we were world-class at, there's another level. Let's go find it."

In order to survive, NVIDIA as a whole had to evolve past old habits and methods. "As in almost all of our experiences in the past, what could have killed us made us better." Huang admits, "NV30 changed the way we organized our company. That's how seriously we look at mistakes. We are willing to make fundamental changes at our core. So architecture is now done differently; we work with developers differently; we work with Microsoft differently; we engineer our chips differently; it's just really, really profound what we've done, and we did all that in a year."

"The 3x family challenges were exactly the impetus [needed] for change," explains Malachowsky, "Nothing about NV4x execution is an accident. The plans, the approaches, were all rather deliberate and aimed at improving anywhere we had a weakness. We've gone toward a much more traditional approach to project management. . . . We've now added some more structure. There is somebody responsible for verification, somebody responsible for implementation, somebody responsible for the different facets, and they really are responsible in that they have delivery schedules. [We have] team leads that are the technical drivers, but now we have more management hierarchy. Coming at it from where we were, we're still much lighter than the traditional engineer organization [but now] the guy at the top is not dealing with every issue simultaneously. . . . So the 4x family and its deployment, execution, architecture, planning for bring up, planning for first production, the software driver, the performance, all of that was deliberate. Not everything worked out exactly the way even those plans went, but we were proactive and thoughtful about just about every aspect having to do with taking that product from conception to the marketplace. Shipping first metal on a 200-something million transistor chip is pretty good, and the performance we're getting out of the box is really great."

Huang agrees, "We now have created an architecture that is much more programmable so that NV40 just runs beautifully out of the box. . . . We've made a machine that is so orthogonal, it just looks beautiful on paper, it's easy to program, developers know exactly how to program it and can almost predict what the behavior will be. . . . One of the good things that we learned on NV3X [is that] the architecture that shows up for the enthusiast or the workstation marketplace and the architecture that we offer you for the entry-level product that you can go down to 7-11 and get out of a Coke machine, that architecture ought to be identically the same. The only difference is performance. The features, the architecture, the programming model ought to be the same. Scalability is very important." This is a large leap from the GeForce4 MX days when previous generation products were branded as current generation technology, confusing customers and frustrating retailers (while giving CPU columnists a lot of carping material).

An especially exciting area of increased activity is the involvement of game developers earlier in the design process. "The first hundred NV40s that left our company did not go to OEM customers, did not go to our add-in card partners. The first hundred NV40s that left our company, not even an exception of one, not one came to my house, went to developers. The reason for that is very clear: This is about content." Huang said in his editor day presentation, "For the first time in our history, NV40 was developed at the architectural level with developers. We brought them into the architectural definition phase. We also learned that we need to provide tools to the developers so they can figure out how to program best to our architecture." NVIDIA has ramped up developer support with new tools that allow programmers and artists to sit in front of the same screen and analyze the performance hit that proposed art or effects will have on different NVIDIA platforms in real-time.

Featuring an adaptively tessellated ocean topped with dynamically computed foam, "explosion, smoke, and fire are present as the ship is destroyed" in the oddly named "Clear Sailing" demo.

Beyond NV40

There is no disputing how hard NVIDIA worked to recover from the spectacularly impressive triple Salchow face plant of NV30. The follow-on NV35 and NV38 chips proved that the company could refocus under pressure, and by all indications, NV40 is everything its creators hoped it would be. (See benchmark analysis and technology details of the GeForce 6800 Ultra in our hands-on preview on page 64.) In a diligent effort to reinvent itself, NVIDIA has not balked internally at admitting mistakes, rethinking fundamental technologies, and reorganizing production methods, all of it, reportedly, without the customary pointing of fingers and rolling of heads. If only NASA could do that.

But is it enough? Generally speaking, the market does not reward effort, it rewards results. In the past few years, NVIDIA has held onto most of its large market share, despite not having the best high-end desktop technology. Brand loyalty and solid marketing, along with a product that could compete (with a bit of fudging here or there) in the same arena as the top performer, kept things going along fairly well. Should the current tussle between NV40 and ATI's R420 go well or badly for NVIDIA, it could have a big impact on sales of its desktop graphics products, but the future of NVIDIA is based on more than just making pretty pixels for your favorite run/kill/die game.

A New Era

NVIDIA's wider market focus starts with what Huang calls the era of "Networked Digital Media." The idea is that the proliferation of digital content spreading across the PC, consumer electronics, and handheld devices through shared standards and formats, together with the arrival of the necessary infrastructure (ubiquitous broadband, cheap storage, and really big LCD TVs) to distribute, archive, and display high-quality media is creating a massive opportunity. According to Huang, NVIDIA has something to offer the Networked Digital Media Era in three key technology areas. The Network, Media, and Connectivity Processing arena is addressed by NVIDIA's nForce MCP (Media and Communications Processor) technology, which is set to capitalize on the transition to 64-bit computing. The Digital Media Software category covers all of the separate technologies contained in NVIDIA's ForceWare drivers, including a high-quality multimedia server, game profiles, multimonitor and HDTV support, system utilities, RAID disc management, an audio mixer, and firewall network security.

Graphics and Video Processing is the largest category, encompassing all of NVIDIA's GPUs and its advanced capabilities in programmable shading, high-definition video processing, and image processing. Although the competition has been incorporating video processing for several generations, even calling their product a VPU (Visual Processing Unit), Huang's view for NVIDIA technology is a very long one: "NV4x is our first-generation architecture where we really reached out and thought of programmability and thought of the GPU as a digital media processor instead of just as a graphics processor. It is a transformational idea for our company, and I believe that in 10 years, our GPUs will be used to drive all kinds of stuff, from car navigation systems to projectors to digital televisions to you name it."

Branding & Licensing

A unified marketing message is essential, and there are currently six brands in the NVIDIA store, each with a distinct purpose and market: GeForce-branded GPUs on the desktop, GeForce Go for mobile, Quadro in workstations, GoForce processors for the handheld market, the nForce brand MCP southbridge for motherboards, and the ForceWare branded unified software environment to pull it all together.

NVIDIA is likely to continue its habit of entering into cross licensing agreements with other computer graphics technology companies, which Huang sees as a liberating force allowing engineers to take best design paths instead of cumbersome workarounds. "What a cross license does is enable you the freedom to innovate. If IBM would like to use our technologies so that they can build whatever computers they want, terrific. If Silicon Graphics would like the same thing, terrific. Evans & Sutherland? Terrific. I happen to believe that intellectual property is used defensively and used wisely to enable you the luxury to innovate in the marketplace. The thing about innovation is that all good ideas are going to be copied. So unless you want to take on everybody in the world and spend the rest of your life in court, your best defense is to keep innovating."

Handheld Devices

NVIDIA is ramping up graphics processors for handheld devices very quickly. Last year's acquisition of MediaQ brought in some advanced ultra-low-power processor technology and the engineers that created it. Now integrated with NVIDIA's small in-house team under a new vice president and general manager of Handheld, the second-generation handheld product, with an incorporated 3.2-megapixel camera, will be released later this year. NVIDIA is also allocating resources for handheld content developer support. (See "Forward Slash" on page 89 for details.) Huang believes NVIDIA's long experience processing pixels for a demanding audience will enable it to compete favorably in this new area. "We will be able to add value [because] computing the pixels for cell phones is arguably even harder [than for PC]. You have such a small footprint, whether it's memory or bandwidth or display size, that you have to take extraordinary care . . . the video that's coming to you is in such a densely packed and compressed [format] that you have to do a lot of media processing to it to finally deliver the beautiful pixel. In the area of 3D, because you don't have that many pixels to start with and because it's so close to your face, you better make sure that every single pixel is highly cared for and comes across beautiful and delicate, otherwise it will look grainy and ugly."

Growth potential in this area is huge, and Huang says it is time to "deliver the type of experiences that we imagine a great cell phone would have, whether it's making a video phone call, watching CNN, using it to listen to music, or playing a small game while you're sitting on a plane. We believe that the time is finally here." Again, there is also the long vision, "Over the course of the next five to 10 years, flexible displays are really going to change the way we think about handhelds. These mechanical cell phone-like devices with buttons, that's going to be the past. Your cell phone is just going to be a single semitranslucent sheet of what appears to look like plastic. It's going to be bendable. You're going to touch it, and it will turn on. It will render the buttons instead of having real buttons. It will be very, very high resolution, and it will be beautiful."

Passionate Pragmatism

In speaking with NVIDIA executives and employees for this story, a common description of the company and its CEO was "pragmatic." But at the crux of NVIDIA's corporate culture is a fundamental contradiction: Bleeding-edge innovation is driven not by pragmatism but by passion. It is much more practical to let someone else make the enormous investment in various R&D paths (some that create new markets and some that fail miserably) and simply stand by and observe which ones actually work and how they are utilized before pouncing on the market opportunities. It is far more sensible to be just behind the curve of change if you don't want to end up under its wheels. Pointing any industry in the right direction is risky. Leading the way with technology before there is a market for it is risky. Passion is risky.

But passion may just be the best game in town. As Chris Malachowsky puts it, "Never underestimate the ingenuity and cleverness of someone who has a goal in life and finds himself slightly challenged. Have you seen the Unreal engine? That thing is gorgeous. There's no geometry behind it, and that's pretty impressive, and that's cleverness. Carmack is clever. Tim Sweeny is clever. These guys and the folks that are like them in the ranks are just smart as whips. They just find a way. The state of the art is continually being pushed." Though recently elevated to the role of NVIDIA Fellow and no longer on the hook for day-to-day engineering issues, Malachowsky sums it up: "We're still in the realm of practical. We need to do a practical implementation and developers need to do a practical deployment of their technology. No one's trying to be ideal. Ideal would probably put us all out of business. But over time we are going to approach ideal."
Words from the wise.

by Joan Wood

NVIDIA Brands

2004 Fortune 100 Best Companies to Work For After the head of HR refers to NVIDIA's work environment as having the pace of "a bullet train at warp speed," it's easier to swallow his other assessment: "NVIDIA is a way of life." Long hours for high pay and great benefits, with an attrition rate of only 4% and free food sounds like a pretty decent way of life to us. Potential candidates are screened "very aggressively" for intelligence, and their work history is checked for signs of "passionate execution." Decision-making is pushed down to the lowest level so the ability to make decisions is important, and there are very few administrative assistants, so be prepared to "get your hands dirty." Education, Internships & Scholarships A BSEEE degree is the most common requirement, and there is an open requisition to hire as many ASIC engineers and semiconductor designers as can be found. Top-tier schools are preferred (Berkley, Stanford, MIT, Carnegie Mellon, Waterloo in Canada). A Computer Science degree is best for software engineering. Ten percent to 20% of all hires are for software, which will increase over time. Top-tier and some second-tier schools are acceptable if they have a good reputation for software (University of Illinois). One hundred to 120 internships are offered every year in hardware, software, marketing, developer relations, and developer technology. About half of the interns are converted to full-time employees. Five scholarships are awarded annually. NVIDIA Gear Want to feel like part of the team without working those crazy hours? The NVIDIA company store (shopnvidia.com) sells everything from NVIDIA T-shirts to the Ultimate Frisbee. Sorry, no discount graphics boards. (Did we mention the free food?) NVIDIA Today $2 billion technology company 2,000 employees Over $1 billion in R&D in last three years More then 600 patents & filings in graphics, digital media, and networking 300 million processors shipped