Thinking Beyond Science

Nvidia, a designer of 3-D graphics chips, had a very rapid rise, shooting in a few short years past apparently stronger firms, including Intel, to dominate the high-performance 3-D graphics chip market. In 2007, Forbes named Nvidia the “Company of the Year,” explaining that “Since Huang [CEO and founder] took the company public in 1999, Nvidia’s shares have risen 21-fold, edging out even the mighty Apple over the same time period.”

Nvidia’s domain of 3-D graphics has nothing to do with wearing special glasses or with images that jump off the page or out of the movie screen. The “3-D” in 3-D graphics describes the process used to create an image appearing on a computer monitor. If you are looking at a still image, it may not be obvious that is was created with 3-D graphics technology-it is just a still image generated by dots of color. However, once you gain real-time control over the position of the imaginary cam-era, the difference is dramatic. Using a mouse or joystick to control the imaginary camera, you can explore a 3-D scene, viewing it from different positions, moving your viewpoint around objects to view them from behind or above, and moving into rooms and spaces at will. This is possible because the computer “knows” the three-dimensional structure of the whole scene.

Many of the fundamental components of modern 3-D graphics technology were developed at the University of Utah as part of a stream of research initiated by Professors Ivan Sutherland and David Evans in the late 1960s. While other computer science programs were teaching high theory, the Utah program was focused on the practical challenge of rendering 3-D images and building flight simulators. The program produced an astounding number of computer graphics superstars, including John Warnock, founder of Adobe Systems; Nolan Bushnell, founder of Atari; Edwin Catmull, cofounder of Pixar; and Jim Clark, founder of both Silicon Graphics and Netscape.

Nvidia was formed in 1993, by Jen-Hsun (pronounced Jen-Sun) Huang, Curtis Priem, and Chris Malachowsky. Huang had been an engineer-manager at LSI Logic. Priem and Malachowsky had been chief technology officer and vice president for hardware engineering, respectively, at Sun Microsystems.

Plus, smaller transistors were faster and consumed less power. The whole semiconductor industry coordinated around achieving a higher level of integration, based on smaller transistors, about every eighteen months. This rate of progress was called Moore’s law. No one could jump much ahead of this pace because all the technologies, from photolithography to optical design to metal deposition to testing, had to ad vance in lockstep. The industry called this pattern of collective advance the “road map.”

On the demand side, management judged that the market would buy virtually all the graphics processing power that could be provided.

There was no sharp call for word processors or spreadsheets that ran one hundred times faster, but there was a very sharp demand for faster and more realistic graphics. Chief scientist David Kirk put it this way:

“There is a virtually limitless demand for computational power in 3-D graphics. Given the architecture of the PC, there is only so much you can do with a more powerful CPU. But it is easy to use up one teraflop of graphics computing power. The GPU [graphics processing unit] is going to be the center of technology and value added in consumer computing.”

Finally, CEO Jen-Hsun Huang believed that Nvidia could construct an advantage by breaking out of the industry’s eighteen-month cycle.

He reasoned that since it looked possible to advance graphics power three times faster than CPU power, Nvidia could deliver a substantial upgrade in graphics power every six months instead of every eighteen months.

The first step in executing the guiding policy was the establishment of three separate development teams. Each would work to an eighten. month start-to-market cycle. With overlapping schedules, the three teams would deliver a new product every six months.

A two-month delay in a six-month cycle was much more serious than the same delay in an eighteen-month cycle. Accordingly, the second set of policies were meant to substantially reduce the delays and uncertainty in the development process.

A serious source of possible delays was a design error. After designing a chip, the company passed the design to a fabricator. After about one month, engineers received back the first samples of actual chips. If bugs were found in these chips, the design would have to be changed, new masks would have to be created, and a new fabrication run initiated. To address this source of possible delays, Nvidia invested heavily in simulation and emulation techniques and organized its chip design process around these methods.

The benefit of a faster cycle is that the product will be best in class more often. Compared to a competitor working on an eighteen-month cycle, Nvidia’s six-month cycle would mean that its chip would be the better product about 83 percent of the time. Plus, there is the constant buzz surrounding new product introductions, a substitute for expensive advertising. As a further plus, the faster company’s engineers will get more experience and, perhaps, learn more about the tricks of turning the technology into product.

As Nvidia gained leadership in performance, it began to focus even more intensely on the problems of the delays, driver problems, and extra costs created by its board makers. First, management tried to negotiate a new arrangement with Diamond Multimedia but were rebuffed because Diamond did not want to see a decrease in its margins.

A senior team then traveled to Dell and made a presentation. They described the obstacles created by current industry practices, showed the economic benefits of the unified driver architecture, and pointed out the lower prices contract board manufacturers could deliver. Dell reacted positively and agreed to offer boards with Nvidia chips, manufactured by Celestica Hong Kong Ltd. In the months and years to come, Nvidia increasingly relied on contract electronics manufacturers for board production and distribution. The contract manufacturers were free to brand the board as they chose; most chose to emphasize the Nvidia name.

Over the next five years, Nvidia continued its pattern of rapid releases, pushing the envelope in 3-D graphics capabilities. During the 1997-2001 period, Nvidia obtained extraordinary gains from integrating the graphics pipeline onto a single chip, achieving a 157 percent average annual gain in performance.* From 2002 to 2007, it achieved an annual performance gain averaging 62 percent per year, about as much as was possible, given the general movement of semiconductor technology. Intel CPUs, for example, increased their processing power (millions of operations per second) at about the same rate per year during that period. The difference is that the effects of Intel’s performance increases have been substantially dulled by hardware and software bottlenecks it cannot control. On the other hand, as Nvidia’s leaders envisioned, the performance of graphics chips is experienced directly and immediately by users. Enthusiasts continued to wait anxiously for each new improvement.

Nvidia jumped from nowhere to dominance almost purely with good strategy. Follow the story of Nvidia and you will clearly see the kernel of a good strategy at work: diagnosis, guiding policy, and coherent action.

You will also glimpse almost every building block of good strategy: intelligent anticipation, a guiding policy that reduced complexity, the power of design, focus, using advantage, riding a dynamic wave of change, and the important role played by the inertia and disarray of rivals.

Source – Good Strategy Bad Strategy: The Difference and Why It Matters by Richard P. Rumelt

Goodreads – https://www.goodreads.com/book/show/11721966-good-strategy-bad-strategy