The Denny's Founding: Three Engineers' Gamble on 3D Graphics

The fluorescent lights of the 24-hour Denny’s on Branham Road in San Jose hummed at a pitch only audible to people who had been awake for more than 24 hours, casting a sickly yellow glow over the scuffed plastic booth where three men hunched over a crumpled stack of notebook paper at 1:17 a.m. on a Thursday in November 1993. The pot of black coffee they’d ordered at midnight had long gone cold, the plate of half-eaten fries in front of 30-year-old Jensen Huang was crusted with salt, and the ashtray (this was still an era when restaurants let you smoke indoors) was overflowing with cigarette butts from Chris Malachowsky, who chain-smoked when he was deep in thought. Curtis Priem, the youngest of the three at 28, had his glasses slipped halfway down his nose, scribbling calculations on the back of a napkin with a ballpoint pen that had run out of ink an hour earlier.

They had all quit stable, high-paying jobs at tech giants just six months prior to bet their life savings on an idea that most of their former colleagues thought was insane: building a chip that could bring the 3D graphics capabilities that only existed on $100,000 Silicon Graphics workstations into the homes of ordinary PC users, powered by the booming market for PC video games. Jensen had just left his role as a chip design manager at LSI Logic, where he’d spent four years working on graphics processing chips for high-end workstations. Chris had spent a decade at Sun Microsystems building the graphics hardware that powered the company’s iconic Unix workstations, and Curtis had been a lead designer at IBM’s T.J. Watson Research Center, where he’d worked on some of the first experimental 3D graphics chips for consumer devices. They’d first met at LSI Logic a few years earlier, bonding over late-night conversations about the future of graphics, and had spent the past six months working out of a tiny garage in Jensen’s backyard in Los Altos, trying to turn their shared vision into a real product.

> 📌 电子版侧边栏：3D图形加速是如何工作的？ > 想象你要在电脑屏幕上显示一个3D的游戏角色： > 1. 建模：先用数学公式定义角色的所有顶点、边和表面，相当于给角色做一个3D的“骨架” > 2. 纹理映射：给每个表面贴上对应的颜色、纹理，比如皮肤、衣服的图案 > 3. 光栅化：把这些3D的表面转换成屏幕上的2D像素点，计算每个像素的颜色、亮度、阴影，以及它和其他物体的遮挡关系 > > 1990年代之前，这三步工作全部由电脑的CPU来完成，而CPU的主要任务是运行操作系统和应用程序，只能分出很少的资源来处理3D图形，所以最多只能跑非常简单的3D场景。3D图形加速芯片（也就是后来的GPU）的作用就是把光栅化这一步从CPU中分离出来，用专门的硬件来并行处理，速度能提升几十倍甚至上百倍。

To anyone outside of their small circle, their bet looked like a terrible one. In 1993, the global PC market was shipping 120 million units a year, but less than 5% of those PCs had any 3D graphics acceleration capability at all.^[3] The only 3D graphics chips on the market were either designed for professional workstations, costing thousands of dollars, or were niche add-on cards for hardcore gamers, like the 3dfx Voodoo, which had launched earlier that year and cost $299 – roughly 10% of the average American’s monthly income at the time. The vast majority of PC users were still using 2D graphics cards that could only display flat, pixelated images, and the only 3D content available for PCs were clunky flight simulators and the just-released *Wolfenstein 3D*, a first-person shooter that had sold 1 million copies in its first month and proved, for the first time, that there was a mass market for 3D video games.

Just two weeks before this Denny’s meeting, Microsoft had announced Direct3D, a new standardized programming interface that would let game developers write code that worked across any 3D graphics chip, instead of having to write custom drivers for every single card on the market. It was a gamble for Microsoft too – the company had never done anything like this before, and most industry analysts thought 3D graphics would remain a niche hobby for years to come. But Jensen, Chris, and Curtis saw something no one else did: Direct3D was the missing piece that would turn 3D graphics from a niche hobby into a mass market. If they could build the best, cheapest chip that worked with Direct

If they could build the best, cheapest chip that worked with Direct3D standard, they wouldn’t just capture the gaming market—they would own the foundational layer of the next generation of personal computing. The question hanging over that sticky Denny’s booth wasn’t whether 3D graphics would matter. It was whether three engineers with no brand, no revenue, and a garage full of prototype chips could outmaneuver the biggest names in tech to build the standard itself.

The argument that night stretched for three more hours, long after the waitress had stopped refilling their coffee and started giving them suspicious looks for occupying the booth after the breakfast rush began. Jensen was the uncompromising optimist, tapping a ballpoint pen against the white Formica table as he laid out his math: the PC market was growing at 30% a year, Microsoft was pushing 3D support as a core feature of its upcoming Windows 95 operating system, and *Wolfenstein 3D* had already proven that regular people would pay $50 for a 3D game. If they could hit a $99 price point for a 3D-capable graphics card—less than the cost of a new pair of work boots—they could sell a million units in three years. “We’re not selling chips,” he told them, his voice hoarse from four hours of talking. “We’re selling the future of how people interact with computers. And if we don’t build it, someone else will.”

Chris was the skeptic, and for good reason. He’d spent 10 years at Sun building graphics hardware for $20,000 Unix workstations, where reliability and precision mattered more than speed or cost. He’d seen dozens of promising graphics startups flame out because they bet on the wrong technology or misread market demand. “Jensen, the only people buying 3D graphics right now are engineers and hardcore gamers,” he said, stubbing out his seventh cigarette of the night. “The average person buying a PC at CompUSA doesn’t care if their solitaire game has 3D cards. They care if it’s cheap and doesn’t crash.” He held up the napkin Curtis had been scribbling on, covered in calculations for their first chip, codenamed NV1. “This thing uses quadratic texture mapping. It’s better than what 3dfx is doing, but no game developer is going to write code for it unless we force them to. And if we try to force them, we lose.”

Curtis, the quietest of the three, was the pragmatist. He’d spent years at IBM working on consumer electronics, where he’d learned that the best technology in the world was worthless if no one could use it. He’d been the one pushing for months to scrap the NV1’s custom architecture and build a chip that worked with Direct3D, the new Microsoft standard that was already being adopted by every major game studio. “Chris is right about the market,” he said, pushing his glasses up his nose. “But Jensen is right about the direction. We can’t build a chip for the market that exists today. We have to build for the market that’s going to exist in two years. The question is whether we can build something that works with Direct3D, is cheap enough for regular people, and is good enough that game developers want to use it. If we can, we win. If we can’t, we’re out of business in 18 months.”

They voted at 4:12 a.m., just as the sun was starting to come up over the Santa Cruz Mountains visible through the Denny’s parking lot window. Jensen and Curtis voted to bet everything on Direct3D compatibility and a sub-$100 price point. Chris voted against it, but agreed to stay on, on one condition: if they didn’t have a working prototype that passed Direct3D compliance testing in six months, they would shut down the company and cut their losses. All three signed the founding documents on a napkin right there in the booth, the ink smudged from the condensation on the table. Jensen tucked the napkin into his wallet, where he kept it for the next 20 years.

---

The six months that followed were a blur of 18-hour days in the cramped two-car garage behind Jensen’s Los Altos home. The garage had no hea