Intel's Patrick Gelsinger on the hot seat
- 06 June, 2008 07:26
Patrick Gelsinger is an electrical engineer. He joined Intel in 1979, worked on the design of the 80286 and 80386 microprocessors, and was the chief architect for the 80486 chip.
On the 30th anniversary of the x86 architecture's birth, he talked about why it has been so successful.
Intel and Microsoft in a sense grew up together. Did you have any interactions with Bill Gates early on?
I had numerous interactions with him over the years. He was one of the smartest and most aggressive guys in the industry. He was always attacking, pushing, striving for more.
I remember at one point he had less-than-flattering things to say to me about the 80386. I was a junior engineer at the time, and he was already a legend in the industry. I started arguing with him. It was quite incredible. Everyone else in the room fell silent as Bill and I were going back and forth in a very aggressive manner. The senior Intel guy -- after we had a little break to cool the air -- told me I wouldn't be needed for the rest of the meeting.
So, how did the 80386 turn out?
The 32-bit 386 was the turning point in the industry. The 80286 was a 16-bit architecture. At the time people said, "What do you mean 32 bits? That's for mini computers and mainframes." They derided us for being extravagant. They said 64K [for memory addresses] was plenty. I had many conversations with press people and analyst about what applications I could possibly conceive of that would need that much memory.
What are some other key events that you witnessed in the evolution of the x86 architecture?
It was at that same time [in 1986] that the PC-compatible machine moved from IBM to Compaq. When PCs moved out from under the shadow of IBM, it really [created] an industry-standard platform.
Also, I'd point to the move to the Pentium Pro [in 1995]. It was a dramatic architectural leap to what's known as an out-of-order architecture [in which instructions could be executed out of order]. We just took the best ideas of computer architectures from the minis and mainframes and implemented them better, because we had a superior canvas to paint them on, called silicon.
The x86 faced a huge competitive threat from RISC computing in the late 1980s and early 1990s. What was your role in that?
At the time, I was leading the development of the 486 at Intel while working on my master's degree at Stanford. My thesis adviser was John Hennessy [now president of Stanford University]. John and David Patterson were really known as the authors of RISC.
John and I had three public debates about RISC. I argued that even though the [CISC architecture of the x86] was a little bit slower, by the time you can develop new software for the RISC machine, we've made the [x86] machine that much faster. I said, "Already, John, your architecture is a failure because it can't deliver the software value that is already available in the x86." It was a particularly pointed comment in the debate, and he turned to me and said, "Who's your teacher, anyway?" And I knew he'd be giving me my master's diploma in just a few days.
Ultimately, you won the argument. Did Hennessy give you your degree?
Yes. We had an overwhelming economic advantage because we had so much of an installed base and so many people developing. The RISC machine could never catch up. But John and I have a wonderful camaraderie to this day.
You worked on the design and development of the 386 but then became the chief architect for the 486 at the age of 25. What was that like?
The 486 was my baby from beginning to end. It was frightening yet exhilarating at the same time. I had only a couple of years of experience, having just finished my degree at Stanford, and I was put in charge of what would be the most important project in the industry at the time. There were 100-plus people in the program, some of them 25 years my senior.
What is the "software spiral"?
The day before we introduced the 80486 [on April 10, 1989], we already had billions of dollars worth of software waiting to run on the chip. When we introduced it, [Intel CEO] Andy Grove made this powerful statement about the "software spiral."
The idea of the spiral was, we'd introduce hardware that was faster than the software and then allow the software to catch up, which requires the hardware to jump ahead again. So software begets hardware, hardware enables software, and that spiral is really what's been driving the industry for many, many years.
It seems that that concept still holds, with multicore chips now available but without a lot of software that can run on more than one processor at a time.
Exactly. I'd say we have made a new turn in the spiral, which is the move to multicore.
When did Intel realize that the game was up for processor clock-speed improvements and that the future lay in multicore processors?
I was chief technology officer for Intel in 2001, and I published a paper that became known as "the power wall" paper. It had this graph, which became famous, that plotted the thermal density -- how much power we were trying to squeeze into a given area of silicon -- and it showed that our chips would have the thermal density equal to a nuclear reactor or to the surface of the sun. It showed that the trajectory we were on was not sustainable. We needed to make what became known inside Intel as the right-hand turn.
There were huge debates internally over the data -- whether we could continue on the curve we were on, whether there were cooling systems that would allow us to continue and so on. And people said, "Hey, if we aren't in the gigahertz race, where are we? Our competition will build gigahertz chips, and the market has been conditioned to see that as a better chip." So there were technical reasons and market reasons why there was tremendous resistance to the argument.
What does the future hold for the x86?
I gave a speech at the Intel Developer Forum in Shanghai [in April]. I included a picture of the Monkey King, which in China had a very powerful tool -- the golden stick. It could be as small as a needle or as big as a pillar that held up the sky.
In my speech, "From Milli Watts to Peta FLOPS," I made the analogy that, like the Monkey King that had the golden stick, we had the architecture that will grow up to be the biggest things on Earth -- petaflops computers -- and also the smallest computers on earth.