At SEMICON West, Dave Patterson (center) sat down for an interview with author John Markoff (left). Playing the sidekick role was Mark Hill (right), who is a professor at the University of Wisconsin, but he was one of Dave's early students at UC Berkeley. Dave splits his time today between Berkeley and Google. He was one of this year's recipients of the Alan Turing Award, along with John Hennessy. For more about that, see my post Hennessy and Patterson Receive the 2018 Turing Award . John said that he wrote the first article on RISC in the much-loved Byte Magazine in 1996, which is when they first met. Now, John said, you have received the Turing Award, the Nobel Prize of Computer Science. Dave replied: I’m surprised but this seems to be a different type of award. People who don’t know who I am or what I do invite me to give talks, so now people think anything I say is important. "So how did you come to computer science?" Dave said he grew up in southern California but got into an advanced math class and did calculus. He went to UCLA and studied math, and at that point, he planned to become an actuary. But in the last quarter of his junior year a math course was canceled and so he had to fill out his schedule. He did a computer programming course (Fortran and punched cards). Then he was hooked. There was no CS major in those days, but in his senior year, he took all the courses he could in computing. One faculty member asked if he wanted to get involved in research, so he went to graduate school. That was the beginning of it all. I'm the first of my family to even get a college degree. The first week I came back and told my Mom 'I could do this forever.' Who knew I pretty much would? "You started in aerospace," John said. "Could you have stuck there?" I'm married to my high-school sweetheart for over 50 years. I'd lost my previous job so I had started working for Hughes Aerospace. I'd done mostly software at school but I was there for 5 years working on hardware. My wife's sisters all had nice houses but we didn't have the money. I was also in grad school for my masters degree. I asked my wife if I could keep going and get a PhD at Berkeley. She had two questions. "If we go to Hughes now, can we go to Berkeley later? And if we go to Berkeley now can we go to Hughes later?". "No, we have to go to Berkeley now" I replied. "OK, we'll be poor but proud," she said. So we went. John asked Mark how he came to be Dave's student? I was looking around at a number of professors and I approached Dave and he gave me a test to do some decimal arithmetic routines for RISC, because that was important back then. [the IBM 360 had a whole instruction set for decimal arithmetic without turning the numbers into binary and back] Apparently I passed the test. But what I saw in Dave wasn't just someone who'd give you technical advice, but someone to grow the whole person. it is the last vestige of the master-apprentice way of teaching. John asked Dave about the roots of the original RISC and Dave told the story of his sabbatical at DEC, and his rejected paper on on-chip microcode. I won't repeat it here, but you can read it in my post Fifty Years of Computer Architecture . Hennessy and I were both funded by DARPA so we knew what the other was doing. We decided that we shouldn't just imitate DEC and IBM and compete, we joined forces. Everyone else was the enemy, not Cal-Stanford. If we had done the classic academic thing and gone head-to-head, it would have made it more difficult to get accepted. "You took the ideas to industry in different ways," John observed. "Hennessy founded MIPS, but you went to Sun but didn't do a startup." I got the job offer from Berkeley in like July but didn’t finish my dissertation until December. For some reason, I read the book Working by Studs Terkel when people looked back at their careers. I learned that people who built things that lasted forever felt good, and people who worked with people, like priests, felt good. So I decided to stay in academia and when asked if I wanted to start a company I gave that little speech. I think I was too dedicated to the idea of academia. I worked with another company [Sun Microsystems] with Bill Joy who recruited me as a consultant and that led to SPARC. "So RISC took over the world, but didn't take over the desktop. Why not?" John [Hennessy] thinks that had us RISC people agreed on a single instruction set it might have got more success. It wasn't clear at the start just how important binary compatibility was, and we didn't have that. It was all high-level languages, and you just had to recompile. Intel also figured out how to have 500 people on a design team that would overcome the disadvantages of the x86 ISA. They also had better process technology. Those two things meant that they could overcome the advantages of RISC. They also had branch prediction that was so good they could hide the disadvantage of the decoding. John switched to teaching. "Dave, when we first met you were trying to teach Smalltalk to undergraduates. Did you continue to focus on undergraduates?" It's funny, I often get asked "Do you teach undergraduates?" and I say "Yes, that's the job description." A big part is the undergraduates. I've always loved teaching and seeing young kids learning stuff for the first time. It's fun to see the light bulbs go off. "Now you are at Google too? Is TPU [Google's Tensor Processing Unit for AI] too specialized?" If you were to poll us computer architects, we'll say the future is special purpose architecture. It was a bad idea to go to multiprocessor, mutlticore since it was bad for software. But there was nothing else. Now the only thing left is special purpose architecture like the Google TPU. A couple of years ago I spent a couple of days at the industry partners meeting at Stanford, and I had dinner with an architect and he was ecstatic. “Wow it’s our turn!" he said. Another aspect of Moore's Law was falling costs, but now they are not. My friends at TSMC say cost per transistor is flat across process nodes for the last few nodes. Intel have said that 10nm is delayed again. This is what the end of Moore's Law looks like. Whole new industries used to come out of cost change. Gordon Bell said that every time you cut the cost in half the industry grows by a factor of 10. When I went to grad school, people said that computers were a weird thing to study. I guess I've gotten smarter over the years since now computers are everywhere. John asked his final question: "Dave, I was watching your talk and at the end rather than personal wealth you have a focus." It struck me that along my career I got to meet lots of millionaires, tens, hundreds of millions. They never seemed particularly happy. It seemed like a better model was just to aim for happiness directly. Psychology changed from just studying sick people to also studying what makes people happy. Having goals in your life. Having friends. You can study how to be happy now. Go directly to the goal. Sign up for Sunday Brunch, the weekly Breakfast Bytes email.![]()
↧