An AnandTech Exclusive: The Jim Keller Interview
by Dr. Ian Cutress on July 16, 2018 11:00 AM ESTIn the past, we've mentioned that in the semiconductor industry there are a few individuals that are more than worthy of the term 'rockstar'. These individuals, no matter where they are, shake up their industry and often do it many times over their career. One such individual is Jim Keller, whose public successes have included being involved in the early designs on the first K8 prototypes, as well as designing Apple's A4/A5 mobile processors, and being a critical element in leading AMD's 'Core Development' team, covering both Zen and K12, specifically focusing on the Skybridge concept. He also co-authored the HyperTransport interconnect standard.
Back in April this year, Jim moved from a stint at Tesla (working on low-voltage Autopilot hardware) to take a position at Intel, as Senior Vice President of the Silicon Engineering Group. Aside from Raja Koduri, now also at Intel, Jim Keller was one of the most requested persons in a straw poll I took via Twitter earlier this year. Clearly someone at Intel was watching in, as this week Intel invited us to a short exclusive* 1-on-1 interview with the man himself, to focus on his new role and position.
Ian Cutress AnandTech |
Jim Keller Intel |
In the last two decades, Jim has overseen a number of pivotal products, but it bears mentioning that modern processor engineering teams are massive groups working on 3-7 year development cycles, perhaps more. In previous companies, Jim has been tasked with developing teams as well as performance, but now that he is under the umbrella of Intel's big development behemoth – officially focusing on 'SoC Development and Integration' – one of the key areas will be how a company the size of Intel, which already has a number of good engineers, can get the best out of a rockstar like Jim Keller. Dr. Murthy Renduchintala – Intel’s Chief Engineering Officer – said that the company has “embarked on exciting initiatives to fundamentally change the way we build the silicon as we enter the world of heterogeneous process and architectures,” which may been seen as a hint of Intel’s future direction.
The Jim Keller Interview
Ian Cutress: For a number of my generation, it was a shock to hear that the famous Jim Keller was set and poised to join Intel. Is it like a Formula 1 driver wanting to drive for Ferrari - did you expect that Intel was going to be on the cards at some point in your career? Why join Intel?
Jim Keller: To be honest, I didn't really ever think about it much. In my career, you know, some people have asked me how I planned it, and the honest truth is I've really just kind of worked on the next interesting thing. And so, I've had a bunch, at least to me, of almost random jobs and interesting experiences and they've all been interesting. When this came up (at Intel), Murthy, Brian, Raja kind of pinged me after Raja joined. I talked to them at some length before I joined Tesla, well to BK [Brian Krzanich] a couple of times at least, and then they pinged me again later. They offered me an opportunity that seemed really interesting so, I took it.
IC: When the offer came to join Intel in 2018, what parts of your brain were firing? How did you expect Intel to differ from the previous companies you have worked for?
JK: Well I'll tell you I've worked different places and I've solely figured out that the culture of every company is different, and while I have ideas about how it's going to be, my plan coming in was to get to know people and get a feel for the DNA of the place and the people and the problems. I was not really coming in thinking I know what to do about it - some places you go in kind of knowing there's a definite plan and agenda and other places you don't. So, I didn't really have a good idea what I'd find here to be honest.
IC: So I've heard that some companies hire talent for the sake of hiring talent, are you just given a carte blanche or a remit to figure out what exactly you want to do?
JK: I was hired to be Senior Vice President of Silicon Engineering, so Murthy had a plan and a structure in place. So generally a large number of the SoC teams and the CPU execution team work for me, so that's a pretty big space, it's quite a large group and they have a lot going on.
IC: When the announcement was made, that you were to take a position at Intel as SVP of the Silicon Engineering Group, focusing on SoC Development and Integration, it sounded like a very vague title. Can you dive into a broader explanation of what you are doing at Intel, and how that excites you?
JK: I'm not sure why you think that's vague, can you clarify?
IC: As Intel has so many varied products in its arsenal, there are a lot of which would fall under SoC development and integration.
JK: The interesting thing about Intel is scale - the scale of the company is big and the company is in a number of markets. They play to drive new technology is big, and working at Intel in the role of developing lots of silicon products for me is really interesting because the potential impact is really broad. It's playing on a big field so that's really interesting. I've had different roles at different companies at different levels and this is a new opportunity for me and I'm quite interested in taking it on.
IC: So the fact that scale could be 5x or 10x more than what you've perhaps previously done, does that excite you?
JK: Oh yeah, sure, yeah that's super cool! I mean, you don't get an offer like that every day that's for sure! And then you know, coming into a culture and environment, like for years the world sees Intel as this big monolithic imposing semiconductor company but I would say it's really an engineering excellence focused company, from my 2 months of talking to people. The culture is very interesting about how people think about their work and go about it. My plan like I said is to get to know people and talk to everybody and see what's really going on and how I can be of service to that. So far, it has been great actually.
IC: With only six weeks into the job, how many faces are familiar? Have you had to go around introducing yourself?
JK: Yeah, not that many to be honest, I mean I know some of the senior technical people, either by reputation or personally, but not a tremendous number. I mean, of my staff I think I knew about 2 of them but I hadn't met them before, out of about 20. I'd say there's VPs of SoC teams - I'm a senior VP so some of my staff are VP's and then there's some senior technical people so I like to have a mix of organizationally focused people and technical people.
IC: Silicon Engineering is a big group, with engineers spread all over. Are you set to lead a specific team for specific projects, or help guide a number of teams across the range of Intel's product line?
JK: You know I do all kinds of stuff. There's two fun things about these kinds of projects: one is building great products that are really different and successful. The other is working with engineering teams to get people engaged - you know to make sure like lower friction in the environment, build better catapults, build better process, build better organization and these things are kind of complicated and you can read about the after effect stories in management books, but you know when you're actually engaging with real people and engineers, some changes are subtle and some changes are not. So I kind of contribute where I need to contribute you know, partly I think if I'm successful then I don't have that much to do, but then what happens is stuff always pops up and I'm like, my brain is kind of energetic so I'm always thinking of stuff.
I'm quite interested in different levels of this, there's the big direction of computing, like I'm not really a visionary, I'm more of a go build stuff kind of person. But to build stuff, there's technical problems and there's organizational problems and even when things are going well, you know there's interesting things to change and there's decisions about when do you do an incremental product, when do you do a new one, you know, Intel's big enough that that landscape is complex. To me, you know, the scope and scale of this is super interesting.
IC: Can you discuss if you're tasked with expanding teams and building teams, or are you already analysing with product roadmaps already?
JK: Intel as a whole is quite a large engineering organization and I didn't walk in thinking I'm sure I have to go expand some teams. We're always finishing projects and starting new ones so there's lots of work and analysis and architecting and judgment about what should we do on the next project and how do you organize to do that properly. But yeah I'm not really here to expand or build teams, I'm here to build products and I think we're well staffed. We hire good people all the time, I mean there's a certain amount of people coming and going, that's pretty normal. I think retention at Intel's quite high.
IC: So how does that work with the heirarchy, between you, Raja, and Murthy Renduchintala? We know that you have worked with Raja at AMD, have you worked with Murthy before?
JK: No I haven't, I knew about him, I knew when he came here, I had quite a few conversations before I joined and obviously a lot since I've joined. He's a very driven character and he really likes to get products done. So far we seem to fit pretty well.
IC: We saw a picture on Twitter of you and Raja Koduri visiting the Intel High Performance Lab in Haifa last week - can you talk about what you were doing there?
JK: So you know, some of my teams in Israel and we're going there to work on some roadmap decisions and then we got a tour of the lab and kind of wandered around and then Raja posted that picture so yeah, it's just normal work. It turns out there's a direct flight from San Francisco to Tel Aviv so that's super good, so we can basically drive to the airport, sleep for 10 hours, wake up in Haifa and that team is great. One thing when I came in is that they asked me if I had ever heard of them, I said "you guys are famous!"
IC: Normally when we speak to new hire engineers at companies, they often describe that the projects they are working on won't be customer facing for either two years or perhaps even five. Is the desire there to implement some short term projects in the next year or so, or do you expect them to be further out, more longer term?
JK: I like the whole pipeline, like, I've been talking to people about how do our bring up labs and power performance characterization work, such as how does our SoC and integration and verification work? I like examining the whole stack. We're doing an evaluation on how long it takes to get a new design into emulation, what the quality metrics are, so yeah I'm all over the place.
We just had an AI summit where all the leaders for AI were there, we have quite a few projects going on there, I mean Intel's a major player in AI already, like virtually every software stack runs on Xeon and we have quite a few projects going on. There's the advanced development stuff, there's nuts and bolts execution, there's process and methodology bring up. Yeah I have a fairly broad experience in the computer business. I'm a ‘no stone unturned’ technical kind of person – when we were in Haifa and I was bugging an engineer about the cleanliness of the fixture where the surface mount packages plug into the test boards.
Also, we have a super cool lab here where we have 2 megawatt server walls that we use to run our CAD system and I like all the details so, I'm not really a future kind of guy. I think about the future obviously because my favorite quote from Ray Kurzweil is ‘People think the next 25 years is going to look like the last 25 years, but the future accelerates, and so the next 10 years of computing is going to be kind of wild’. But the next 6 months is going to look a lot like the last 6 months and so there's I'd say a creative tension between the day to day work and the accelerating future. That's a super interesting problem especially when you have thousands of engineers working on it.
IC: What have you learned since joining Intel?
JK: I've learned that there's a lot of things going on, that there’s a high level of technical excellence, and what the teams capable of is super good. The Intel DNA is to ‘get stuff done’, and people are quite passionate about it so I quite like that. There's a lot of changing technology stuff and there's some uncertainty obviously in technical future, so you know people worry about that. Yeah on the whole it's quite an interesting landscape.
IC: Have you had time to examine Intel's competitors for the products you are interested in?
JK: Oh of course, yeah. I mean technology, you have to think about it, there are fundamental principles of technology, so how does computer architecture work and software work, and where is the transistor, the foundry roadmap go, and then you kind of think about what do you want to do, what's the best you can do, what's the competitive landscape look like, what are the customer’s needs, so that's space is multidimensional and you have to consider everything. You always compare what you're doing next to what you did last time right? That's a super fine idea but you also have to compare to what your competitors are doing, you have to compare it to what the best thing you could possibly do is, right? Then there's creative tension between those because the difficulty of achieving different things depends on how the lay of the land looks, and there's nothing easy about it.
IC: There are a number of elements to Intel's portfolio that must be new to you. What elements about Intel's range of products are you looking forward to getting into?
JK: Which part do you think is new to me?
IC: I mean the range of Intel’s products might mean that you may not have necessarily focused on certain areas in previous roles.
JK: Well, I'm kind of a computer expert. And an AI expert. And an SoC expert. I know a lot about GPUs and software stacks, I've worked on networking processors, I've worked quite extensively on transistor level design and foundry technology. Somebody once said to me that this must be all new to you, but it all pattern matches. It is complicated but the details are interesting, and the number of new faces to who does what is task in itself, but the thing is that that is new and is the next 10 years of computing.
We're going through a somewhat radical transformation with AI, up in the data center and in client. So not only is that new, but that's an industry new. But the mechanics of building processors is something I know pretty well. The CPU designers at Intel have designed some pretty innovative components of CPUs, and we've been building processors for a long time right, like 50 years. Certain details of processor design today look the same to me as 25 years ago, and certain things are just outrageously different because of the scale. The first branch predictor I put into a computer was a 2 kilobits of SRAM array and now you know, we probably have 100 megabits (across the chip), which you know is hilariously different.
So you know the expression when ‘a difference in scale creates a difference in kind’? I think of the difference between a neuron and a brain, or an ant and an ant colony, but we have lots of interesting breakpoints in computing based on the scale situation. Then we have breaks in computing because of things like the AI discovery of gradients of descent some years ago just transformed the computing model which has then turned into the wild-west in terms of computer design and so I see newness there. But on the execution side, that is skilled craftsmanship and Intel has lots of skilled craftsman, and some of their tools are sharper than others, but I'm a craftsman on the computer design side.
IC: Do you expect to be working on products that will be on the leading edge manufacturing nodes regularly?
JK: Of course
IC: How are you involved in how new processes are developed?
JK: Well in terms of how do they put metal layers on wafers, no. In terms of how do we define the next generation technology node optimized for performance frequency density power? You bet! In the first 2 weeks we sat down and walked through the metal stack for Intel's 14, 10 and 7nm and I have lots of opinions about that. So you know, I have a simple theory about this kind of stuff: as I tell people my opinions, one of two things happen. They like what I say, or they correct me when I'm wrong. So yeah, I'm going to be quite involved in that.
IC: I'd love to hear your opinions but I don't think the PR team will let me listen to them.
JK: There's somebody here shaking his head!
IC: I understand the reasons why!
JK: Intel has been a spectacular foundry developer and that's a space that is super complicated and you know, they led the world for 20 years and somebody has a hiccup and everybody is like ‘oh woe is me, what's going to happen next’. Personally I'm thinking I know what's going to happen next: fix it and keep going.
IC: There's a lot of people interested in that story and also how you fit into that story as well, we'll maybe come back and address that at a later date perhaps.
JK: You know, that's a complicated problem and again, these things are more sophisticated and more dimensional than people tend to think. There are lots of levers to make positive steps, and I've successfully designed successful projects in let's say fairly trailing technology where some innovation is architectural, some of the innovation is at the silicon level, some is at the system level, some of it is the software level, and it's always interesting to look at the whole pallet of that and then go make great products.
And again, the special charm with Intel is we play in lots of different pieces of it. Intel has set the standard for a lot of the software technology that goes across multiple platforms and as an architect I look for the coherence in that space and as a guy who builds products. I want to make sure the mechanics of doing stuff and the craftsmanship at every level is high. So yeah, I'm definitely going to roll with the technology guys. I have for years and I think CPU performance is super important and graphics performance is super important and density is super important and I'm quite interested to figure out how to go apply it. Like Intel has invested in 3D XPoint, which is really interesting technology, and we're now a player in wireless, and our range of components and power, I think of it as half a watt to megawatt problem. So that's you know, that's a pretty broad range.
IC: Have there been ideas in the back of your mind that you've wanted to do, that now can be realized at Intel?
JK: I kind of work and focus more on the next really interesting problem. We announced I think that we're going to go into the graphics business and we've been developing some really interesting graphics. We've made some pretty serious moves and are already a serious player in AI, and the scale we can go deploy that at is super interesting. But you know, the pillars of our business are the PC world and the server world, and building, making sure that our customers there are super happy and that we're doing the next great thing.
IC: Do you expect to be public facing when it comes time to announce future products?
JK: I doubt it, I'm not really that public of a person, so I like to work on technology and every once in a while, I'm sure they'll drag me out. I'm not exactly a PR person.
IC: In the market today we have high-performance CPUs, we have super GPUs, hardened IP blocks, mass deployed accelerators. Where is the future of compute from your perspective? Should the goal simply be 'exascale', or something more nuanced?
JK: The computer world is getting pretty broad right, like I said, and we got interesting points from a half a watt to a megawatt. We have the scale and the scope of engineering team that will address quite a few of those pieces.
IC: Do we have to have another x86-64 moment?
JK: Well 64-bit's a lot of address space so we haven't ran out of that yet! I think it's more interesting if you go look at the new applications and the type of data that they're processing and how do you accelerate that kind of data processing. There are all kinds of people who have problems that take days and days to compute right? There are things we'd like to compute in seconds to minutes that take days.
So every once in a while somebody says ‘Whoa, what happens when computers are fast enough?’, and I always think, ‘really?’. Like again, Ray Kurzweil has a graphic of computer performance I think from 1870 to today which is log linear. So when some people ask me where I think the future is going – well I think it's been a log linear scale for 100 years, now it is 104 years, so what do you think is going to happen next? I think it's going to go for a while.
But I’m sort of solving it as visionary as I can get because then I'll look at what's the right thing to do next for our server product right now, and a GPU product, and an AI product, and a client product, and a power constrained product, and a power unconstrained product. You know, the choices you make with every design point are interesting and there's better and worse choices and there's also tradeoffs about the customer needs, competitive needs and expediency. Solving that whole thing is charmingly complicated.
IC: For you personally, what's the best way you take advantage of the newest upcoming technology? A good workstation? A self-driving car? High-end smartphone? Are you a gamer?
JK: I'm a kite surfer so I don't know what to tell you about that. I'm like the least computer savvy computer designer I think. I just had somebody help me log in to one of my online accounts. So I use my smartphone, I love my self-driving car, I use technology all the time kind of randomly but I'm not really a technology nerd except as tools to go build computers.
IC: Thank you Jim, thanks for your time!
JK: Sure thing, it was fun!
Many thanks to Gavin Bonshor for his speedy transcription.
*We've been told that Jim took a couple of other interviews with business focused outlets, who may or may not publish their discussions. So this technically might not be an exclusive, but out of the technical press the only place you'll read something is here at AnandTech.
Additional: There were a few comments here and on social media about what exactly this interview was about. I decided to test my recording equipment and put something together, and go through the interview analyzing what was being said.
129 Comments
View All Comments
porcupineLTD - Monday, July 16, 2018 - link
From my understanding the equipment necessary for EUV is manufactured slowly and at very low capacity and you need to book orders over a year in advance, apparently GloFlo and Samsung are first on the booking list and they expect to have volume production ready in late 2019. Even if Intel has mastered EUV (witch seems unlikely considering they are having problems with DUV in witch they have decades of experience) they could not go into volume production since they cant procure the equipment they need.name99 - Tuesday, July 17, 2018 - link
Which part of"Use EUV for their existing 10nm line. (Sure, EUV has lower volumes. Which, given that the actually useful 10nm volumes right now are ZERO, seems like a strange problem to get hung up on.)"
did you not understand?
mode_13h - Wednesday, July 18, 2018 - link
Uh, which part of "cant procure the equipment they need" don't you understand?I'm not saying @porcupineLTD is right, but your rebuttal isn't even responding to their point.
name99 - Wednesday, July 18, 2018 - link
Q: Does Intel own ANY EUV equipment?A: yes
Q: Is ANY number of chips made with that EUV equipment more than zero?
A: yes
Which part of this do you not understand? Let's say a single ASML machine can create 150 wafers an hour (this numbers varies depending on details, but is kinda the right ballpark).
OK, so you ran that 24/7 (at 50% downtime) and you can process 50,000 wafers per month.
OK, now there are issues of how many passes does each wafer need through EUV, etc, but the end result is not negligible, it's a reasonable fraction of the output of a fab.
It's certainly enough to ship premier versions of IceLake, priced to constrain demand until you get a few more EUV machines on line. But you see commercial advantage in sitting around doing nothing, learning nothing about real-world EUV, and continuing to insist that Cannon Lake is too a real product?
mode_13h - Thursday, July 19, 2018 - link
You act like that was obvious and common knowledge, but it obviously wasn't. You attack others for your own failure to make your case.0ldman79 - Tuesday, July 17, 2018 - link
Very true.Right now Intel's 10nm process is about as useful as the carburetor from the late 70s that could get 100mpg... that kind of thing is awesome, but where is it? I've seen as many reputable reports of 100mpg carbs as I have working 10nm Intel CPUs.
kn00tcn - Monday, July 16, 2018 - link
just dont understand why you keep writing 'raju' for months & months, it's been written in article titles, article contents, people you've replied to, etc... can only think of max 1-2 other people that did it even though that's a vague memoryis it stuck in some auto correct dictionary?
kn00tcn - Monday, July 16, 2018 - link
"it's been written" = *the real name of 'raja' has been in sightarashi - Monday, July 16, 2018 - link
I'm convinced that HStewart is a Markov Chain.Bulat Ziganshin - Wednesday, July 18, 2018 - link
Intel is ALREADY producing mobile CPUs with 10 nm, do you really expect that they don't started 7 nm DEVELOPMENT long before that?>Jim was very clever about not giving too much away
and it's why these interviews are meaningless. People just not going to say anything real, it's pure PR action with all those "I'm excited" stuff.