[7] Jim Keller, Silicon Wizard 🧙‍♂️

hi everyone you join me on a very special occasion this is part two of our interview series with jim keller he is currently cto of a new ai company called tens torrent but in the past he's worked at dec apple amd intel uh broadcom cybite tesla jim am i missing anything there uh i worked at amd twice oh yes that's true so don't know what to make of that i think i have a couple of questions uh for you about that it's uh i wanna actually let's start with that that's a good place to start right it's um a lot of audience a lot of my audience at least know you from your work um at amd on the zen the sky bridge platform and you know now amd's gaining market share with their product line and you're off on to bigger and better things but there's been a lot of confusion as to your exact role at amd during that project you know some people believe you're integral in nailing down zens and twos and three high level micro architecture others believe you put people in place signed off at high level then focused on the armed version the sky bridge k12 so can you give a any clarity as to your role there you know how deep you went with zen versus k-12 or your involvement in things like infinity fabric yeah it's uh well it was it was a complicated project right because amd when i joined they had bulldozer and jaguar and they both had you know some charming features but they weren't successful in the market and the road maps weren't aggressive they were falling behind intel and so that's not a good thing to do if you're already behind then you better be catching up not falling behind so i took the role i was president of the cpu team which i think when i joined was 500 people and then over the next three years you know the soc team the fabric team some ip teams joined my my little gang i think when i left it was 25 2 400 people i was told so it was you know so i was a vp with you know a staff right so i had senior directors reporting to me and senior fellows and my staff was like 15 people so i was i was hardly writing rtl well that's you know that said um but we had a whole bunch of things so like i'm a computer architect not really a manager um i i wanted the management role which was the biggest management role i'd had at the time i'd been you know the vp of a startup but that was 50 people and we all got along and so this was a fairly different play for me but i knew that the technical changes we had to make would be involved in getting people aligned to it and i didn't want to be the architect on the side arguing with the vp about why somebody could or couldn't do the job or why this was the right or wrong decision so mark papermaster i told him my theory and he said okay you know we'll give it a try and uh it worked pretty good because then i had let's say direct authority as it were but people don't really do what they're told to do right they're do they do what they're inspired to do right and so you have to lay out a plan and part of it was finding out like who are the right people you know to do different things and sometimes somebody's really good but their ideas like people get very invested in what they did last time or they believe things can't be changed and i would say my my view was things were so bad that almost everything had to change and so i i went in that with that as a default thing does that make sense now it wasn't that we didn't find a whole bunch of stuff that was good to use but you know you had to prove that the old thing was good as opposed to prove the new thing was was good so we changed that mindset architecturally i had a pretty good idea what i wanted to build and why and i found people inside the company mike clark leslie barnes j fleischmann there are quite a few debjit now there's quite a few really great people that once we described what we wanted to do they're like yeah we want to do that and like architecturally i had some input there was often decisions analysis uh people had different opinions i was fairly hands-on doing that but again i wasn't doing block diagrams writing rtl and uh i had yeah we had multiple projects going on you know with zen there was the the arm the arm cousin of that the follow-on that integrated them both there was some new ssc methodology but we we we did more than just cpu design we did methodology design ip refactoring uh very large organizational changes and i was like hands on top to bottom with all that stuff does that make sense sure it's uh a few people consider you the father of zen do you think you ascribe to that that position or should that go to somebody else yeah or one of the uncles you know you know there was a relationship a lot of really great people on zen so and it was built you know there was a methodology team that was worldwide the soc team was partly in austin and partly in india the floating point in cash was done in colorado the core execution front end was in austin the armed front end was in sunnyvale we had good technical leaders you know i was in daily communication for a while with suzanne and steve hale who who kind of built the front end of this encore and the colorado team so like there was a really good people mike clark's a great architect um yeah so we had a lot of fun and uh you know success success has a lot of authors failures as one so that was the success so let's and then some other you know team stepped up we uh moved excavator to the boston team where they took over the finishing the design and the physical stuff harry fair and his guys did a great job on that um so there was some fairly stressful organizational changes that we did going through that and you know the team all came together so yeah i think there was uh there was a lot of camaraderie in it so i don't claim to be the i was brought you know the instigator and the chief nudge and you know part architect part you know it's a transformational leader and it was fun was everything that you did during that time sorry go ahead no you have there's a little bit of a lag between here and yeah whatever country you're in right now i was going to say was so is everything that you worked on now out amd or is there still you know kind of road map stuff still to come out do you think from the ideas that you helped propagate so when you build a new computer and zen was a new computer there was already work underway on what they called sock 15 which is the new fabric and bus you build in basically a road map right yeah so i was thinking about you know what are we going to do for five years like chip after chip and you know we did this at apple too when we built the first big chord apple we built big bones right so you can't when you make a computer faster is two ways to do it you make the fundamental structure bigger or you tweak features and zen had a big structure and then there was obvious things to do for several generations of follow-up and and they've they've been following through on that so at some point you'll have to do another big you know rewrite and change i don't know if they started that yet um what we had planned for architectural performance improvements were fairly large over a couple of years and they seemed to be doing a great job executing to that but yeah i've been out there for a while four or five years though yeah yeah it's it's i think they they said that zen three the last one that just came out was a rewrite so i think some people are thinking that that was still under your direction yeah it's hard to say you know like even when we did zen like we did a clunk from scratch design clean design at the top but then when they built it there was a whole bunch of pieces of rtl that came from bulldozer and jaguar which were perfectly good they just had to be modified and built into the new zen structure so so hardware guys are super good at using code when it's boots them and so when they say they did a big rewrite you know they probably took some pieces and you know re-architected them at the top but when they built the code you know it wouldn't surprise me somewhere between 20 and 80 of the code was the same stuff or mildly modified but that's that's pretty normal the key is to get the structure right and then reuse code as needed as opposed to taking something that's complicated and trying to tweak it to get somewhere so if they did a rewrite they probably meant they fixed the structure so i mean moving on from that you're you're several companies on you're now a company called tens torrent um with with an old friend of the bishop basic um but you've been jumping from company to company each company you know for for basically your whole career you're always finding another project another opportunity another angle not to be too blunt but is tense torrent gonna be a forever home well first you know i was a digital for 15 years right now yeah that was two different careers because i was in the mid-range group where we built computers out of echo ucl you know they were refrigerator-sized boxes and i was in the alpha team where we built little microprocessor little teeny things right which at the time we thought were huge you know 300 square millimeters at 50 watts just blew everybody's mind and so i was there for a while um i went to amd right during the internet rush and we did a whole bunch of stuff in a couple years you know we started out there on hyper transport two-piece servers like it was kind of a whirlwind of a place but you know i got sucked up or caught up in the enthusiasm of the internet and i went to scibite which got bought by broadcom and i was there for four years total and we delivered several generations of products and then it was at pa semi and uh we delivered a great product um but they didn't really want to sell the product for some reason or they thought they were going to sell it to apple so i actually went to apple and then apple bought pa sent me and then i worked without teams i was you know between pa semi and apple that was seven years so i don't really feel like that was jumping around too much then i jumped to amd i guess and that was fun for a while and uh then i went to tesla where we delivered hardware three so that was you know kind of phenomenal standing start to driving the car in 18 months i don't think that's ever been done before and that product shipped really successfully they sold they built a million of them last year um in tesla and then you know intel was a different kind of whirlwind so now you could say i jumped in and jumped out but i i sure had a lot of fun with him so so yeah i've been around a little bit um i like to think i mostly get done what i set out to accomplish you know my success right there is pretty high in terms of you know delivering products that you know have lasting value i'm not the guy to tweak things in production i like either a clean piece of paper or a complete disaster you know that seems to be the the things i do best at it's good to know yourself i'm not an operational manager um so tense torrent is is more the clean piece of paper ai space is exploding the company itself is already many years old um but we're building a new generation of parts and and going to market and starting to sell stuff uh i'm cto and president i have a big stake in the company both you know financially and and also i say uh commitment wise to my friends there so i plan on being here for a while but uh that sounds good uh if if i can just cycle back one um i know it's still kind of fresh so i'm not sure what kind of ndas you're still under um but your work at intel was that more sort of a clean slate or can you sort of go into any detail about what you did there you know beyond the sort of pr stuff that we've all had i can't talk too much obviously the the role i had was senior vice president of silicon engineering group and the team was 10 000 people so they're doing so many different things it's just amazing it was something like 60 or 70 socs in flight at time literally from you know design to you know prototyping debugging and production so it was a fairly diverse group and there my staff was you know vice presidents and senior fellows so it was a big organizational thing i had thought i was going there because there was a bunch of new technology to go build i spent most of my time working with the team about both organizational methodology transformation like new cad tools new methodologies new ways to build chips a couple years before i joined they started what's called the you know socip view of building chips versus intel's historic monolithic view and that to be honest wasn't going well because they took the monolithic chips like they built great client and server parts and broke it into pieces but you can't just break it into pieces you have to actually rebuild those pieces and some of the methodology goes with it and yeah we we found a bunch of people were really excited about working on that you know i spent a lot of time on you know quality ip quality density libraries characterization process technology like you name it it was my my days were kind of wild like some days i have 14 different meetings in one day you know just be click click click click because there's so many things going on all those meetings how did you get anything done i don't get anything done like i told you i was i was the senior vice president you know it's like evaluation set direction make judgment calls uh you know let's say try some organizational change people change and you know that adds up after a while you know the key thing about getting somewhere is knowing where you're going and then put an organization in place that knows how to do that and that takes a lot of work yeah so i didn't write much code but i did send a lot of text messages that kind of code right i mean intel now has a new sort of uh engineering focus ceo and pat gil singer would you ever consider going back if the right opportunity came up i don't know i have a really fun job and a really you know explosive growth market so i wish him the best you know i think i i think it was a good choice i hope it's a good choice uh we'll we'll see what happens he definitely cares a lot about intel and he's had real success in the past so he's definitely going to bring a lot more technical focus to the the company um but i like working with bob swan just fine so he was you know we'll see what happens i mean if i go through your career it's you've kind of you know gone between high performance computing and you know low powered efficient computing and now you're in this sort of world of you know ai acceleration um has it ever got boring no no it's really weird well it's changing it's changed so much and at some level it doesn't change at all like computers at the bottom they just you know you add ones and zeros together you know it's like it's pretty easy zero one one zero one one one zero zero it's it's not that complicated but you know i work on the vex 8800 where we built it out of gatorades that had 200 or gates of each chip like 200 right and now when at 10 store like our little computers are we call them tencit cores are four trillion operations a second per core and there's a hundred of them so like the building block shifted from you know 200 gates to four terraps so that's kind of that's kind of a wild transformation and then the tools are way better than they used to be like what you can do now like you can't build more complicated things unless you know the abstraction levels change and the tools change and there's been so many changes on that kind of stuff and when i was a kid i used to think i had to do everything myself and you know i worked like a maniac and coded all the time and and now i am you know i know how to work with people and organizations and listen you know stuff like that you know people skills like i probably get a pretty uneven scorecard on the people skills but um i do have a few and uh that's interesting would you say that engineers need more people skills these days because everything's complex everything's got separate abstract layers and if you want to work between them and yeah now so here's the here's the fundamental truth people aren't getting any smarter right so people can't work across more things that's just that's dumb right you do have to build tools and organizations that support people's ability to do complicated things right so like like it's interesting the vax 8800 team was like 150 people and the team that built uh first processor at apple hurricane that was in the a but the the first big processor i guess the second custom apple core um was 150 people now the cat tools were unbelievably better we used thousands of computers to do simulations we had tools that could place and route 2 million gates versus 200 you know so some things changed radically but the number of people an engineer talked to in a given day didn't change at all like if you have an engineer talk to more than five people a day they'll lose their minds you know so they you know some things are really constant in terms of that constant idea you've you've spoken about um cpu instruction sets in the past um and one of the biggest requests for this interview i got was around your opinion about cpu instruction sets um you know specifically how we deal with fundamental limits on them how we pivot to better ones you know and what's your skin in the game in terms of your arm versus x86 versus risk five i think at one point you said most compute happens on a couple of dozen opcodes so iso doesn't matter am i remembering that correctly it's a very sad story yeah yeah yeah yeah no it's not even a couple of dozen eighty percent of the execution is like six instructions you know load store add subtract compare and branch you pretty much covered it you know if you're writing in you know pearl or something you know maybe maybe call and return are more important than compare and branch but you know instruction sets only matter a little bit right you can lose 10 or 20 because you're missing instructions or you know for a while we thought variable length instructions were really hard to decode but we keep figuring out how to do that you basically predict where all the instructions are in tables and once you have good predictors you can predict that stuff well enough so fixed length instructions seem really nice when you're building little baby computers but you're building like a really big computer the predictor to figure out where all the instructions are isn't dominating the die so it doesn't doesn't matter that much and when risk first came out an x86 was half microcode so if you look at the die half the chip was a rom there may be a third or something and the risk guys could say there's no rom and a wrist chap so we get more performance but now the rom is so small you can't find it actually the adder is so small you can hardly find it right so what limits computer performance is predictability and the two big ones are you know instruction predictability and branch predictability and and data locality and now the new predictors are really good at that they're big like the predictors are way bigger than the adder right and that's where the you get into the the cpu versus like gpu or ai engine debate yeah like the gpu guys will say look there's no branch predictor because we do everything in parallel so the chip is way more adders and subtractors and that's true if that's the problem you have but they're they're crap at running c programs yeah it's yeah it's about crafting the software or the compile time right if you can do that well then and then gpus were built to run shader programs on pixels so you're given like eight million pixels and so you know the big gpus now have six thousand threads so you can cover all the pixels with each one of them running you know a thousand programs per per frame but it's sort of like you know an army of ants carrying around grains of sand whereas big ai computers they have really big matrix multipliers and like a much smaller number of threads essentially that do a lot more math because the problem is inherently big whereas whereas you know the shader problem was the problems were inherently small because there were so many pixels so there's genuinely three different kinds of computers cpus gpus and ai and then nvidia is kind of doing the tweener thing where they're using a gpu to run ai and they're trying to enhance it and some of that's obviously working pretty well and some of it's obviously fairly complicated and so it's an interesting and this happens a lot like like general purpose cpus when they saw the vector performance of gpus added vector units and sometimes that was great because you only had a little bit of vector computing to do but if you had a lot a gpu might be a better solution it's i think you've said you said before that going beyond the sort of matrix you end up with your sort of massive graph structures especially for ai and ml and the whole point about tens torrent is it's a graph compiler and a graph compute engine not just a simple matrix multiply yep yeah so there's you know from from old math now i'm a mathematician so mathematicians are going to cringe a little bit but you know there was scalar math like a equal b plus c times d right and when you had a small number of transistors that's the all the math you could do yeah now we got more transistors and you could say hey i can do a vector of those like an equation properly in a step and then we got more transistors we could do a matrix like multiply right and then as we got more transistors you sort of wanted to take those big operations and break them up because if you make your matrix multiplier too big the power of just getting across it's a waste of energy so you want you find you want to build this optimal size block it's not too small like a like a thread and a gpu but it's not too big like covering the whole chip with one matrix multiplier would be a really dumb idea from a power idea so then you get this array of medium sized processors where medium is something like four tera ops like that's which is hilarious to me by the way because i remember when that was a really big number and once you break that up now you have to take the big operations and map them to the array of processors and ai looks like a graph of very big operations it's still a graph and then the big operations are factored down into smaller graphs now you got to lay that out on the chip with lots of processors and have the data flow around it and that seems to be a very different kind of computing than running a vector or matrix program so let's call it you know scalar vector matrix roger and i used to call it spatial i would say graph would probably be a better word and then uh correct me if i'm wrong but you're putting in vector engines into your graph chord next-gen so it's becoming this general purpose cpus that have vector engines on them yeah and that turns out to be um when you're running ai programs there's some general purpose computing you just want and then there's sometimes in the graph where you want to run a c program on the result of a ai operation and having that stuff be tightly coupled is nice actually on the same chipset the latency is super low and the power to get back and forth is reasonable so yeah we're we're working on an interesting roadmap for that and by the way that's that's a little computer architecture architecture research area like what's the right everybody has this idea what's the right mix with you know accelerated computing and general purpose computing and how are people using it and how do you build it in a way programmers can actually use it right that's that's the trick which we're working on so many people are asking you what do you think about arm versus x86 you know which one has the legs which one you know as to performance do you do you care much um i care a little like here's what happened so when x86 first came out it was super simple and clean right and then at the time there was multiple 8-bit architectures x86 the 80 6800 6502 i programmed probably all of them way back in the day and then x86 oddly enough was the open open version right they they licensed that to seven different companies and then that gave people opportunity to put in tape but intel surprisingly was both licensed it in one long run but then they went to 16 bits and 32 bits and then they added virtual memory virtualization security 64 bits and then more features so what happens to an architecture is as you add stuff you know you keep the old stuff so it's compatible so when arm first came out it was a clean 32-bit computer and compared to x86 it just looked like you know way simpler and easier to build but then they added a 16-bit mode and the it instruction which is awful and a kind of a weird floating point vector extension set with overlays and a register file and then 64 bits which partly cleaned it up and they added you know there was some special stuff for security and booting and you know so it slowly got more complicated now risk five shows up and it's it's the shiny new cousin right because there's no legacy and it's actually an open instruction set architecture and and people build in universities where they didn't have time or interest to add too much junk like some architectures have so relatively speaking just because of its pedigree and age it's early in the life cycle of complexity and it's a pretty good instruction set they did a fine job so if i was just going to say i want to build a computer really fast today and i want it to go fast wrist five is the easiest one it's the simplest one it's got all the right features it's got the right top eight instructions you know that you actually need to optimize it doesn't have too much junk so yeah so modern instruction sets have too much bloat especially the old ones because they have too much legacy baggage construction sets that have been iterated and added to will have too much blood yeah that's what always happens right so as you keep adding things like you know the engineers have the struggle right you have this really good design there's 10 features so you add some features to it the features all make it better but they make it more complicated and as you go along every new feature you add gets harder to do because the interactions with that feature and everything else gets terrible and and the marketing guys and you know the the old customers will say don't delete anything but meanwhile they're all playing with the new fresh thing that's you know only does 70 of what the old one does but it does it way better because it doesn't have all these problems right so this is you know i've talked about diminishing return curves there's a bunch of reasons for diminishing returns but one of them is the complexity of the interactions of things slow you down to the point where something simpler that did last would actually be faster and that's happened many times it's it's some result of complexity theory and you know human nefariousness i think so do you ever see a situation where x86 gets broken down and something just gets reinvented or will it just remain sort of legacy and then just new things will pop up like risk five two well well then with a fairly you know clean slate that obviously had to carry a long baggage for this and that but you know they deprecated a lot of like you know the old 16-bit modes and bats there's a whole bunch of junk that disappeared and sometimes if you you're careful you can say all right i need to support this legacy but it doesn't have to be performant and i can isolate it from the rest of the client emulator and stuff yeah emulate it or support it and we used to build computers such you know you had a front end you know fetch dispatch execute load store l2 cache and if you looked at the boundaries between them you'd see like a hundred wires doing random things that you know were dependent on exactly what cycle of the you know or what phase of the clock it was now the interfaces tend to look like it's like instruction boundaries i send an instruction from here to there and you know and then i have a protocol that says act so the computer inside doesn't look like a big mess of stuff connected together it looks like eight computers hooked together that do different things there's a fetch computer and the dispatch computer and the integer execution computer and a floating point computer and if you do that properly you can change the floating point without touching anybody else and again that's that's less of an instructions that thing than what was your design principle when you build it and then how did you do it and were you faithful to that because the thing is you get to a problem you say boy if i could just add these five wires between these two boxes i could get rid of this problem but every time you do that every time you violate the abstraction layer you've created a problem for future gym future gym like like i've done that so many times it's like if you'd solved it properly it would still be clean but you know some point you always have to hack a little bit and then that over time that kills you it's i've seen a number of talks where you where you speak about the concept of abstraction layers in pretty much a lot of aspects of engineering but also kind of you know life as well so that you can you know independently upgrade different layers without affecting those above and below and providing you know new platforms to build upon what point in your life did that kind of ethos click and what what what what what made what happened in your life to make that you know that be a pervasive element of of your personality oh pervasive element of my personality that's pretty funny i know i repeat it a lot maybe i'm trying to convince myself um the like when we built ev6 so dirk meyer was the other architect and then we had a couple other strong people we divided the design into a few pieces we wrote a very simple performance model that got it but when we built the thing it was a relatively short pipe for an out-of-order machine because we're still a little weak on predictors and there was a lot of interactions between things and it was a difficult design to build and we also built it with at the time we you know digital had a custom design methodology so we had 22 different flip-flops like people could roll their own flip-flop and we frequently would build large structures out of transistors and i remember somebody asked me what what elements were in our library and i said both of them end devices and p devices right and then i went to amd and k7 was built with a cell library now the engineers were really good at laying down the cell libraries in a way they got good performance but they only had two flip flops a big one and a little one and they had a clean cell library so they had an abstraction layer between the transistors and the designers and this was before the age of really good place in route tools and that was way better and then the interface that we built on ev6 which was later called the s2k bus because we licensed it to amd had a lot of complicated transactions to do snoops and loads and stores and reads and writes and all kinds of stuff and uh a friend of mine who was a digital research lab i explained how it worked to him one day his name is cruz kuroshko he went to google later and became one of the security gurus like he listened to me and he said he just shook his head he's like jim that's not the way you do this and he explained how virtual channels worked and how you could have separate abstract channels of information right and you get that right before you start encoding commands and saying what a wire does when it wiggled so the result of that let's say educational seminar slash ass kicking was hyper transport which had a lot of the protocol the s2k bus but it was built in a much more abstract way so i i would say that was probably you know my trans my uh move from amd to or digital to amd where we had the ideas of how to build high performance computing but the methodologies were like say integrated like transistor up to architecture was literally could be the same person yeah and the fbmd you know there was mark you know architects microarchitects the rtl people write verilog but they literally translated it to gates the gate library people it was much more of a layered approach and k7 was quite a fast processor and our first swing at k8 we kind of went backwards because bruskieski was my favorite circuit partner at the time he and i could talk about big designs but we we saw it as transistors but that's a complicated way to build computers and uh and since then i've been more convinced like the abstraction layers are right you don't overstep human capability that's the biggest problem right if you want to build something bigger and more complicated you better solve the abstraction layers because people aren't smarter and if you put more than 100 people on it it'll slow down not speed up and so so you have to solve that problem if you have more than 100 people you need to split into two abstraction layers yeah yeah exactly and there there's reasons for that like human beings are really good at tracking you know your inner circle of friends is like 10 20 people it's like a close family right and then there's this kind of 50 to 100 depending on how it's organized that you can keep track of but above that you don't you don't you read everybody outside your group of 100 people send me strangers and so there you have to have some different contract about how you do it like when we build zen we had 200 people and half the team did the front end and half the team did the back end and the interface between them was defined and they didn't really have to talk to each other about the details behind the contract right and that that was important now they get along pretty good and they work together but they didn't constantly have to go back and forth across that boundary speaking about abstraction layers um you've said on stage in interviews in the past that you're not worried about moore's law you know process node side evolution of semiconductors will eventually get worked out by someone somewhere would you say your attitude towards moore's law is apathetic or do you draw or you know trust engineers that will work out super proactive yeah subs like yeah yeah that's not empathetic at all like i know a lot of details about it well you you get into these kind of people conflate a few things like intel's tendon animator slipped right and people said oh this because moore's law is dead tsmc roadmap didn't slip at all right now some of that was alignment right tsmc's roadmap aligned to the euv machine availability so when they went from 16 to 10 to 7 they did something you know and tsmc has been really good at kind of doing these half steps so they did seven without euv and then seven plus with the uv and then five with more and then five plus they tweak stuff and then they um and the uv machines for a while people weren't sure if they were going to work but now asml's market caps twice intel's like so you know and then there's a funny thing i realized that the locus of innovation like we tend to think of tsmc samsung intel as the process leaders but a lot of the leadership is actually in the equipment manufacturers applied materials sml so who's building the innovative stuff and euv machines the worldwide sales so the number something like tsmc is going to bought like 150 uh e machines by 2023 or something like that like the numbers are phenomenal because not too many years ago people weren't even sure they're going to work and now there's x-ray lithography coming up and again you can you can say well it's impossible but bloody everything has been impossible um and then you know the fine print this is a richard feynman you know sentence he's kind of smart he said there's lots of room at the bottom and i personally can count and if you look at how many transistors how many atoms are across transistors there's a lot right and how many transistors you actually need to make a junction without too many quantum effects there's only ten so so there's there's lots of room there it's i recently put out a video about ibm's two nanometer announcement and it and the title was there's a lot more rumors in gordon more yeah a lot more so you're not empathetic either yeah well there's also this funny thing there's a belief system when everybody believes technology's moving at the space and the whole world oriented towards it because like technology isn't one thing you know there's people who figure out how to build transistors like you know what the process designers do it like intel or tsmc or samsung is they used equipment which can do features but then the features actually interact and then there's some really interesting trade-offs between like like should how should this be deposited and etched and how tall should it be and why and what the space so they're the you know they're told they're the craftsman using the tools so the tools have to be super sharp and the craftsmen have to be super knowledgeable right and that's a complicated play and and there's lots of interaction and at some level you can see because the machines themselves are complicated you have this little complexity combination where the machine manufacturers are doing different pieces they don't always coordinate perfectly or they coordinate through the you know essentially the machine integration guys who designed the process and that that's complicated and that can slow things down but it's it's not due to physics fundamentals like we're making good progress on physics fundamentals and uh i i put up your i think i mentioned before your scaled ml talk where you have the one that you have in comic sans where you have the printed x and you say you know as time goes on the way you print the x because of the laws of physics and there's still several more steps to go on euv and high n a euv coming in a couple of years but you mentioned x-ray you mentioned x-rays what's the timeline for that it's not even on my radar yet i'm forgetting what the the machines were available for commercial some part of the well typically when the when a technology that comes along they use it for like one thing the first time like you know when the euv was first using drams it was literally for one stop or maybe two so um i'm trying to remember twenty three twenty four it's not that yeah that means they're already up and running people are playing with it yeah and then the wild thing is you know the when they went from optical like the uv was about a 10x reduction in wavelength right so they'd got these crazy multi-patterning you know interference kind of stuff that you saw those pictures of and then ev they could print direct but actually they can use the same tricks on euv so ev is going to multi-patterning i think in three nanometer and and then there's there's many tricks you can do with that so yeah the physics is is really interesting and then you know with along with the physics you know of the optics stuff and then there's you know the purity of the materials which is super important and then temperature control so things don't go around too much like everywhere you look there's a you know interesting physics problem and uh but there's there's lots to do and thousands of different innovations needed i think i think i think yeah there's literally hundreds of thousands of people working on it so you know every group of 10 people about innovation there's there's more than enough innovation bandwidth so uh pivoting pivoting to a popular question we had it's um so one of the things that we've noted you doing as you go from company to company is you know the topic of building a team um you know and as you build more teams we've seen certain people take engineers from a team they've built a previous company to the next company you know it's have you got any insights into how you build your teams and have there been any different approaches at the companies that you worked for on this well the first thing you have to realize is are you building a team or finding one so there's a great museum in venice uh the david museum and the front of the museum there's these huge blocks of marble yeah 20 by 20 by 20. how they move them i don't know and the block of marble is sitting there and you know michelangelo could see this beautiful sculpture in it it was already there right yeah the problem was removing the excess marble so if you go in the company that's a thousand employees i guarantee you there's a good team there you don't have to hire anybody we didn't hire anybody when i was at amd i hardly hired anybody you know we moved people around we deployed people let's say and uh but there was plenty of great people there when i went to town the team from scratch because there was nobody at tesla that was building chips and you know i hired people that i knew but then we hired a bunch of people that i didn't know at some point and then this is one of those interesting things if you only hire you know i've seen leaders go from one company to another and they bring their 20 people and then they start trying to reproduce what they had before and that's that's a bad idea because 20 people is enough to reproduce it but but it alienates you know what you want it when you build a new team you ideally you get people you really like you know either you just met them or you work with them but you want some differences in approach and thinking because everybody gets into a local minimum so the new team has this opportunity to make something new together and some of that's because you know if you had 10 really great teams all working really well and then you made a new team with one person from each of those teams that may well be better because they will you know re re-select which the best ideas were because every team has pluses and minuses and so you have to think about if you're building a team or finding a team and then what's the what's the dynamic you're trying to create that gives gives it space for people to have new ideas or you know some people get stuck on one idea and then they work with new people and also they're doing this credible thing you think wow they're great they used to be not so great what happened well they were carrying some idea around that wasn't great and then they met somebody who challenged them or the environment forced them or you know whatever you know happened and all of a sudden they're doing a great job i've seen it happen so many times i used to ken olsen at digital said there's no bad employees there's just bad employee job matches when i was younger i thought that was stupid but i work with more people i've seen it happen so many bloody times that you know i've even fired people who went on to be really successful because they weren't doing a good job because they were stuck and emotionally they were committed to something that wasn't working and the act of moving to a different place freed them up you know i don't get thank you i should so how much of that also comes down to sort of company culture i mean when you're looking for the person for the right position or whether you're hiring in for the new new position do you try and get something that goes against the company grain or goes with the company grain do you have any tactics here or is it really depends on what you're doing if you're trying to do something really innovative it's probably mostly against if you have a have a project that's going really well bringing in the instigators is going to slow everybody down because you know you're already doing well you have to read the group and the environment and then there are some people who are really good they're really flexible they go on this project i'm going to fit in and just push but on the next one i'm going to you know you can see them they're building their network and team and on the next project they're ready to do a pivot and everybody's willing to work with them it's like trust is a funny thing right you know if somebody walks up and says jump off this bridge you'll be fine you're like but if you'd already been through a whole bunch of stuff with them and they said look trust me you jump and you're going to be fine it's going to suck but it's going to be fine you'll do it right and teams that trust each other are way more effective than ones that have to do everything with contracts negotiation and politics so that's probably one thing if you're building a team you know building or finding a team when you start seeing people let's say doing politics which means manipulating the environment for their own benefit um they gotta go unless they're the boss and then you know then you have to see if they deliver like some people are very political but they really they think their political strength comes from delivering yeah but but people randomly an organization that are political just cause lots of stress do you recommend that sort of uh early mid-career engineers they you know bounce around regularly from project project just so they don't get stuck in a hole so it sounds like that seems to be a common thing you learn fastest when you're doing something new working for somebody that knows way more than you so yeah you're relatively early in the career and you're not learning a lot or you know the people that you're working for aren't inspiring you then yeah you should probably change um so there's some careers where i've seen people bounce around three times because they're getting experience and they end up being good at nothing and they would have been better staying where they were and really getting deep at something so you know creative tension there's great attention between those two ideas that kind of leads into into a good question actually because i wanted to ask about you and you know your mentors you know going through your early career who did you look up you know to for leadership knowledge skills is there anyone you idolize oh yeah lots of people yeah i well it started out with my parents like i was really lucky my father was an engineer and my mom was super smart you know kind of more verbally and linguistically and the weird thing was when i grew up is i was sort of more like her you know thinking wise but i was dyslexic and i couldn't read and my father was an engineer and so i grew up thinking i was like him but i was actually you know intellectually more like my mother and they were both you know smart people now they came out of the 50s and my mom raised you know family and she didn't start her career as a therapist until later in life but they were you know they were pretty interesting people and i got that and then when i first started at digital i worked for a guy named bob stewart who was a great computer architect he did the pdp 1144 1170 back 780 vex and the ci interconnect and somebody said every project he ever worked on earned a billion dollars back when that was a huge number so i worked for him and he was great but there was a half a dozen other really great computer architects there and while i was at dec had deck research labs and i got to meet guys like butler lamson and chuck thacker and new oldham and you know i worked nancy kronenberg was like one of my mentors when i was a little kid practically who's one of the chief people in the vms operating system and so you know that was kind of lucky um so yeah did i idolize them well they were they were both daunting and not because i was a little bit of a you know i don't know i didn't quite shut up who they were at the time now i was more a little oblivious to what was going on you know like my first week in digital well we got trained on this drawing system called valid which is kind of before the metro graphics era and uh this guy walked in and he was asking us questions and telling us about hierarchical design and i explained to him why that was partly a good idea and probably stupid and so he had like an hour debate about it and then he walked off and somebody said that's gordon bell i said who's that you go he's just digital yeah really well he's wrong about half the stuff you just said i hope i straightened it out but you know i think that's just i'm low in you know serotonin activation or something so you know that's that's more of a mental problem with me than the future i think so so would you say you've matured not a bit is that where the fun is no no i mean there's a whole bunch of stuff when i was young it was like i get nervous when i gave a talk or like believing you know it used to be like i realized i had to understand the people around me better but you know i wasn't quite convinced um it was like just rather they you know just do the right thing or something so there's a bunch of stuff that changed like i'm now i'm really interested in what people think and why they think it and you know i have a lot of experience with that and every once in a while you can really help debug somebody or you know get the group to work better you know i don't mind giving public talks at all you know i just decided that the energy i got from being nervous was fun so you know i still remember like walking out on stage at intel at some conversations like 2 000 people and i was like i should have been really nervous but instead i was just really excited about it so some of that kind of stuff changed but that's partly conscious and partly just practice um i'm still really excited about like computer design and stuff i had a friend of mine's wife said what do they put in the water like all you guys ever do is talk about computers it's really fun you know changing the world you know it's great it also sounds like um you spend a lot more time sort of studying the human experience if you're understanding how people how people think how people operate compared to being you know mouthing at golden bell for now it's funny people occasionally ask me like like like i tell people i read books and you learn a lot from books books are fun by the way if you know how a book works somebody lives 20 years and then passionately writes their best ideas and there's lots of those books and then you go on amazon and find the best ones like it's hilarious right like really condensed experience in a book written and you can even select for better books like who knew right um but i've been reading a lot of books for a long time like it's it's hard to say well read these four books and it'll change your life because sometimes a book will change your life but reading the thousand books will change your life that's for damn sure and you know there's so much human experience that's useful it's like who knew shakespeare would be really useful for engineering management right but like what are all those stories you know power politics and devious guys and the minions doing all the work and the occasional hero who's saving the day and you know it's like how does that all play out like you're always played 500 years ago and it applies to corporate america every single day of the week so if you don't know textbook or machiavelli you don't know nothing i think i remember you saying that before you went into your big first management role you read 20 books about management techniques and you ended up uh realizing that you'd read 19 more than anybody else yeah pretty much yeah i i actually contracted with venkat rao who's famous for the ribbon farm blog and a few other things to you know figure out so i really liked his thinking about organization from his blog and he had a little thing at the bottom that says click here to buy me a cup of coffee or get a consulting you know or consult with me so i sent him an email so we started yakking and we spent a lot of time talking before i joined amd and he said you should read these books and i did i thought well i thought everybody who's a big management job did that nobody does yeah you know it was hilarious like 19 is generous i read 20 more management books than most people most managers i've ever read where they read some superficial thing like good the great you know which which has some nice stories in it but it's not that deep a book management wise you know you'd be better off reading carl young than good the great if you want to understand management do you find yourself reading more fiction or non-fiction as a kid you know i read all the non-fiction books and then my my parents had a book club so when i i didn't really learn to read until i was in fourth grade but somewhere around seventh or eighth grade i read all the books in the house so you know they had john updike and john barth was one of my favorite authors when i was a kid so there was a whole bunch of stories and yeah there's weird ones like um i'm i'm drawing a blank on the name uh i'll think of it a second doris lessing dork blessing wrote a series of science fiction books that were also psychological you know in inquiries and i read that and just i couldn't believe it like you just you know everyone saw stuff like that kind of blows your mind you know and it happened obviously at the right time but then now i read all kinds of stuff i like you know history and anthropology and psychology and you know mysticism and there's there's so many different things i've probably read less fiction books in the last 10 years but you know my when i was younger you know i read probably mostly fiction i mean it's i i did get a few comments um from the audience because when you are um being interviewed by lex you know you said that you read two books a week you know and you're very adept at quoting from key engineers and futurists and stuff i'm sure if you started tweeting what books you're reading you'll get a very large following yeah a sort of passive jim keller book club yeah that'd be fun i would say i used to read two books a week now i i read a lot but you know it tends to be blogs and all kinds of crazy stuff um but i've never i don't know like doing lex is super fun i don't know that i have the attention span for social media to do anything like that i'd forget about it for weeks at a time so i'd be sort of your clothes i'm sure people just want lists i think but how do you make sure that you're absorbing what you're reading rather than having your brain diverting about some other problem that you might be worrying about i mean well first of all i don't care at all [Laughter] i don't really care about that i know people that read books that are really worried if they're going to remember them and they spend all this time you know highlighting and analyzing it you know i read for interest right so what i really remember is like like you know like people have to write to write 250 page books because that's like a publisher role like and it doesn't matter if you have 50 pages of ideas or 500 so you can tell pretty fast i've read some really good books only 50 pages because that's all they had you know you read 50 pages you think wow it's really great and the next 50 pages is huh it's the same you know and they realize well he just fleshed it out no you know i wish you just published a shorter book but uh you know that is what it is but if the ideas are interesting that's good i i meditate regularly and then i think about you know what i'm thinking about which is sometimes related to what i'm reading and then if it's interesting it gets incorporated but you know your brain is this kind of weird thing you don't actually have access to all the ideas and thoughts and things you've read but your personality seems to be well informed by it and i trust that process so i don't worry that i can't remember somebody's name because their idea may have changed who i was and i don't you know remember which book it came from sorry as long as you passively absorb it at some level yeah well you know there's there's a combination of passive and active and you know i told lex like a lot of times i'm working on problems i i i prep my uh dreams for it and you know it's it's really useful um and that that's a fairly straightforward thing to do it's you know you before you fall asleep you call up your your mind what you're really working on and thinking about and then my personal experiences sometimes i really do work on that and sometimes that's just a problem in the way of what i actually need to think about and you know i'll dream about something else and i'll wake up going well one way or the other it was really interesting so i mean on on on the topic of time i mean here we are discussing and you've discussed with like you know personal health study meditation family but also how you execute professionally right are you one of these people who only needs four hours sleep at night i like seven well i don't i i you know everybody i added it up one day that you know my ideal day would have like 34 hours in it you know because i like to work out i like to mess around with my kids i like to sleep and eat and you know i like to work i like to read so i don't know it just work works the weird one because that that can fill in lots of more time than you want to spend on it but but i also really like working so it's you know it's it's a challenge to kind of tamp it down well i mean when when there's a deadline what what kind of gets pushed out of the way first because i mean you've worked at companies where getting the product out time to market has been a key element of what you're doing is exercise now for about the last six years the key thing for me is once i have too much to do in some place i find somebody who wants to do it more than me like i mostly work on unsolved problems [Music] you know i was the laziest person at tesla as best i could tell well tesla had a culture of like work 12 hours a day to make it look like you're working and yeah oh i worked you know nine till seven like i wasn't uh it was a lot of hours but i also you know i go running at lunch and work out you know i had a weightlifting room in deer creek on the right next to the the big machine shop so i go down there for an hour and work out and eat so and you know amd and intel they're big big organizations i had a really good staff so i'd find myself spending way too much time on presentations or working on some particular thing and then i'd find some people who wanted to work on it and i'd give it to them and you know go on vacation god speaking to press people like me and filling up your time what's what's your feeling about doing these sorts of press interviews and you know more sort of marketing and corporate and discussion which aren't really necessarily related to actually pushing the envelope it's just talk it's not just talk you know i you know i worked on some really interesting stuff so i like to talk about it like when i was in intel i realized one of the ways to influence the intel engineers like everybody thought moore's law was dead and i thought holy crap intel's the moore's law company like it would be early at drag you know if your main thing was dead and i thought it wasn't so you know i talked to various people because then they amplified it and debated and it went back inside and you know i reached more people inside intel by doing external talks so that was that was useful to me because you know i had a mission to you know build faster computers that's what i like to do so and then when like i talk to people they always bring all kinds of stuff up and you know like the work we do impacts people guys like you think really hard about it you talk to each other and then they talk to you and you ask all these questions and it's it's kind of stimulated it's fun it makes me orient like if you can explain something really clearly you probably know it like there's a lot of times you think you know it then you go to explain it and you're stumbling all around like i did some public talks where they were hard to do like the talk actually seemed simple but getting to the simple part you have to get your ideas out and reorganize them and then throw out the bs and you know it's a useful thing is that is that feynman or sagan said that if you can't explain a concept to a college at a college level first year college level then you don't really understand it yeah that sounds poorly like feynman he did that really well lectures here in fun physics yeah it was quite interesting fireman's problem is he had such a brilliant intuition for the math that his idea of simple was often not that simple like he'd just see it you could tell so he was like well i got from here to here by these three leaves like he calculated some orbital geometry in like five steps and he was like so excited about how simple it was but i think he was the only person in the room that thought it was simple and uh yeah he was i presume he had you know he had the ability to visualize things in his head and manipulate them so i remember you you're you're saying at one point that when it comes down to you know circuit level design that's the sort of thing you can do yeah yeah i'm much more of a visual thinker yeah if i had one super power it's like i feel like i can visualize how computer actually runs so when i do performance modeling and stuff like that i can see the whole thing in my head and i'm just writing the code down so i don't um which is really useful skill but you know i probably partly was born with it partly developed and partly it's a it's something that came out of you know what you know my late adult diagnosis of dyslexia yeah that's i was going to ask how much of that is nature versus nurture yeah it's it's hard you know there's there's this funny thing with super smart people often things are so easy for them that they can go a really long way without having to work hard so i'm not that smart definitely so you know persistence and you know what they call grit is super useful especially in computer design when you know lots of stuff takes a lot of tweaking you have to believe you know like you can get there but a lot of times there's a whole bunch of subtle iterations to do and practice with that actually really works so yeah everybody's a combination you know if you don't have any talent it's pretty hard to get anywhere but sometimes really talented people don't learn how to work so they they get stuck with just doing the things that are obvious not the things that take you know that persistence through the mess you know to the other side also i guess identifying that talent is is critical as well if you don't know you have it yeah but on the flip side you may have enough talent but you just haven't worked hard so people give up too soon sometimes so but you got to do something something you're really interested in like you know when people are like struggling like i don't know if i want to be an engineer in marketing or this or that it's like man what do you like like or especially people that think well i want to be an engineer but my parents or somebody wants me to be a manager like oh my god you know you're gonna have a tough life because you're not chasing your dream you're chasing somebody else's and the odds of you being excited about somebody else's is low and if you're not excited you're not gonna you're not gonna put the energy or learn so you know that's that's a tough tough loop to begin so i mean to to what extent do you spend your time mentoring others either inside organizations or externally with co-work previous co-workers or students do you ever envision yourself doing something on a more serious basis like the jim keller school of semiconductor design yeah so no it's funny because uh i'm mostly mission driven like we're gonna build zen we're gonna build autopilot you know and then there's people that work for me and then as soon as they start working for me i start figuring out who they are and then some of them are you know are fine some of them have big problems that need to be let's say you know dealt with one way or the other so then you know i'll i'll tell them what i want sometimes i'll give them some pointed advice sometimes i'll i'll do stuff and you can tell some people are really good at you know learning by you know following and then people later on told me that i was mentoring them and i'm thinking i thought i was kicking your ass you know it's a it's a funny experience i have you know there's quite a few people that said i impacted your life in something and but some of it was like you know you know i went after them about their health or diet because i thought they thought they looked you know not energized by life you know you can really make really big improvements there and it's worth doing by the way or they're doing the wrong thing and they're just not excited about you can tell they should be doing something else so they either got to figure out why they're not excited or get excited and then a lot of people start you know fussing with themselves or with other people about their status or something and the best way to have status is to do something great and everybody thinks you're great having status by trying to claw your way up is terrible because everybody thinks you're a climber and you know and sometimes it is don't have the confidence or skill to you know make the right choice there so but but it mostly comes out of you know being mission driven like like i do care about people at least i try to um and and then i see the results i mean it's really gratifying to get a big complicated project done and you know where it was when you started and then you know where it was when it was done and then people people when they work on successful things associate the leadership and the team they're working with which was being part of that so that's that's really great um doesn't always happen but it's i have a hard time doing that just quote mentoring people because what's the mission like somebody comes to you and says i want to get better well better at what well then that's you know like well i want to be better at you know playing violin it was like i'm not good at that or you know whereas i say hey we're going to build the world's fastest you know autopilot chip and then everybody working on it needs to get better at doing that and it turns out three quarters of their problems are actually personal not technical right yeah so to get the autopilot check you have to go go debug all that stuff you know and you know there's all kinds of problems you know like health problems and parental childhood problems and you know partner problems and workplace problems and you know career stall problems and jesus the list is so bloody long we have to take them all seriously like you know it turns out everybody thinks their own problems are really important right like you may not think their problems are important but i tell you they do they have a list you ask anybody what's your top five problems right they can probably tell you or even weirder they give you the wrong five because you can see the idea right that happens too but they don't give you the five they think you want they give you the five that they think you want to hear rather than the actual yeah or the you know people have no fly zone so their biggest problem may be something they don't want to talk about so you know but if you if you saw if you help them solve that then the project will go better and then at some point they'll appreciate you and then they'll say you're a mentor and you're thinking yeah kinda i don't know you mentioned about you know projects succeeding and you know people being proud of their products do you have a proudest moment of your career project accolade specific moments in time oh i have there's there's there's a whole bunch of them i worked with this woman a little bit becky loop and intel and we were debugging some quality things and it turns out there was a whole bunch of layers of stuff and then you know we were going back and forth on how to analyze it how to present it and you know and i was frustrated with the data and what was going on and one day she came in with this picture it was just perfect and it was like like i was you know i was really excited for her because she'd you know got to the bottom of it we actually saw a line of sight to fix and stuff but that kind of stuff happens a lot like like probably the epiphany yeah well working with somebody working with a group of people going in there like it was a mess and then it got better you know like you know the tesla autopilot thing was wild and zen's success has been fantastic like you know everybody thought the amd team couldn't shoot straight and now i was very intrigued with the possibility of building a really great computer with the team everybody thought was you know out of it like nobody thought amd had a great cpu design team but you know the people who built zen they had 25 30 30 years of you know works you know history at amd like that was mike and leslie they've been there for 25 30 years suzanne plummer yeah you know the life is yeah you know they're kind of lifers but but you know they had done many great projects there you know they both had good track they all had good track records um but you know what what did we do different well we set some really clear goals and then we reorganized hit the goals and we did some really thorough talent you know analysis of where we were there was a couple people had really checked out because they were frustrated they could never do the right thing and you know i listened to them oh jesus i'll have to listen to people we had this really fun meeting it was like one of the best experiences of my life suzanne called me up and said people on the zen team don't believe they can do it and i said great i'll drive to the airport i was in california i'll see you there tomorrow morning eight o'clock make sure you have a big room with lots of whiteboards there's like 30 angry people ready to tell me all the reasons why it wouldn't work so i just wrote them all down on a whiteboard and we spent two days solving them and it was wild because it started it was like me defending against you know the gang but people started to jump in and you know i was like whenever possible somebody would say i don't know how i know how we fix that you know now give him the pen and then get up on the board explain it and it worked out really good but we you know the thing was the honesty of what they did was great here are all the problems that we don't know how to solve and we're putting them all on the table they didn't like give you two and hold back 10 and say well you solved those two against you know there was none of that kind of kind of stuff they were serious people that had real problems and they'd been through projects where people said they could solve these problems and couldn't so they were probably calling me out it was like i just another bullshitter and uh and i'm not a bullshitter but i told them you know something we can do some some i don't know but i remember like mike clark was there he was like we could solve all these problems like and then you know i walked out our thing was pretty good but then you know people walked out of the room feeling okay but two days later the problems all pop back up and so you have to just you know like how often do you have to go convince somebody as many times as necessary and the leadership team on zen did it right that's why they they got through it it wasn't just you know me hectoring them from the sidelines there was lots of people and lots of parts of the team that really said yeah i'm willing to really put some energy into this which is great yes at some point i'd like i'd love to interview some of them but um amd keeps them under lock and key from the likes of us that's probably smart who knows is is there somebody in your career that you consider sort of like a silent hero that has hasn't got enough credit for the work that they've done or a person yeah most engineers most engineers oh it's there's so many of them it's unbelievable you know how many people you know like engineers you know they don't really get it you know compared to lawyers are making like 800 bucks an hour in silicon valley you know they you know engineers so often they just want to be left alone and do their work and you know crank out stuff and there's so many people there just bloody great you know i've talked to people it's like this is my eighth generation memory controller and they're just proud as hell because it works and there's no bugs in it and the rtl is clean and the comments are perfect you know they're all over the place like like i really like that i think but they don't self-promote or the company doesn't come yeah you know like engineers are more introverted and you know you know conscientious and introverted tend not to be you know the people who self-promote aren't you a little like me you've you've learned how to be more extroverted as you've grown well i decided i wanted to build bigger projects and to do that you have to pretend to be an extrovert and you have to promote yourself because there's a whole bunch of people who are decision makers who who don't go i'm going to do the work to find out who the best architect is they're going to say who is the person that everybody says is the best architect or the loudest or the you know capable so to some level if you want to succeed above you know you know you know principal engineer you have to understand how to work in the environment of people who play there and and some people are super good at that naturally so they get pretty high in organizations without much talent sometimes without much hard work and you know so you're you know the group of people you know director and above that you have to deal with has way different skill set than most of the engineers and so if you want to be part of that gang even if you're an engineer you have to you have to learn how that rolls and it's not that complicated read shakespeare young couple books machiavelli you know you know you can learn a lot from that so as i i know we're coming to time very shortly here but uh i i want to address a few questions that i think the audience are interested in getting your thoughts on such as uh do you do you care much about intel versus amd or intel versus apple from your perspective now uh you know i worked at all those places i really like working there um i think you know there's always new waves of stuff i'm really intrigued by the you know the wave of ai startups and the next revolution of computing so you know at the high level you know the extent of the corporate companies with stock prices and boards and you know products you know they're going to go fight it out they do it i care about the engineers and all the companies you know i work with lots of people there i really like them but now i don't think that much about them right now in terms of you know like what are they going to do next well we'll see you know apple's always been so insular about how they do stuff like their product touch point is you know aluminum box or a glass box yeah well they've always been kind of a secret society the engineers are great but you know they i think they miss out on participation in the wider technical world you know intel and amd sell components that go into many people's products so their interface to the public is more you know tangible in terms of the what the engineers do right because yeah they're chips you measure their performance and you care about how many pc express lanes they have and and whether the chip has bugs so this it impacts people more so i'm interested in what they do but they're big corporations with a life of their own so in terms of uh process design i mean currently you know the the the eda tools you know there's some amount of automation in there and advances in ai and machine learning are being expanded into processor design do you ever envision in visit do you ever envision a time where an ai model can design a purposeful multi-million device chip that will be unfathomable to human engineers yeah well so already the complexity of like you know high-end you know amd until apple chip is almost unfathomable to any one person right but but if you actually go down into details today you can mostly read the rtl or look at the cell libraries and say i know what they do right but if you go look inside a neural network you know that's been trained and say why is this weight five eight .001584 three no business is is is isn't that more data than design though well here's somebody told me this so so scientists traditionally they do a bunch of observations and they go hey when i drop a rock it accelerates like this and they calculate how fast accelerate and then they curve fit and they realize holy crap there's this equation right yeah so we you know so physicists for years came up with all these equations and then when they got to you know relativity they had the bend space and quantum mechanics they had to introduce probability but still there's mostly understandable equations there's phenomena now where there's a phenomena that a machine learning thing can learn like predict you know physics is you know some equation put inputs equation outputs right you know function output right but if there's a black box there where the you know the ai network says inputs black box of ai outputs and if you look in the box you can't tell what it means like there's no equation right so now you could say well that's the design of the neurons is obvious you know the little processors little four teraflop computers you know but the design of the weights is not obvious right and that's where the thing is now now let's go use an ai computer to go build an ai calculator what if you go look inside the a calculator and you can't tell why it's getting so now i know don't you understand the weights you don't understand the math or the circuits underneath them that's possible yeah so now you have two levels of things you don't understand but what do you design you might still be designing the human experience right like yeah commuter designers used to design things with transistors and how we design things with high-level languages so so those ai things will be building blocks but but it is pretty rare that there's going to be parts of science where the function is not intelligible like stephen wolfram said was something really interesting like he said physics physics there was physics by explanation that was aristotle you know 1500 years right he was wrong about a whole bunch of stuff and then there was physics by equation like newton copernicus you know people like that and he says you know stephen wolfram said now there's going to be physics by by program right there's very few programs that are you can write in one equation programs are complicated and he says why isn't physics like that well pro you know in computing world now we have program by ai which has no intelligible equations in it or statements so why isn't physics going to do the same thing so it's it's going to be those abstraction layers down to the transistor eventually each of them will be replaced by ai some internal intelligible black boxes that resembles the transistors makes things that we don't even understand as devices yeah like people have been staring at the brain for how many they they still can't tell you exactly why you know the brain does anything twenty-one fat and salt and they see chemicals go back and forth and electrical signals move around and you know they're they're finding more stuff but you know it's fairly sophisticated one one of the future aspects of computing is security and you know we've had a wake of side channel attacks potential can of worms to do with the tricks that we use to make fast computers to what extent do you approach those security aspects when you're designing silicon these days are you proactive do you find yourself you do you find yourself specifically proactive or reactive no so the the market is sort of dictating needs so that you know the funny thing about security first of all is you know it only has to be secure if somebody cares about it right and then you know for years you know security like an operating system was virtual memory this process virtual memory couldn't look in that process virtual memory but the code underneath it in the operating system was so complicated that you could trick the operating system into doing something so so they basically you went from security by correct software once you couldn't prove the software correct they started putting additional hardware barriers in there and now we're building computers where the operating system can't see the data the user has and vice versa so the operating system can never so we're trying to put these extra boundaries in but every time you do you made it a little more complicated um at some level security worldwide is mostly with security by obscurity right like yeah nobody cares about you in particular because you're just one out of seven billion people like somebody could crack your iphone but they mostly don't care about it and most most you know it's it's a funny thing there's a funny arms race going on about this but it's it's definitely kind of incremental a lot of the discovered side channel attacks and they weren't that hard to fix but there'll be some other things and you know i'm not a security expert the overhead of building security features is mostly low the hard part is thinking it out and deciding what to do and you know everyone thought somebody will say something like well this is secure because the software does accent i always think yeah just wait 10 minutes and the software get more complicated which will introduce a gap in it so there needs to be real hardware boundaries in there and there's lots of computers that are secure because they don't talk to anything like there's both you know places where the computers are either behind such a hard firewall or literally disconnected from anything it's the only physical attacks work and then the guards so yeah it's going to be interesting but it's it's not super high in my thinking i mostly follow what's going on and then we'll just do the right thing but i have no face in security by software let's say because that always kind of grows to the point where it kind of violates its own premises you know it's happened many times you've worked at tesla and when you you designed a product specifically for tesla and you worked at companies who sell products to for a wide array of uses you know beyond that sort of custom workload analysis do you consider you know the myriad of possibilities of what you're building will be used for you know and the ethics behind what they might be used for or do are you just there to solve the problem of building the chip for the purpose well the funny thing about general purpose computing is it can really be used for anything and so the ethics is more a net good better than the net bad and you know for the most part i think the net good is better than the possible downsides so but people do have serious concerns about this the there's all big movement around ethics and ai and to be honest that the ai capabilities have so far outstripped the thinking around the ethics of that i don't know what to think about it like what what the current systems can do is already you know a stripped despair you know knows what we think and and what we want and what we're doing and the question is you know how many people have it have that and one reason to build let's say you know lower cost ai and programmable ai is you know i hope and we're talking to quite a large number of ai software startups that you know ai hardware and computing is in more people's hands because then you have a little mutual standoff situation as opposed to one winner take all but you know the modern tech world this has been sort of a winner's take all but there's actually you know literally several dozen very large companies that have you know let's say competitive relationship with each other so that's kind of complicated so i think about it some um i don't have anything you know really good to say besides you know the net benefit so far has been a positive and having technology in more people's hands rather than concentrated seems better but we'll see how it plays out um you've worked for a number of big personalities you know elon musk steve jobs to name two um and you it appears you still have you know strong contact with elon your presence at the neurolink demo last year with lex was not unnoticed i was i was invited by somebody in the neuralink team by the way you know i'm in elon i wouldn't say i have a lot of contact with him at the moment so but yeah i knew that development team never liked so i went over there to talk to those guys it was fun so you don't stay in touch with elon or not i haven't talked to him recently now it was it was very much a professional not a personal relationship when you worked for tesla yeah yeah because i was going to ask that um because elon's a big believer in cryptocurrency and regularly discusses it you know and as it pertains to demands of computing and resources for something that has no intrinsic value do you have any opinions as it comes to cryptocurrency not much not really i mean humans are really weird where they can put value in something like gold or money or cryptocurrency and you know that's a shared belief contract and what it's based on the best i can tell hasn't mattered much i mean the thing that the crypto guys like is it appears to be out of the hands of some central government whether that's true or not i couldn't say and whether how that's going to impact stuff i have no idea um but human you know group beliefs are really interesting because when you're when you're building things if you don't have a group belief that makes sense then you're not going to get anything done and group leaves are super powerful and they they move currencies obviously to politics companies technologies philosophies self-fulfillment you know you name it so that's a super interesting topic but the details of cryptocurrency i don't care much about except as a manifestation of some kind of psychological phenomena about group beliefs which is actually interesting but um it seems to be more of a symptom than a fundamental you know or a random example let's say um and yeah for a final question i wanted to ask you about going beyond silicon i mean we've worked been working on silicon now 50 plus years fully this full silicon paradigm has been optimized do you ever think about what what's going to happen beyond silicon if we ever reach a theoretical limit within our lifetime oh yeah yeah like you know my computer started you know with abacuses right and then yeah you know mechanical relays and then vacuum tubes and then transistors and integrated circuits and you know now the way we built transistors like it's like what i don't know 12th generation transistor they're they're amazing and there's more to do the optical guys have been actually making some progress because they can they can direct light through polysilicon and do some really interesting switching things but that's sort of been you know 10 years away for for 20 years but they actually seem to be making progress the economics of biology it's like 100 million times cheaper to make a complicated molecule than it is to make a transistor like that the economics are amazing like once you have something that can recon replicate proteins like i i know a company that makes proteins for a living and we did the math and it was literally a hundred million times less capital per per molecule than we spend on transistors now when you print transistors something interesting because they're organized and connected in very sophisticated ways rigid rays but you know our bodies are self-organizing you know they get the proteins exactly where they need to be so there's something amazing about that and there's so much room as if i'm going to say at the bottom of how chemicals are made and organized and you know convinced to go a certain way i was talking to some some guys who were looking at doing a quantum computing startup and they were using lasers to quiet down atoms and hold them in these 2d grids super cool so you know i think we've you know barely scratched the server zone what's possible physics is so complicated and you know apparently arbitrary that you know who the hell knows what we're going to build out of it so yeah i think about it um and it could be we need ai kind of computation in order to organize the atoms in ways that you know takes us to that next level but you know the possibilities are so unbelievable it's literally crazy so yeah i think about that i mean it's really great jim i mean thanks you for spending so much time with me i think i've got another 20 questions unfortunately we haven't had time to get to okay um but but good luck uh i said it before in the last interview but again good luck with tense tauren and uh look forward to seeing what you come up with okay great hey good to talk to you again [Laughter] [Music] you

Channel: TechTechPotato

Views: 57,518

Keywords: jim keller interview, jim keller, jim keller amd, jim keller intel, jim keller tenstorrent, what is jim keller doing now, jim keller zen, jim keller cove, jim keller ian cutress, jim keller techtechpotato, ian, cutress, techtechpotato, anandtech, people skills, jim keller engineering, engineering, arm vs x86, arm vs x86 vs risc-v, intel vs amd, intel vs apple, jim keller abstraction, abstraction, company culture, building teams, beyond euv, AI chip design, beyond silicon

Id: AFVDZeg4RVY

Length: 97min 11sec (5831 seconds)

Published: Mon Jun 21 2021