짐켈러 “엔비디아 시대는 결국 끝난다, 다음은…” / KBS 2024.03.29.

yeah so people sometimes think I move around a lot I've been working on computer design for 40 years so it's somewhat natural to work in several places so when I joined Apple we were we were starting to make iPhones and we wanted to make very high performance iPhones and I think we went from a a pretty standard phone chip to a very high performance phone chip and we solved most of the technical problem uh when I went back to AMD um it was clear that AMD didn't have the right design or design team or something they didn't have a very good product and you know I worked to pull together the team the design methodology and the design to create Zen which became a great product when I was finished with that project I was looking for another hard thing to do and I talked to Elon about building an autonomous driving chip which is a really exciting challenge so I went to T Tesla and we built a chip to drove a car in 18 months and it became the world's best autonomous driving chip I I like to work on the Innovation side and the team build I mean there's there a MD is a really interesting company they're a great Graphics Company run by a great CEO think great um they invested in multiple Technologies outside of Nvidia for years you know when when AI started to run on top of Cuda it wasn't because they got lucky one day they've been working on it for 10 years right it's a big investment and they the thing I really I like about companies that have some Visionary leadership is now big companies almost always have their day of the sun when something happens right for many many years Nvidia great GPU company building games now they're providing ai ai software stack and systems so they're becoming more vertically integrated um sometimes that works out for a while but it's hardly ever stable so many companies now are investing a lot of money to go build their own AI chips in house some of those will work some of them won't first of all gpus are not the best thing to run AI software on well they're they're better than regular CPUs but gpus were built for graphics and and we decided to build a processor that an AI chip that's built around tensor processors right which means you have a very fast math now the GPU companies AMD Intel Nvidia are all adding tensor processor units to their chips but their software model still depends on the GPU software Stu which has been successful so far but I don't think it's the most option and and that's the position that we're you know the design we're doing chat gbt or llm find a website that I can buy a plane ticket get me the best price to go to Soul 10 years from now almost all your devices will be smart your refrigerat you could say hey what's in my refrigerator and please put an order for this that other thing and it's going to happen and then at what point will you say your life changed yeah but people always think about is you know people ask me how are you going to compete with Nvidia it's a $ 1.5 trillion doll company they saw hundred billion doll of very high priced AI computers that's not our Target our target market is there's hundreds of AI startups software and Hardware that want to build their own products you know at lower cost and they need more open access to that kind of right so the RIS five architecture came out of Berkeley University started by David Patterson who I know was a great guy um it started out as a student College project but then multiple companies said this is really interesting because we can build our own computer and we can do our own kind of changes to it and so that got out in the world and then at 10 stor we think think computers are going to evolve a lot in the next 10 years right so I want to be able to build computers with the ideas best ideas we have as fast as we can and I can't do that by either using Intel architecture which is somebody else's I can't license it or arm where I can't change it right so a RIS five I can change it in the short run you know the arm architecture and software is very refined Intel is very refin find it's very expensive so if you want to make a product today that's probably a good answer if you want to own your own product and you want to be a evolve over time and go maybe someplace where nobody else has gone then risk 5 is the only choice it's the only open [Music] architecture so everybody has their ups and downs I worked with Samsung in the '90s they helped us build Alpha chips work with Samsung and Apple and they they were a great partner work with Samsung at Tesla where they help build the autopilot chip I work with Samsung when I was at Intel where they build a number of chips for us and my personal experience I I can't tell you about Samsung as a whole my personal experience is every project I've ever worked with on Samsung has been successful they go very fast and then you hit a plateau or it's management change you know I don't know all the details that's Samsung but uh you know Samsung's been a great company for a very long time I'm also doing business with DMC so every time we build a chip you you look at all the pieces and then say what's the best what's the best solution we talk to lots of people for our next generation and samon the best let's say whole package and you know and I I like working with them obviously so we we spend a lot of time talking about how do we work together a product that's you know that's really great so they had a really good plan well Samsung still makes a lot of products and they have lots of really good technology like you say they still make the best DS in the world I think uh they make some of the best camera sensors so I'm talking to a couple companies now and they want me to help them solve their problem but they want the problem to be in this one area they said we think our problem is AI and I think your problem is not AI your problem is how your organization does design I can't I can't fix that from outside if you have access to the leadership technology you'll buy it but if you don't you will invest more money on the follow on technology and I think China's got lots of great Engineers I'm not a political analyst what I do know is uh Chinese companies have invested a lot more money on AI semiconductor production equipment production because of restrictions and they're making good progress so mors law uh was defined as you know transistor density doubles every 18 months and then that slowed down the 2ear and then they sort of shifted it the computer performance will double every two or three years right and then and then that started to slow down for for a number of reasons for a while it slowed down but there's a breakthrough so and uh I've given a talk about that I called it cascading diminishing return curves right right so so the the one curve slows down but then the invention happens and you got right and your brain is something like 10 to the 18th operations a second for 20 watts that number is unbelievably better than the best GPU today we're like six orders of magnitude away from the same performance per per L right so people say and I people told me Jim you don't understand you can't sequence atoms at low cost with no power it's was like you literally did nobody will ever make a computer with the same efficiency as a brain even though everybody has right and even weirder every every cell in your body has a factory in it to make a whole copy of like so we have technology problems but we don't have a possibility problem say I'm interested in that [Music] kind he what should I study to be a chip designer and I'd always think math physics programming engineering like get a good Basics and then read a lot of books right if you want to build something new or do something creative you have to spend some part of your life being creative or you know go see really good movies or have something you really like to do but the basics matter more than the details of today like today everybody all the college students are studying P torch 5 years from now pie torch won't won't exist the next 10 years could be f and you you need to have your mind trained to be open you need to be somewhat Broad and really good s you know there's a really good book called structure of Scientific Revolution by joh which talks about how you make substantial changes and that's a fairly complicated topic and not that many Engineers have read that book despite the fact that their whole life is in technology change so it's a it's a Curious [Music] Thing

Channel: KBS News

Views: 1,040,049

Keywords: AMD, KBS, KBS NEWS, KBS뉴스, NEWS, format:기타, genre:경제, location:시사제작2부, person:서영민, source:영상, type:디지털, 뉴스, 뉴스_기타, 맞춤칩, 반도체마이더스의손, 반도체의신, 삼성잃어버린10년, 삼성전자, 애플, 엔비디아이후, 짐켈러, 테슬라

Id: 2pw-YZ7KuFY

Length: 10min 48sec (648 seconds)

Published: Thu Mar 28 2024