 Good afternoon everyone the head of EC and it's my pleasure to I guess pass the baton to team Rogers So team is an associate professor in our school He is actually working in the areas of hardware and software engineering and specifically on Programmable accelerators like GPUs Team is really the recipient of many different awards. He has won many research awards teaching awards Best paper nominator and nominations I'm just only going to mention that he has won the NSF career award and the College of Engineering Exceptional early career teaching award So right now team is working in expanding the rates of efficient hardware in the twilight of Moore's law and The rest will be on his talk Thanks, thank you very much to me tree and thank you all for having me here It's kind of an interesting presentation to create and give and I'm glad to have the opportunity So like everyone I'm going to start off with my origin story I am my maps a little bit bigger than the first presenter, but not as big as the second presenter So I'm from Canada that little like rim of of land right above The US here and I was born in the small Canadian province of Nova Scotia to two loving parents Anthony and Colleen Rogers and I grew up most most of my life growing up I was raised in a very small town of about 800 people called Chester, Nova Scotia on the coast of the Atlantic Ocean I had a happy childhood. I spent my winners playing hockey like a stereotypical Canadian my summers playing baseball I was never good enough at either of them to play beyond high school But I was pretty good at school. So I decided to keep going with that at 2001. I graduated high school And then I decided to move on and study and you know for someone in my field I'd never actually programmed a line of code at that point in my life I would just like math and I like science and I was like, okay I'm gonna go do something interesting and build things and so I went and I decided to go to a college called McGill University in in Montreal And this was a huge jump for me coming from a town of 800 people into you know I know many people from major are from giant cities But to me Montreal was a giant city several million people, you know huge huge French influence there and I spent, you know four years studying electrical engineering, which is not actually what I ended up doing now And I actually did an honors degree there. I wrote honors here for two reasons one We spell it weird there's a you on there and every time I write my CV word complains of me because that's the British way To spell it in America. You don't use the you And I had an advisor there that I actually you know did my honors thesis with Dr. David plant And I did my honors thesis and something very far away from what I'm doing right now the optical reception of gigabit ethernet Which was huge in 2005 And it was a very analog type of thing and I ended up sort of veering into the into the digital world And part of the reason why I did that was I thought that analog stuff was super interesting and Dr. Plant was also okay with me signing off from my honors contract after the deadline has passed because I've actually forgot Or I put it off too far and he was he allowed me to To research with him even though that I passed the University deadline. So anyway, I studied electrical engineering here and then I decided to not go to school right away I was ready to go make some money and get a real job. So I actually moved To Vancouver, Canada. So still staying above this this line Over in on the west coast and I actually moved with no job I moved there because my wife put it in a pharmacy school and I just followed her there And I was unemployed for like a few months and it was like trying to find a job and I found a job And I quote I put real job in quotations here because I actually made video games. I It was a real job, but it was actually a really fun job I worked for electronic arts. They have a big development office in Vancouver And I worked there for about five years and it was actually a lot of fun It was a super fun job. They had they have like a soccer field and an inline hockey field and basketball courts You can go play at lunch and I like played in intramural leagues in between coding up these games and you know It was actually a pretty cushy nice job And I liked it a lot and then I got promoted and then I decided to quit Because you know, I I'm not a heat like if you're passionate about video games This is where you should be but I wasn't passionate about video games I really wanted to understand machines better and try to make them more efficient That was what the most fun part of my job was so I decided to go to graduate school and pursue a degree in computer architecture And so I went to graduate school in the same city I actually took the bus out to talk to my talk to my future advisor from my job at electronic arts and we seemed to have You know see see eye to eye Professor Torah Mott he was my PhD advisor and I applied and then got in and then went to school here in Canada out here at UBC and I spent five years doing my PhD During the course of West I got to do some pretty cool internships I went down to Seattle to do an internship at AMD research I didn't put pictures on here of all the all the folks who you know, I'm gonna call out So I worked I worked with several folks there Brad Beckman and Gabe Lowe at AMD research Gabe is that now a fellow there and then also worked down in Texas at NVIDIA Mentored there by a guy named Steve Keckler all of which who's had you know vast influence in my Choice of problem and just trying to understand You know what it's like to be an industry and what it's like to innovate in this space So I did that and you know, I finished my graduate degree and then straight out of grad school I interviewed and managed to score a job here at Purdue and I came here in 2016 Oh, and I also mentioned that while I was a grad student. I also had two kids My wife and I gave birth to my daughter Erica in 2011 and then my son Nate in 2013 These are their first Halloween costumes. That's why they look strange What was the sunflower and Nate Nate was a Peas in a pot or something. I don't know just my wife picked them out But I thought, you know, they were kind of embarrassing pictures. They haven't seen this presentation. That's okay So so yeah, that's sort of my journey on how I got here So, you know, they gave us a list of questions on like what you're supposed to put in in this talk And so one of the things is like how did you get the job? And if I'm being completely honest and now I have ten years so I can be completely honest I got this job because of my very first paper It's the paper it went on to win a whole bunch of awards. It won me a bunch of fellowships independently funded my PhD, etc My most cited paper people in the field probably think about this paper when they think about my work And so, you know that you know, I obviously did a bunch of other stuff in graduate school But that paper is the reason I got top ten interviews and I bring this up because you know That paper the story of that paper is not all smiles and sunshine We did that paper and we submit and it took a lot of work It took me like a year and a half we submitted it and then this is an empirical field We found a bug in the code that we wrote that study the design of the paper and it changed the results and It changed the results in a negative way and we had to withdraw it So we actually had to withdraw the paper because the results were wrong and the idea was actually dead for like 10 months And my advisor said to me at one point. This is actually a sad little paper Now I love my advisor And I have very good Relationship with him now and if you say if you you know if I reminded him that he said this he might not remember it because it's also one of his you know Landmark papers, but you know, this is sort of you know, it was a really trying time Right, we submitted this thing and we thought it was gonna work And then and then it got we had to retry because of an error, but I stuck with it in the end I didn't give up on it I kept pushing on the idea and actually one day while I was running and this is a piece of advice I'd like to give anyone who's trying to do any kind of creative or deep Piece of work Exercise right exercise is one thing that I love to do and I love to like think about things while I'm doing it And I actually came up with an idea to save the paper so something we had done we could change and when we changed it the results got better and Then I sort of worked on it And I worked on it actually at night when I was an intern because I got this idea when I was interning and Working for AMD during the day and then working on this paper at night We submitted it and then it got in and then sort of the rest is history so You know, I think that some one of the questions here is like how do you succeed and one of the things that I want to sort of point out on how What I think it takes to succeed is to just keep trying Okay, there's this actually trend on Twitter academic Twitter of like posting all your rejections, right of like things so, you know, that was a huge You know, you know, not done life is not just one sort of unbroken stream of successes, right? And when I first became a assistant professor, and I think many of us have these same stories You submit you submit you submit you get rejected rejected rejected my first seven grant applications all got rejected 15 pages were the work a whole bunch of mental effort and didn't work out And then eventually it hit and then once I started to hit they started to hit continuously and then things got things got a lot Better so just sort of persisting and trying to find a way to make something work is really the best piece of advice that I can Give you generally in any technical field don't give up There's always a way just keep pushing at it. And you know, that's really what what has made me Able to succeed and I really do owe this to my parents all these intangibles that people have about themselves This comes from my background of my parents sort of teaching me that you know, you don't give up on things keep pushing and The other thing that I think has been helped me succeed is to be really passionate about creative ideas When I come up with an idea and I publish paper. I feel like a rock star I know society doesn't see us that way But I feel that way and I have an anecdote actually that sort of nicely matches this You know a couple days after I submitted that first paper I went to a rock concert with my with my wife or yeah, she was my wife at the time She was actually pregnant with my first child And the guy on stage was talking about like their career and they were saying that like they were playing in garages They were a couple years, you know just you know milling around nothing was hitting for them He thought about like going back and working as a janitor or something And then he wrote one song and it took off and then they were super famous Okay, and I sort of I felt that connection with what he was saying with the research I had been doing that it was like I was milling around trying really hard and then I did one thing and I felt like that thing was going to take off and It turned out to actually be true in my case And you know ever since every idea I sort of feel this passion that like I'm inventing something My name is on this thing and you know, I think it's I think it's pretty special Another thing that's helped me to succeed is to do stuff that if you thought about it for a really long time It's probably a bad idea. Okay, so quitting my cushy job that I really loved to go back to graduate school a Lot of people thought it was stupid idea. I talked to my colleagues like why are you doing this? I actually went back unfunded in Canada the funding situation is a little bit different But I decided just to do it anyway because I wanted to and you know I love the fact that I did that because it set me down this career path Having kids is a poor grad student Many of us maybe have done that people told me that was a stupid idea too. One of my favorite things. I've ever done Pushing on an idea even when your advisor said that maybe it's a sad little paper Now again, I'm bagging on my advisor, but I love the guy if he hadn't said this I probably wouldn't have kept pushing on it just my personality, right? I just I want to prove that he's wrong And that sort of has you know been good for me even moving here, right? I'm a Canadian. I went to school in Canada moving down here with my family and everything was sort of a jump So everything that I've done all the favorite things I've done. I sort of just did them and sort of dealt with the consequences afterwards and I think that You know those are the things that I enjoy the most so I managed to talk for over five minutes And I'll talk about anything about what I do so obviously that's why I'm a professor So what am I passionate about so what am I what do I actually do? I do computer architecture so what I do is I design systems and the hardware that runs those systems that Really power the world. It's very empirical field. Like I said, we can have bugs that could make your results incorrect and the nice thing about it is that What we study really depends on what people are running for software So these last two folks that you know had really cool things that they were that they were doing That's driven by software that's running on hardware We're sort of like the money of Science right like economists You know sort of trying to figure out like where the money is going to go We're providing the computation that's going to allow science to do whatever it wants right we're sort of embedded in everything because everything runs on a machine and how that machine is built is determined by us and I love the I love that fact and that means that it's always changing always moving. It's never static There's always going to be something new that comes up now. It's machine learning. Thank God for machine learning from a you know From a sort of interest in our field perspective, but it's sort of you know Continues to push things in weird new directions that we can design new hardware for which is exciting So what is my research specifically look at so there's this general tension between how easy a machine is to use and how Efficient that machine is so I've thrown a few design points out here of you know typical types of machines You might see I you know up in the in the top left hand side here is something that's really easy to program Which is like a traditional CPU core for anyone who's ever programmed before this is probably the first thing we ever programmed on There's actually a lot of really smart stuff in the background trying to make that run well, but it's not super efficient You can go the other way be a hardware designer and know exactly what you want and specify it and build an ASIC for it But that's really hard So only a subset of people that can really do that well and there's all these designs in between So what we're genuinely trying to do is create a piece of hardware that's efficient But that isn't so challenging to program and there's a whole space of different things that we can explore there to try And make that happen So you know it's not a research talk, so I just want to give you a flavor of some of the things That we do so one thing that we took a look at is Could we run object-oriented code on a GPU? So anyone who's ever programmed before you probably maybe programmed in an object-oriented language. Does that work? Well, well, no It doesn't work. Well, so you can come up with some clever tricks to basically Optimize that program to try to make it run well on that really efficient piece of hardware We also study giant hierarchical machines So you know as as machine learning in particular gets more demanding it needs more compute We need to stitch a whole bunch of machines together to do something useful Those machines are now being composed of these things called chiplets because Moore's law is ending and it's really hard to make big chips So we have to make you know bigger chips out of little ones and then put them together on the substrate What does that do to programmability? How do we make that work? Well, and then finally we're you know The most recent thing we're doing is actually looking at different types of accelerators in the data center So outside of machine learning, which is obviously drawing a lot of this all of the web services that you use on your phone Etc are running with these things called microservices on the day in the data center could we develop accelerators for these things? We think the answer is yes, and we've got you know some work in that direction. So Coming back to sort of a more higher-level point. Why do I love this job? I love this job mostly because I can do what I want. I have no boss I've worked in industry and you know, you're always answering to somebody Trying to Not sure what this keeps ringing. Can I just take this off set fine? All right What's that? Okay. All right. Thank you Yeah, so I don't really have a boss. I mean Dimitri's here But you know, he's not gonna tell me what to do I get to decide what I want to do based on what I think is interesting and where I think a field is going and what I want to pursue You know Intellectually, there's really no other job like this, right? You know, you talk to people in the real world with real jobs They don't get to do all this stuff It's it's a perfect mix of innovation trying to come up with new things mentorship mentoring students And then teaching which is like this whole other side of things where you get to stand up in front of a bunch of people and and show them something that they never knew before and there's no other job in the world that's like this, right? You know, I think people in our field You know with a phone call 24 hours, I could probably triple my salary and live on the beach but you know, I don't want to do that and The reason is that this job is just so unique and I love the fact that you get to do all these different things and do What you want and you know, you don't have you're not beholden to some sort of corporate bottom line Which you are when you work in industry and the the other sort of big piece of this is that you have an exponential impact you go and you graduate students and then they go off and do stuff and You're creating these other entities in the world in the form of your students that are going off and doing fantastic things And I've four of the students here that I've graduated have gone off and and done that Some of them have taken the money by the beach and good for them You know if that's what you want, I'm glad and happy and you know Some of them have actually gone on to be professors themselves, which is just kind of a cool thing So what do I think is next so what do I think is next for our industry? Well, Moore's law is ending and this is actually a great piece of information for For us as architects, there's going to be billions made in the fallout for decades Basically the big companies like Intel have been able to bury any really interesting architectures because they could always beat you on process node because Moore's law would kill you Moore's law is dying and you know that we've been saying this for a long time, but now it actually is You know, we can't make these things any smaller So we're gonna have to find other ways to get you that efficiency and that means changes in architecture changes in systems Increase specialization. We're seeing that with accelerators. That's the stuff we do We're right in the meat of what's gonna have to come in as Moore's law dies and that is super exciting I've got tenure now and this is happening. So what am I gonna do in this space? You know, I'm not exactly sure yet. I actually just got tenure. I think officially on August, right? I think it takes forever by the way if you're a student right, I think I submitted the package like 18 months ago or so anyway, but whatever it takes a long time, but I officially got tenure I think in August and so you know just trying to figure out what's next I actually am intrigued by this ideal of Idea of commercialization just mostly because it seems like something. It's a bad idea that maybe I should do anyway But you know, it's also kind of maybe satisfying to take a product and commercialize it right because we're academics We send things to journals and conferences other academics look at it and say oh This is interesting but to really test something to see if it really is worthwhile I think you know capitalism actually is pretty good vehicle for that. I think I hope So that's something we're sort of thinking about Many thank yous my graduate students some of whom are here if you're not here If you're not on this slide and you're here just slack me a photo. Sorry knee I didn't have a photo of you. I think I'm at I took that off GitHub. That's like a digitization of your face, right? So I my apologies because I put this together last minute So I didn't actually ask solicit photos, but you know, I owe everything to my students You know everything I've done as a professor everything. I've done in order to get tenure You know is based on their work and I'm that's sort of the thing that I'm the most proud of is the is the people that you know We've managed to train and create and continue to go on to do cool things and that's it I think I did it in 10 minutes ish. Thank you very much Thank you so much. I don't know about you. I learned a lot Including that we can lower team salary. It was All right, so time for questions Yeah, for somebody doing System architectures, what do you think is gonna be in my laptop in 15 years from now? And how do you think your group would have been contributed to that? So there's gonna be many different things in your laptop in 15 years We're already seeing that you know right now your laptop is primarily a CPU and a GPU for the most part They're doing most of the work We're not gonna be able to keep selling you new laptops with current CPU GPU technology Because of the fact that Moore's law is slowing down So if we want you to buy new stuff, we're gonna have to put new things in there that makes your battery last longer that makes stuff last longer so the diversity of things that you're gonna see in your laptop is gonna increase and That's exciting from my perspective because we design hardware So the more hardware designs that people are willing to pay for The more innovation will be in that space as opposed to everything just having to run through your Intel CPU Which it sort of still does today So I would I can't say for sure what'll be in there You'll probably I mean you There are already machine learning accelerators in a lot of your phones, etc So that's one way piece that's already sort of working its way in what's gonna happen in 15 years. That's You know, whatever happens, we'll do something to try to make that better But I think you'll see a vast diversity of different things inside your laptop in 15 years Very inspiring story The question is on your little sad paper that might have generated many good papers now Yeah, it was a good paper. He called us that paper Maybe to motivate me. I don't know but yeah it go on but Did you help did it help you to generate creative ideas around it? How to make it successful? Because it was a sad little paper. I think so because it was like this thing is dying You know, I think it depends on your personality I feel like if there's no sense of urgency, I won't do anything not everyone's like that But I am so I think the fact that it was a sad little paper gave me a sense of urgency that hey I better do sing but tenures like that too, right if you have to you know, you have to get tenure You have to you know do a bunch of stuff by this date or else you're not gonna get it So it definitely deadlines focus people, right? And I think that that's definitely true for me Yep great talk a Technical question, so there's a lot of talk about with the end of Moore's law about Advanced packaging and heterogeneous integration playing a more important role What does that mean to a computer architect? Well, it actually means a little bit of what we did in in this middle project here, right? We're basically Chiplets are gonna appear right we're gonna package things and and not these monolithic dies We're gonna have some kind of interposer that connects chiplets together And we're gonna have weird characteristics that we didn't have when everything was all happy and cozy on the same die, right? I mean we're sort of gonna it's almost like we're gonna go backwards because for decades. It was like You know the IC there wasn't much we could put on it then Moore's law came and then everything went into the IC now It's like okay. We still want to put more stuff on here, but the IC can't get any bigger So we're gonna go back in some ways and that's gonna have all kinds of interesting challenges from you Communications perspectives from how do you program something like that? Like is your laptop always gonna be composed of chiplets and all the software that needs to run on that? So I think it'll pose a lot of interesting questions I mean, I think that this is true with a lot of fields and science they go in cycles, right? I think we're kind of gonna bounce back a little bit and have to look at things that were important pre Moore's law if that makes sense and One quick question. So if we're working towards efficiency of the design, how does that work towards efficiency? Of like energy consumption on the chip like is it possible for us to get the same amount of output with much less power draw in the future? So it's been mostly pot that Moore's law has made that somewhat easier for us, right? Because you know basically the things get smaller they get more efficient and with the same design you can basically get more energy efficiency What the argument that folks in my field make now is that does that ends you'll have to get efficiency in other ways So you'll have to you know one very concrete thing that's happened today is specialization where you have specific Hardware components that do a subset of things very very efficiently But then how do you use the subset becomes a big big challenge? How do you program those things you can't just you know dump down some Python and have it work? so I think that I Think to sort of answer your question We're gonna have to find ways to make things energy efficient without the underlying technology being better, which is part of the challenge Okay, thank you. I Think that was the last question based on the time we have All right, let's give it on the flaws again for