Video
Đăng bởi Ban biên tập OpenER ·
Tác giả giữ toàn bộ bản quyền (All Rights Reserved) — không phải giấy phép mở. Bạn vẫn xem và học miễn phí trên OpenER; muốn tái sử dụng cần xin phép tác giả.
DAN GARCIA
Okay, perfect. Let me look at participants. Welcome, everybody, to the last class, 61C. So happy to see you all here.
45 — see if we can get to it. Boy, it would be great if we could actually get to a hundred.
Boy, that would be wild — one out of every 10 students showing up for the last live class. [LAUGHS] What a stretch that would be.
By the way, before I start, let me share that it's important for you to sign up and do the course evaluations, course-evaluations.berkeley.edu. We need to get up to 90%.
You get a free point. So please. This is just wonderful, wonderful to sign that up. So do that while you're waiting.
And we will start recording. Actually, recording now. And we will start in a second. All right.
Bora, you feeling okay? You ready?
BORA NIKOLIC
I am good.
DAN GARCIA
All right, good stuff.
BORA NIKOLIC
Are you driving the slides?
DAN GARCIA
I'm driving the slides, yep. You can all see that. Hopefully, you can all see. We're not doing mm-hmm today because of audiovisual reasons.
You'll see the reasons why. But hopefully, you can see us on this week. Probably should pin ourselves too. I just realized we should probably — is there a way to — ahh.
When you're driving screen, you lose all the windows. Let me see if I can pin myself.
BORA NIKOLIC
That's right. I can pin myself.
DAN GARCIA
Spotlight here. Spotlight myself and spotlight you as well. How about that, Bora?
BORA NIKOLIC
I got it perfect.
DAN GARCIA
And that way our TAs are not being spotlighted. They don't have to worry about being online. Perfect, okay. So let's get rolling.
Welcome. We got to 77.
BORA NIKOLIC
All right.
DAN GARCIA
Ladies and gentlemen, welcome to the last class in 61C, lecture, I think, 45. It's amazing. Friday, week 15 — woo!
BORA NIKOLIC
I mean, can you imagine that we're almost done?
DAN GARCIA
Can you believe it? I can't believe it. This is incredible. I mean, what a road this has been.
Holy moly.
BORA NIKOLIC
Yes.
DAN GARCIA
If you had told me the bumps in the road, I was, ah, come on. There won't be bumps in the road. Bumps in the road? They're mountains.
You're all here together.
BORA NIKOLIC
I can tell you this was the craziest semester that I've had in my 21 years at Berkeley, period.
DAN GARCIA
Me too. Me too.
BORA NIKOLIC
I mean, and the craziest one. And then — yeah. And then comes the spring 2020.
DAN GARCIA
Well, there's like three things that are distinguished this year. One, it's the craziest semester.
The second is that these students are walking into the final with the highest grade I have ever seen in 61C. And I've taught this, I think, 10 times.
I've never seen this histogram, this total course points, going to the final. So these students are as primed as they are.
All that work you guys put into project 1, 2, 3, and 4 and the studying for the exams, wonderful. It has really paid off in terms of your performance.
BORA NIKOLIC
But not only that, we get some feedback about the performance in interviews.
Our friends at Apple told us that all the firmware positions that needed basically understanding of C and how does it work with hardware were sold to Berkeley students, Berkeley undergrads, five days after posting.
That's it.
DAN GARCIA
Wow, wow.
BORA NIKOLIC
I mean, nobody failed. Apparently, everybody passing through.
DAN GARCIA
Wow. Wow.
BORA NIKOLIC
I feel really good about that.
DAN GARCIA
Congratulations. Congratulations, everybody.
BORA NIKOLIC
I mean, I really feel good about that. It all feels like it was worth it.
DAN GARCIA
And you made Berkeley proud. I mean, this is what we are. Our job is to set you up. We're the coach.
Like, we're the ones on the corners in the boxing ring. You can do it. You're a lean, mean 61C machine. And we send you off into the round.
And you've all done very, very well. So great job, everybody.
BORA NIKOLIC
And you also create opportunity for the next class and the class after that. And it feels good. It feels like it's worth doing all of this.
DAN GARCIA
I love it. I love it.
BORA NIKOLIC
So what we have done in this class — let's try to recap it a bit.
DAN GARCIA
Sure.
BORA NIKOLIC
So we tried to convey six really great ideas in computer architecture. So we drilled on the abstraction. So we have been talking about building really complex systems.
But we can't build these complex systems by having everybody always touch every gate at the very bottom when writing a very complex system.
DAN GARCIA
Too much detail. Too much detail. You'll just drown. You'll drown in detail, exactly.
BORA NIKOLIC
And too many things going on. So we build these layers of abstraction. And these layers have served us really well.
We have been able to tackle them layer by layer, going from top to the bottom and back from bottom to the top.
And I'm hoping that everybody feels really comfortable now understanding how does a computing system work.
That's the first big, big point that was drilled down throughout the class. The second one was this thing of Moore's law that got us where we are now.
And there is the scare that Moore's law may be going away, that the technology is not going to continue scaling at the same pace as it did before.
And that's true. But it is still getting better. And new things are going to happen. And we just need to understand that we are going to have to deal with power limits.
We need to figure out how to go towards designing systems with slightly better technology and a little bit more energy efficient. So we'll have to deal with design under power constraints.
And that's going to be big. I mean, that's the task for your generation. You're going to be dealing with that.
There is no miracle technology that is going to come and suddenly start scaling into subatomic levels. You guys will have to figure this out. That's the mission for your entire generation.
And you know what to do.
DAN GARCIA
And quantum may be coming. That's going to be exciting. When quantum is here, that changes everything. I can't wait to teach the new 61C with quantum as one of the things we'll use.
I just can't wait for that.
BORA NIKOLIC
Yeah. I mean, but don't hold your breath right away. I mean, we've got to make do — DAN GARCIA: (LAUGHING) I agree. I agree.
BORA NIKOLIC: --with CMOS for another 10, 15 years or so.
DAN GARCIA
I agree.
BORA NIKOLIC
Exactly. Exactly. I mean, we talked about that. We are seeing glimpses of this using specialization.
And yeah, we have to change 61C as we go along, the big thing. I'll get back to that. So the third thing that we did is this principle of locality and memory hierarchy.
And that is huge. It looks like we are working with this infinitely fast and infinitely large memory system, although it is not. It is limited.
I mean, we most of the time are working with those few registers that we got in L1 cache.
And everything else is there just to show up right in time for us to grab it and do computation.
DAN GARCIA
It's a beautiful abstraction to have the abstraction to make you think you get this benefit of infinite size and infinite speed.
L1 speed at size of disk — that is just a great lie that we tell you. And it just works. I mean, it is wonderful.
A lot of, obviously, engineering behind it, but it's just pretty cool that you can think that way.
BORA NIKOLIC
The other thing, keep in mind one of the challenges you'll hear a lot of people talking about, oh, there is some other technology that is going to come and rescue us.
We are going to do non computing and so on. And typically, when you drill down, they forget about this principle of memory hierarchy.
And the life — working without being able to build these luxury applications that we have running on top of the operating system with multithreading and many processes and all of that working simultaneously and not impacting each other, that is going to be really tough like.
I don't think we're going to see much of that. So keep in mind this is a very basic principle. Don't violate it.
We do need infinitely big and infinitely fast memory built out of cheap components.
The fourth thing that came initially to help out with the issues with power limitations and slowdowns of Moore's law is the parallelism. That is a huge thing that we drill through.
So if you took this class 10 years ago, there was almost no parallelism.
DAN GARCIA
That's true. That's true. Gate level — there was some gate level, I'm sure.
BORA NIKOLIC
But that's not how people taught it. Dan, you taught the class back then.
DAN GARCIA
I did. I did.
BORA NIKOLIC
So things now, parallelism is in the forefront. And what we are going to see going forward, we are going to see specialization, a different kind of parallelism.
We are going to have these units that are really good at their stuff. It's not just CPUs, and GPUs, and everything else.
There is going to be a whole bunch of different units that are really good for their particular stuff, for particular things.
We need to figure out how to run them all together concurrently and efficiently. And we need to program them really efficiently. And that's a pretty big thing, pretty big deal, how to do that.
The fifth big principle is about the performance measurement and improvement. And this is one of the first — we didn't tell you that early on.
But that was one of the big Patterson's contributions from late '70s. Before that, building computers was like a black magic.
And the first edition of Patterson, Hennessey's book actually set up the right way of thinking about computers. First, let's try to figure out how do we evaluate performance.
And then let me know what we are dealing with. We know what to improve on.
And there, what we talked about in the class is just IPCs and the iron law of computing but also different ways of how we measure performance.
Because taking just one angle of measuring the performance often gives us a wrong idea how the actual system is going to perform from user's point of view.
And we mentioned this multiple times. AMD is known for beating all the benchmarks. And users don't like it in the end. They've gotten a lot better over time.
But this is something that we are going to see, especially going forward in the era of specialization. We have the Apple M1.
It's a different system. And it is going to be doing some things better. But some other things may not be. Some applications may not be running quite well.
And then the sixth big idea we just touched on is dependability by redundancy. It is important. It is what makes our systems run.
And we can rely on them for doing really, really important things, like health care, and life support, and finances, and everything else.
DAN GARCIA
And self-driving cars. And self-driving cars. When there's a mud splotch on my — on sensor on the right, does my car continue to drive? It does.
How does it do that? All of that.
BORA NIKOLIC
Now, that also is entangled with everything else.
I mean, we need to — there is this really, really important thing, where we need to figure out how do we measure how dependable our system is. We have put some numbers over there.
But that's something — if you look at all these ASIL standards, people are trying to matricise that. How often are we allowed to have our self-driving engine fail?
We are going to see that — we are talking about quantum. Those are going to fail a lot more often, at least the way how we see them now.
They're going to be less reliable than what we have now. We'll have to live with that. They're going to be wonderful. But they're not going to work all the time.
This is going to get to the forefront. That's it. I'm going to switch over to Dan.
DAN GARCIA
Sure. And what's interesting is I remember Dave has this wonderful line from point 5, which was great metrics yield great performance. Great metrics yield great improvement.
If you don't have a metric — my user interface is better than your user interface. Really? On what metric? There is no metric.
It's a soft, squishy one. And that's why it's hard to measure these things that are squishy. You make a really clean metric. You can optimize.
And by the way, you know what they're optimizing for now? This is something probably, Bora, in the last five years, I think. Used to be, what's your performance?
What's your spec rating? Now it's spec rating per watt. It's, how power-efficient are you? That's a whole new way of thinking.
How efficient are you per watt? Everything's mobile. Everything's on the laptop. Certainly, the Mac Pro plugged in, who cares how many watts you're using?
But if you're doing anything else — laptop, mobile, watch — you need to have performance per watt.
So that's a big new idea way of thinking that wasn't there, I think, five years or maybe even three years ago. It was just raw performance.
BORA NIKOLIC
Yeah, and that's a huge thing as well, I mean. And it also needs to be shaped up over time. And it's not just instructions per watt now. It is often some particular task.
How many facial recognitions can we do per watt and per second.
And that's a pretty big deal. When we are looking at a phone, that facial recognition engine has 20 milliseconds to recognize you. Otherwise, you are not going to like your phone anymore.
You're going to give up on that. 20 milliseconds under a limited power budget is a pretty challenging compute task, a lot more challenging than project 4.
[CHUCKLES] But project 4 does lay a good foundation for building something like that.
DAN GARCIA
Sure. Let's keep on moving. Good stuff.
BORA NIKOLIC
Yeah, you go.
DAN GARCIA
So this is — I mean, Bora talked a lot about this. This was the picture I used to teach when I taught this in 2001 with Dave Patterson.
This was our course — machine structures, the old school model, just thinking about a single machine — a single core, even — and trying to look at all the layers of abstraction.
We talked about great idea number one.
With the ISA being the most important, we spent a lot of time talking about the design of this new, wonderful, open ISA that we have, that we've learned about, that our team, Berkeley's team, has contributed to.
Just love that. So we spent a lot of time thinking about this slide. We also talked about this slide, which is — it's a lot more complicated.
You've got not one machine but multiple machines. And every machine has multiple cores. How do they talk? How do you have cache coherency?
What are the issues? What's the fourth C? Parallelism came in at many layers. We've got parallelism with requests, high level.
We talked about thread level parallelism, parallelism instructions through pipelining, parallel data. Let's do some SIMD and do things all at once. And hardware descriptions all operating at the same time.
So we talked about parallelism, the many, many layers. And I think all of you are fluent in all five layers.
You can imagine a future project 4 might involve somebody touching every single one of them, some project that actually involves if you really optimize it, you touch on every single one of those.
And that would be really exciting.
Maybe future TAs could think about a project that actually makes them — to get that number, to get the top of the score, you actually have to do something explicitly about that in a clever way.
BORA NIKOLIC
May make you wish to take 61C again.
DAN GARCIA
Yeah, right?
BORA NIKOLIC
By the way, we have a hundred people now.
DAN GARCIA
Hey! We got [INAUDIBLE] — yeah, that's great. Wonderful, wonderful.
BORA NIKOLIC
Ah, it was a hundred a minute ago.
DAN GARCIA
Yeah, yeah, yeah. It came out just before — just as you said it. And this is also a big idea here. We have connected — okay, I'll tell the story.
This was the first time. My own version of this class I took a hundred years ago, 1980 — boy, man, 1986. I'm dating myself. Bora, you're probably similar.
You and I are the same age, I remember learning this, and sitting in class, and realizing — it was called 6.004 at MIT. It's basically 61C. They are almost identical classes.
And I remember saying this is the first time I understand it all. This is the first time I understand the connection between how software works at the highest level.
Here's C. And above that, here's Python and things above that. Here's something. Here's Snap! above that. All the way down to — BORA NIKOLIC: Dan, there was no Python back then.
DAN GARCIA
I know, right? Right. Python was only on the TV show, Monty Python. It was a gleam in Guido's eye.
So it came down. I assembled down to machine code. Now I have an executable program, just 1's and 0's. How do you run that?
At the point of me walking into that class — and hopefully, you walked into this class, you didn't know what that was. Those 1's and 0's meant nothing to you.
Now they do.
You could literally — if I give you on the final exam 1's and 0's from a final executable, a simple thing, you could actually go all the way back up to C and understand what the heck that means.
And you could also build a machine that does that. And you did that in project 3. So I think of this where it's almost like the two connections between hardware and software are connected.
And I want people — I want the 101 of you to actually — if you can, get ready here. I want you to clear out. Step back from your computer.
Nobody else is on the screen except for Bora and myself. I'm going to try to do this if I can. I got a background. I got to do this thing.
I want you to stand up. I want everybody to stand up. Here, let's do it. Everybody stands up.
This is a moment. This is a moment for celebration.
BORA NIKOLIC
I'll zoom out. I'll zoom out.
DAN GARCIA
I can't zoom out.
BORA NIKOLIC
I'm zoomed out.
DAN GARCIA
Bora has more — Bora has higher and higher. I'm going to show you my chest. That's all I got. I'm standing up.
I'm standing up, okay?
BORA NIKOLIC
I can wave.
DAN GARCIA
Everybody standing up. Here we go. This is the moment where you made hardware and software connection, where the connection between hardware and software comes together. [IMITATES ANGELIC SINGING] The first time ever.
I want you to call home. Get your phone out. Call home. We've done it!
Celebration! [MUSIC - LUDWIG VAN BEETHOVEN, "SYMPHONY NO. 9 IN D MINOR, OP. 125"] Woo!
Do a dance. [CHOIR SINGING GERMAN] We did it.
Woo.
Woo. Thank you so much.
BORA NIKOLIC
And zoom back in.
DAN GARCIA
Zoom back in. Now, I will say this is all weird. And it's all virtual. So you don't get the equivalent.
But when we would do this, Bora, we'd do this live, we asked the entire class to stand up, and walk around, and dance with us around the class.
And we would have everybody — I just want you to remember that moment, that there's never been a time before where you've really been able to connect that.
And the moment that you actually finish, the last tie when you connect the two railroads, the last tie that was connected was when you figured out the control logic to make the RISC-V machine work.
That was the last — the moment that hit, that's when that connection made, where you could actually teach it from the top down and the bottom up, and build it there, and tell your friends.
So I encourage you to try to teach your friends how this works, to share some of the wonders of it. All right.
BORA NIKOLIC
And there is a benefit of being remote this semester. We didn't make you call home right away at that point. [LAUGHTER] Otherwise, we would have.
DAN GARCIA
Right. That was a big deal. It's hard to do to this asynchronously when we make a video and ship it off to you. We don't know when these things happen.
But nice to have a shared experience. These shared experiences are definitely missed sometimes. All right. We've got a pretty wide open calendar here, folks.
You've got two weeks. This is a Friday. This is the summary lecture. You've got all the rest time to put things together, to get yourself gathered together.
You have a unique exam situation. The exam situation is going to be the same as it was in the midterm. We give you 24 hours. We've already announced.
This is nothing new. It's going to be three sections — a Quest, a midterm, and a final.
Roughly a 1, 3, 2 kind of a model in terms of the number of half hour segments, where the first question will be about half an hour-ish, the midterm probably 90 minutes-ish — we definitely leave room for flex — and the final section for about an hour-ish, totaling up to three hours.
And you can take it any time over 24 hours. And the same recording rules — I'm seeing a question. Same recording rules is there. Nothing changed.
Do everything the same. But only three sections you have to worry about, and not two.
BORA NIKOLIC
And we are going to have — we are here to help.
DAN GARCIA
And points definitely — points distributed similar to the time limits. That's exactly right, yes.
BORA NIKOLIC
Yeah, we are here to help during the RRR week. There will be extra office hours. We will hold them. There will be a dense offering of office hours.
So come armed with your questions. I'll definitely hold an office hour. I can even try to hold it in person, but it's limited to 12 people.
I don't know how is that going to work. But I can try.
DAN GARCIA
I may be with my nose to the grindstone working with CeCe on the team to get the PrairieLearn. So I may not offer that many office hours.
But I definitely do know that we care about you.
And we are putting our hours into where they need to be spent to really make sure we give it a very nice beta-tested exam for you guys, definitely.
And you have to complete it in three hours. A question from Josh — yes, you have to complete it in three hours, but not in a three-hour window.
Three hours overall time — like, this plus that plus that is three hours. And how you divide those three sections up is there. Good stuff. Good question.
All right. So we said this before. And by the way, I ought to make this clarification. Somebody got stumbled one time.
I was really embarrassed when they said, oh yeah, I aced the midterms. That means I don't need to take the final, right? The midterm part of the final, I didn't take it.
I just left it blank because I aced it, right? No! No. No, these are harder questions.
You've got to take that whole final. So everyone takes the whole final. Even if you aced Quest or midterm, you've got to take the whole final. Don't make that mistake.
That would be embarrassing. And if you did well in any class — and this is for 10 — this second part of the slide, I talk about every class I teach.
If you did well, please give back. The usual path is academic intern, tutor, or reader, and then TA, and then head TA, and then summer instructor.
The highest pantheon of elevation of a student instructor is the summer instructor. So some of our students have done that this semester. And it's been great to work with them.
So — BORA NIKOLIC: And this is — DAN GARCIA: Go ahead. BORA NIKOLIC: --really, really important. We really rely on you guys.
I mean, if you have done well, if you like this class, join the teaching crew.
DAN GARCIA
Yeah, definitely.
BORA NIKOLIC
This semester, we went with a lot of new TAs because the previous TAs, a set of lots of TAs that were around before, graduated. And guess what?
In May, another round of former TAs are going to graduate. We need new ones.
DAN GARCIA
Right, right.
BORA NIKOLIC
And that's really important. Also, it's a great experience. I mean, one thing is to take a class.
A completely different level of understanding is to actually teach someone else how to do things. And you [INAUDIBLE].
DAN GARCIA
You might think you know something until you try to teach it. [LAUGHS] BORA NIKOLIC: Yep, exactly.
DAN GARCIA
So there's two questions that are coming up. I'll just answer live.
Do you have to start at a 61C AI and then jump to this?
But if you have AI experience — if you have lots of experience for another class, sometimes we can jump into that.
You can be jumped into a paid position of a tutor reader without having to go through 61C AI. If you have no experience, you have to go through at least an AI experience.
But if you do have experience, it's a good question, anonymous attendee, for that one. And exam strictly timed this time.
That means, again — actually, the exam is same timed as it was in the midterm.
It's just that we're not going to write an exam that's much longer than hopefully an hour and a half for the TAs and the three hours.
So basically, it's even the same exam style as the midterm, same idea, but basically a smaller scope. You don't have, I think, whatever, six hours or something for the midterm.
Now it's definitely three max, three max divided into three pieces. All right. I love this. I'll do my again — (IMITATING GODFATHER) the Godfather answers all of life's questions.
You come to us. The reason the Godfather answers all life's questions is a wonderful meta point here. But what we mean by this is you need somebody to vouch for you.
You need somebody to tell — if you're going into research, somebody's got to say, this person is good. You go into a job.
Sometimes on a job you don't need to vouch as much because you can be in an interview. And you self-vouch by showing off that you know the stuff in an interview.
But most of the time, if you go to the next level of academics, you need to have somebody tell you that they believe. I'm from a position of authority.
And this person is a rock star. Somebody's got to tell you that. And so the way you do that is get involved in research.
And that's one of the reasons that Berkeley is the top university in the world, or at least one of them. So — BORA NIKOLIC: And you — DAN GARCIA: Go ahead, Bora.
BORA NIKOLIC: --heard the pretty strong pitch from Dave Patterson.
Go to grad school. I mean, if you can.
DAN GARCIA
Yeah, definitely.
BORA NIKOLIC
If you can, by any means. Don't stop with a bachelor's degree. I mean, you can. Nobody will fault you at that.
But go for it. The thing there is it becomes a lot more personal. Instead of having a thousand people, it's your advisor and a few people.
And it's a big commitment for that advisor to take you on. So somebody has to vouch for you, like Dan and I.
DAN GARCIA
And they know you great. I mean, they know you for years.
I mean, I just love writing letters for people I've worked for for several years, either in GamesCrafters or other things that I've done, because I can say — I can speak — it's not just, well, they've taught for me, but they've done this research.
They've done these things. There's a lot of elements to a person that you can really boast about. This person has dedication, and passion, and blah blah.
And I can just continue to talk about that. If they're a student — in fact, if you ask me for a letter, I won't give you a letter if you're my student.
I don't know you well enough. Only when you go through my teaching, research, or service activities do I get to know you enough to write a letter that means you.
Otherwise, it's a meaningless letter. It's called actually a weak letter. And weak letters hurt your case more than no letter at all. So I always say no to any student.
But I say yes to every former person who worked with me in a teaching, research, or service role as well. All right. So do you want to take this one over, Bora?
BORA NIKOLIC
Yeah, of course. I mean, I'll paraphrase this a bit and butcher it, possibly. Dave would say this always better than I would.
There are many opportunities in this field. So we'll talk a little bit more about that.
For huge majority of students in the class, this was the entry into the systems area of computer science or what's commonly known as computer engineering. How do we make systems work?
There is a bunch of upper division classes that will help you understand that. And there are many career opportunities that are out there.
Dave Patterson and John Hennessey won the Turing Award three years ago.
And they made a point that this is a golden time, golden opportunity, golden era of computer architecture.
These golden eras happen from time to time when we have a big transition to a new paradigm. And we are now transitioning to this domain of specialization.
Somebody has to do it. We'll have 1,200 candidates in this class that can participate in that. But it probably needs many more people than that.
There are two directions that Dave mentions. One thing, there is a huge gap now between what we call productivity programming and performance programming.
So far, before this class, you have learned about productivity. But in this class, you have encountered what performance means.
And by the way, a side benefit of that, whenever you see a job listing or internship listing for performance programming, you're most likely already qualified for that.
I mean, you'll get more qualified as you take undergrad — more upper div classes. But you're already good enough to hit those internships.
One of the big things is that this gap between productivity and performance is growing every year. And it's the gap between high level applications, frameworks — even Python down to C is huge.
It's 1,000x in [INAUDIBLE]. And we have seen that. One of the big challenges is make productivity frameworks run as fast as performance.
I think that Dave has been quoted, whoever does that will win a Turing Award. It's not an easy thing to do.
And then the other thing is about this domain specialization and domain-specific applications.
Building hardware, building efficient compilers for that and efficient frameworks, it is something that is pretty hard. And we have seen our glimpse of that.
You have seen a little bit how that is done. The first successful coprocessor was a floating point processor. That's the first accelerator. Then we are seeing vectors.
And your vector accelerators, you're using them in the class now. And I'm hoping you're having fun.
But there, in project 4, you're actually using accelerator using relatively low-level programming. These high-level frameworks do help.
But they don't always find the right thing. And they don't compile always quite the right way. There is a big challenge of doing that.
And what we are going to see more and more of these classes of accelerators showing up. And frameworks that are going to try to hit them are going to be really important.
DAN GARCIA
It's tough. Nice summary. Dave would be proud. That was great.
Do you want to talk about this one too? Yeah.
BORA NIKOLIC
Yeah, we are seeing a glimpse of this showing up. We are seeing these systems on a chip that have not just — CPUs are not going anywhere.
Whoever is telling you that the computing is going to completely change and we are not going to see this general purpose course, I would love them — ask them to show you how are we going to program bare metal, that kind of stuff, and never live with — never use an operating system again.
Good luck with that. Good luck with user acceptance of that kind of a thing.
But we are going to see these systems that have incredible performance when they are ran properly, when there are going to be multiple processor cores, GPUs, accelerators, interfaces, memory systems of different kinds.
And we use the picture here of Apple A13 Bionic. But M1 chip is an example of that. And we are going to see more of those.
And by the way, one funny thing about this class — we use this Anandtech often as a source of our pictures and so on.
And we brought you Anand live to — DAN GARCIA: No other class that I've ever seen. No other class in the history of Berkeley — BORA NIKOLIC: Yeah, never happened before.
DAN GARCIA: --as far as I know. It's true.
BORA NIKOLIC
Yeah, and he's not a 14-year-old kid anymore. [LAUGHTER] He's hiring. And I think he's the one that had a few of our undergrads.
DAN GARCIA
That's great. That's great.
BORA NIKOLIC
Things are going to go in this. And what we are seeing now — I mean, you have been following these news. Amazon has now their own systems on a chip.
They have been announced to be in the cloud. Apparently, they have the client one. So Alexa is going to start speaking, is going to get custom hardware.
(IN A STAGE WHISPER) And there are rumors that it may be this time. We don't know.
DAN GARCIA
Woo. Woo. That would be nice coup. That'd be a great coup.
BORA NIKOLIC
And this is going to be all over the place. And they're all going to be different. Somebody has to figure out how to work with all of that.
DAN GARCIA
Outstanding. Outstanding. Good stuff. So a little taste of getting involved with some of the projects I'm working on — as you know, I'm passionate about CS for all.
I hope you know that. I care about bringing computer science to people who are really just passionate about learning, learning CS who may not have had the resources to get there.
So I do have some projects to improve CS10 and to improve Snap! the language. And if you're interested, I have a project manager of BJC, who is Lauren Mock.
She's our former head TA. Just shoot her an email saying I want to join one of Dan's groups. What do you have? And then she'll cue up your names.
Then we'll have a big meeting at the beginning of next semester and do that. I also have had this project I've been working on since I was — gosh, I was a sophomore.
I was a sophomore in college. And I created this game theory tool called GAMESMAN. And I've continued to keep it rolling. And it was my master's project.
And I have now a group that I've been running for now 20 years. Started it in 2001. So next year is going to be the 20th anniversary of GamesCrafters.
So please do that if you care about game theory.
We call it computational game theory, using computer horsepower, i.e. parallelism — nudge, nudge, wink, wink, say no more, say no more — to try to attack game theory trees.
Bora.
BORA NIKOLIC
And we do also have to do development first. I mean, 61C is — it's never twice the same class that is being taught.
I would love to change some of the infrastructure that we run.
As the RISC-V harder becomes more available, we may go for changing our compute infrastructure to be RISC-V — DAN GARCIA: Oh, that'd be so good. That'd be so good.
BORA NIKOLIC
Yeah. Right, you can imagine building your own core. And then fast forward, you're running software on that core — DAN GARCIA: Oh, wow. Wow.
BORA NIKOLIC: --and the accelerators and so on. It'll be a few years. But we will need a big effort to put that together in the meantime. I did mention this before.
But now this is an open call. We do have a bunch of research projects. And we do involve undergrads. And why am I — and the next three slides are pretty important.
My graduate students and senior undergrads are telling me nobody is knocking on their lab doors. Well, perhaps there is nobody in Soda, and the lab doors are closed, and there is nobody inside.
It's a lot harder to make a connection. I'm trying to make a connection. And this is open call.
If you think you're doing really well in this class and then you have it, you can be involved in research. Please join us. And here is stuff that we do.
My group is a little different than the others. But we build a ton of chips, a lot of chips. And guess what?
Pretty much every single one of them has a RISC-V — or every single one in this picture has a RISC-V home core in there.
DAN GARCIA
Wow.
BORA NIKOLIC
Look where they date, to May 20, and that was a Raven-1 chip. Had the first RISC-V core.
Remember, RISC-V ISA was formed by — the first draft was in August of 2010. Six months later, we taped out the chip based on that.
DAN GARCIA
Wow.
BORA NIKOLIC
The chip didn't work. Its internal codename was a train wreck because it was just — DAN GARCIA: [LAUGHS] Oh, that's brilliant. You're learning.
BORA NIKOLIC
But that is what the reason — that's a beauty. I mean, it basically was like a 61C project. I mean, it was built — DAN GARCIA: That's great.
BORA NIKOLIC: --pretty much with an understanding that what you guys have now. But we get better over time.
So the thing that you have in the corner over there, EAGLEX, is a 20-core RISC-V pretty capable — DAN GARCIA: Wow. Wow.
BORA NIKOLIC: --system that booted Linux and has all kinds of other things on it. And you can play with it.
And we need — some of our current TAs are helping us out, write the applications for it. But you can run — write applications through its lab applications for that particular processor.
Next slide.
DAN GARCIA
That's great. Boop.
BORA NIKOLIC
A few big things — I'm associate affiliated with a few research labs. And we do traditionally engage quite a few undergrads.
At that lab, has this big infrastructural project that is called Chipyard. We have a whole bunch of — a good collection of RISC-V cores and tools that go along that.
And there is an automated software framework that lets you build your own chips. You can try those chips. You can build them on an FPGA. Or you can build them in an ASIC.
We typically have undergrads. And several of your TAs have been working on that as well. Join us, please. Next slide.
Here is the — what is in there, there is a generator now. We make you write these cores from scratch when we start, when we build this Chipyard thing.
We give you cores. We have pre-configured cores written in a more abstract way.
But you can make a single core RISC-V. Or you can make a few cores RISC-V. And then we have custom accelerators.
And we are practicing that future that we are imagining, where the domain specialization is going to be the thing. So we make all kinds of things.
We don't always know what is that going to be. Is it going to work well? And there is plenty of surprises over there. Oh, we thought this was going to work.
Not really well. But that's what the research is. So try it out. And it is going to help you see the future a little bit earlier.
And that's it. Back to you, Dan.
DAN GARCIA
Beautiful. Sure. Let's talk about opportunities next semester. So if you're thinking about what courses — I think that was one question.
What courses should we take after 61C? Here's a little taste of it. 152 is an obvious course if you like CPU design, memory, and architecture in general. Please think about that.
162 is the OS side of that. So if you think — if you appreciate the VMOS and just want it on more, and more, and more.
And I think 162 will really teach you to be a better programmer as well, even if you don't end up studying systems or OS. You'll be a better C programmer.
BORA NIKOLIC
It's an intense class.
DAN GARCIA
Yeah, it's really hard. But it's great.
BORA NIKOLIC
These two classes really build up on top of what we have done. I mean, what we have done follows.
DAN GARCIA
They build your muscles in that way. I offer three DeCals. I'll just give you a little taste of them. One, I mentioned GamesCrafters before.
Please do join that. This time has changed. It's not Monday, Wednesday — this old slide, 10:00 to 11:00. I think it's 4:00 to 5:00.
Check my website. I think it's 4:00 to 5:00 PM next semester. So sorry about that.
There is an iOS DeCal, if you want to learn how to program iOS devices, and UCBUGG DeCal, if you want to learn how to make Pixar short films.
That's an amazing course. That's also a lot of work. That's three units. You don't need anything other than 3D interest to do that.
So that's just great. You want to cover this one, Bora?
BORA NIKOLIC
More on the EE side, many of you probably have taken 16A. And that's good enough to get you into a lot of entry into the EE side of things.
If you want to get into 120 series, which are signals, and systems, and communications, you probably should take 16B to get going with that.
But if you are interested more into circuits and physical devices, you can go straight from 16A into 105 and then dive into 140 series. There is also devices that is 130 series.
The next one I'm teaching — I'm guessing I'm going to see many of you in 151. So we just got our teaching assignments. I'm teaching EECS151, which is Intro to Digital Systems and Circuits.
And that class just basically follows exactly where we stopped. We are going to repeat some of that — there'll be other students who might have not seen RISC-V in that class.
We are going to repeat RISC-V and then just get going, learn a new language, Verilog, and build a bigger RISC-V system and get it to work.
It is one thing — I often get questions. It has two labs. And they're separate. You can sign up for LA and LB.
Generally, not a good idea to sign up for both. That's seven units, and seven Berkeley units is a lot of units.
I mean, there are students who do that, but do think about it before signing up for seven units of heavy lab project class.
Usually, people take five units, so the lectures plus one of the labs. And then in the next semester, they take the other lab.
DAN GARCIA
Wonderful. I just looked up on the side. GamesCrafters is 3:00 to 4:00 PM next spring. So think about that if you want to do that.
This is my slide I created a couple of years ago. And I still think it's true.
With all the whiz bang, oh, there's cloning, and TVs on your wristwatch, and all the thing, I think the biggest invention that we're going to see in our lifetimes is going to be the ability to directly connect your brain to a machine.
And I think we've already seen Elon Musk — I think one of our Computing in the News, if you remember that, Bora, is we brought — Elon Musk is trying to do that and play with that exploit.
That is a very exciting space.
BORA NIKOLIC
A bunch of our alumni. And I think today posted they're hiring again.
DAN GARCIA
Wow. Wow. I mean, this is, I think, the hottest, most interesting space in terms of future, like blue-sky research.
Boy, the Matrix is coming, folks, just the ability to be able to have input and output, to be able to download a dream and upload a dream.
Blows my mind in terms of science fiction. But I think it's going to be here soon. So very excited about that. And may be able to download all your thoughts of life.
You have some Albert Einstein level intelligence. Download it so you have it kind of a thing, surviving it. But wonderful stuff.
So Jose Carmena is one of our colleagues who was working that very, very, very hot research area — brain-machine interfaces. So please connect with him if that's interesting to you.
Bora, project 3 awards. Ladies and gentlemen, the awards!
BORA NIKOLIC
We finally — we got our first 10 winners. Congratulations.
DAN GARCIA
Woo! Congratulations, everybody, for your hard work in project 3. Man, oh, man, that was a lot of work to put that together.
BORA NIKOLIC
So what you won is our — each one of you won a pair of Apple AirPods Pro.
DAN GARCIA
Wow.
BORA NIKOLIC
We will need your shipping addresses. We are going to contact you to — if you would like to receive the award. We can give it to you in person.
I mean, maybe. We might be able to arrange for that. But more likely, we are going to ship it to you. And we'll need your address.
DAN GARCIA
Congratulations.
BORA NIKOLIC
Congratulations.
DAN GARCIA
Congratulations to all that group. Wonderful, wonderful.
BORA NIKOLIC
This is spectacular. I mean, sometimes we have small prizes, $30 prizes. These are a — DAN GARCIA: This is a big deal.
BORA NIKOLIC
We have a couple of dozen of these.
DAN GARCIA
Again, we've never done this before. Usually, we give you a handshake and a certificate. We've never given a hundred dollar gift for this. Congratulations.
BORA NIKOLIC
Well, this is $259, $259 each.
DAN GARCIA
Wow. Wow. Thank you, Apple. And thank you to the students who put their hard work in, folks.
Great job, everybody. Great job, folks.
BORA NIKOLIC
Yep.
DAN GARCIA
And there are more awards coming I understand, Bora.
BORA NIKOLIC
Yeah. So we have more awards for project 4, top performance in project 4, and also top performance in the class. And our top TAs are going to get awards.
DAN GARCIA
Wow. Wow. Great job, folks. Again, wonderful.
BORA NIKOLIC
Speaking of — DAN GARCIA: All right. Speaking of TAs — BORA NIKOLIC: Speaking of TAs — DAN GARCIA: — let's give thanks to all the hard work done by our TAs.
This is all a set of just thank yous at this point. This is like the end of an Oscar speech. First, we want to thank Stephan and CeCe.
They were outstanding head TAs up and down, handling all the situations we did. Thank you so much for your hard work and dedication to this class. And this is just pages after pages.
We have a huge staff. So thank you Jonathan, Nikhil, Sunay, Daniel, Cynthia, and Kevin, again, for all of this.
Caroline, Jie, Kimberly, Max, Anson, Taejong, Jerry, Alice, Justin, Seah, Ryan, Emily, Ryan, Ivy, Robin, Luke, Avi, and Lu — almost done — Ryan, Dayeol, and Anthony, thank you so much.
Just give a hand for our TAs and our staff! Amazing group of TAs who were there. By the way, our TAs are overworked at a level — I don't even want to know.
I mean, I want to know because I want to fix it. But our TAs put in so many extra hours that you guys don't even realize.
So please — thank you so — let's give our thanks to the TAs. They really — BORA NIKOLIC: It's like Oscar ceremony — we need to thank our families, friends, producers.
DAN GARCIA
Yeah, yeah. [LAUGHTER] That's right. Let me thank my family for letting me steal every weekend for the entire semester. That's definitely where the — prize goes to my wife.
BORA NIKOLIC
Exactly, and staying up till 2:00, 3:00 AM.
DAN GARCIA
The prize goes to my wife for not having a weekend with her for the whole — all right.
I also want to thank — there's another slide that I think might have gotten lost here. I want to thank also the tutors and the AIs for our family.
This is a lot of work by a lot of people. This is where the wheels meet the road, all that contact, all that face-to-face time you had with your tutors and your AIs.
Thanks to those folks. Great job, everybody. Shoo, here we go. The future for future Cal alumni — what is the future?
Look at this. This new millennium is amazing. Ubiquitous computing — anywhere you go, you've got a computer on your phone, in your hand as fast as a computer from two generations ago.
What can you do with it? Nanotechnology — making things smaller and smaller and small. The internet of things means that everything is an IP address now.
I have light bulbs in my house that are IP addresses, Bora. This is crazy. If you told me, Dan, you're going to have light bulbs, I was like, come on.
A light bulb lamp, yes, I have — BORA NIKOLIC: Dan, they're not secure. They're not secure.
DAN GARCIA
I know they're not. Wait — why did it go red? Thank you so much, hacker. So I've got this going on.
We've got AI. The parallel revolution is hitting. Domain specialization, as Bora mentioned, exactly that. Again, rapid changes in technology.
Post-PC era — we talked about the eras of computing. Remember the lecture.
I talked about that, where some of the post PC-era, the power of the cloud is in your hand and on your watch. It's incredible. You've got the world's best education at Cal.
Never give up. "The best way to predict the future — is to invent it!" Alan Kay. So the future is up to you.
That's it. Folks, I want to thank Bora for the teamwork. Bora, I really have enjoyed this year with you, my friend, this semester with you.
It's been a challenge because of all the other things that came out. But I couldn't imagine somebody better to ride the train with you.
So thank you so much for all the things you've done.
BORA NIKOLIC
Dan, and thank you as well and to your family for letting us borrow you.
DAN GARCIA
[LAUGHS] Yeah. Yours as well.
BORA NIKOLIC
Lending us to you.
DAN GARCIA
And yours as well. Oh my goodness. Let me get out of this. Get out here.
Don't leave yet. Don't leave. We've got one more — let me see. I might have — Bora, do you have one more thing?
I might have one more — BORA NIKOLIC: Yes, one more thing.
DAN GARCIA
I might just have one more thing. I don't know if many of you folks are fans — BORA NIKOLIC: We've never done this before.
DAN GARCIA
Yeah, we've never done this before. This is brand new. I don't know if we have fans of the Hamilton soundtrack, Disney+ and everything.
But I might need some help with this from folks who might need to — so TAs, if you can unmute, I might need — all the TAs — Anson, Ryan, Lu, Bora, I need some help here.
BORA NIKOLIC
Dan, I'm here. I'm here.
DAN GARCIA
But you have to go to — BORA NIKOLIC: I'll give it a try. I'll give it a try.
DAN GARCIA
So I need everybody to give me a little bit of a backup here going on. So we'll see — shoo. We'll see if this works. And there is some Zoom lag.
So I want to apologize already for — BORA NIKOLIC: We have never done this.
DAN GARCIA
The Zoom lag is going to be just very unfortunate. And this is much better live in front of the class. But we'll try. So folks — BORA NIKOLIC: Let's give it a try.
DAN GARCIA: --give this a try. TAs, Ryan, Anson, Lu, you've got to unmute. You got to say with us. They don't know.
They're like, what do you mean? You didn't tell us this. I didn't tell anybody. Nobody knows what they're supposed to be doing.
But we're going to give this a shot anyway. Bora, at least you've got my back here. Here we go. All right?
BORA NIKOLIC
Exactly.
DAN GARCIA
Ladies and gentlemen — BORA NIKOLIC: Let's do it.
DAN GARCIA
All right, on 3. Let me turn the volume down just a little bit so I'm not overwhelming myself. Okay. On 3.
Here we go. Woo. [MUSIC - LIN MANUEL MIRANDA, "TEN DUEL COMMANDMENTS"] (RAPPING) 3, 4, 5, 6 — BORA NIKOLIC: (RAPPING) 1, 2, 3, 4, 5, 6, 61C. It's the six big ideas.
Computer architecture.
DAN GARCIA
Number 1 — BORA NIKOLIC: Number 1 — DAN GARCIA: --it is all about the layers of abstraction. If you understand, no need for further action.
BORA NIKOLIC
Number 2 — DAN GARCIA: If you don't, grab a friend, peer instruction. They'll tell you that's Moore's law was an exponential function.
BORA NIKOLIC
Number 3 — DAN GARCIA: Have a second cache just in case the first one is the miss.
Locality of time and space, this hierarchy, with smaller costly faster, always a local copy of the bigger, cheaper master.
BORA NIKOLIC
Number 4 — DAN GARCIA: If not fast enough, parallel! Though pruning nondeterminism can be hell.
You widen up your registers, accelerate with SIMD, or you go multi-core and then you speed it up with MIMD.
BORA NIKOLIC
Number 5 — DAN GARCIA: Measure performance for improvement. The use of solid metrics, yo, is a movement.
To get — BORA NIKOLIC: Number 6 — DAN GARCIA: --dependability, add one, two, or three backups for redundancy.
1, 2, 3, 4, 5, 6 — BORA NIKOLIC: 2, 3, 4, 5 6 — DAN GARCIA: 61C!
BORA NIKOLIC
61C!
DAN GARCIA
Good luck on your exam! Woo! Folks, what a pleasure it's been teaching this class with you.
We are so excited for you to go rock star that final exam, crush that, get the highest A grade in the history of this class, no curve. You guys are going to be great.
You guys are ready. We love you guys. All right, good stuff. Good stuff.
Thank you again for coming to this last class, everybody.
BORA NIKOLIC
And we've got to vacate this. We have a class after us.
DAN GARCIA
Yeah, that's true. We've got a class right now, actually, Chem 1A. So we've got to get out. Take care, folks.
Thanks, everybody.
BORA NIKOLIC
All right. We'll answer the questions in — hit me up with the questions next week.
DAN GARCIA
Perfect. All right.
BORA NIKOLIC
All right, bye.
DAN GARCIA
Take care, folks. I'm writing the exam. All right. Thanks, folks.
Bye.
BORA NIKOLIC
Bye.
DAN GARCIA
Good stuff.
Đăng nhập để đánh giá học liệu này.
Đăng nhập để bình luận — chưa ai bình luận về học liệu này.
Không có khóa học liên quan.