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.
PROFESSOR DAN GARCIA
Yeah. All right. Ladies and gentlemen, welcome to week 15, last Monday live session, 61C. Woo!
Great to see you all here. Delightful, delightful. Do know this — I'm going to say it early and often, like voting in Chicago — or vote early and often.
Make sure you come on Wednesday for a live session, for James, our awesome guest who's here is going to give a live session on GPUs. Come here, same Bat-time, same Bat-channel.
And also, make sure you come on Friday for our last live lecture. So this is three live, live, live.
It's as if we weren't even in a virtual world, and we're just having live sessions, Monday, Wednesday, Friday, this same time, okay? 4 to 5, come to same channel, same link.
We'll see you there. Perfect. And by the way, Bora, we need to make the actual webinar. Let's not forget to do that.
PROFESSOR BORA NIKOLIC
Yeah, I am on it.
PROFESSOR DAN GARCIA
We may have forgotten to do that, so let's make sure we do that.
PROFESSOR BORA NIKOLIC
I'm on it, Dan.
PROFESSOR DAN GARCIA
All right.
PROFESSOR BORA NIKOLIC
Yeah, and also the award, preliminary award ceremony on Friday.
PROFESSOR DAN GARCIA
Yes, and the awards happen. Remember, the awards are going to happen on Friday, so we're excited about that. All right. Today's agenda — we always do the agenda here.
Welcome to week 15. Welcome to week 14. Cross that out. Wait, can you fix that in post, Bob?
Yeah, sure, I can. Week 15, our guest is James Percy from Apple Computer. We're going to have a couple of Computing in the News pieces. We'll talk about important final exam information.
We're going to drop a mic on some of that stuff.
And then I'm going to do a review of the schedule, and then we'll have an ask us anything, and we'll have some time there. But let's jump into our wonderful guest, James.
So our guest is James Percy from Apple Computer. I'm going to now add you to the spotlight so you're on stage. Welcome, James. Great to see you here.
James has been involved in computer engineering in many, many roles, I think, since your undergrad at Princeton. I'm thinking back to the time. He was an Engineering Manager, you can see here, at ATI.
Then he went to AMD as the Director of Engineering, and then joined Apple since 2013, has been the GPU leader.
So he's been one of the key, key — as you get up in the top of the pyramid, the group meetings become smaller and smaller.
And James is in smaller and smaller group meetings with more and more important people. I'm sure you're meeting with Tim Cook every other week or something like that.
It's crazy. But anyway, it's a delight. James, it's been great to have you. We've been working together this last semester, thinking about this.
And he also has been a guest at 61C a year ago. So thank you, again, for returning to 61C.
The purpose of this is just to meet you, and get to know you, and talk about your path and advice you have for students who are thinking about similar careers, or thinking about your time back into the school.
So first of all, I always start with, how did you get to where you are? I'll just throw a couple questions out there.
Did you always know you wanted to be what you are now? And if not, how did you get to where you are now? So tell us your path.
Tell us your career path.
JAMES PERCY
Sure. I'll try to go through that [INAUDIBLE]..
PROFESSOR DAN GARCIA
And you can go as far back as you want to. In second grade, Mrs. Hawkenhauser — PROFESSOR BORA NIKOLIC: Kindergarten is fine.
PROFESSOR DAN GARCIA: --always told me I'd be a great — no, so go ahead. Anything you want to do.
JAMES PERCY
So going back, way back, I was always interested in computers, going back as far — I forget how old I was, probably fourth or fifth grade, which would have been in the '80s, believe it or not, interested in computers since I was a kid.
We used to play computer games. Back then, you'd have to run stuff on DOS. And you'd have to set it up to manage your memory and stuff like that.
And so that got me very, very interested in how do computers work, and how do operating systems work?
And back in those days you really had to hack your way through just to get games to run. And so that was kind of like my early interest in computers.
And that kind of turned into me becoming a Computer Science major in college at Princeton. And I went into Princeton knowing I wanted to be in engineering.
I liked math a lot, liked physics a lot. I discovered that real math, meaning studying math in the university is really hard, like, really, really hard.
And despite being very good at it, I was not that good at it.
PROFESSOR DAN GARCIA
That's my story. That's my story, too. I was like a superstar in high school, and then, wait — this math is a lot harder than it was in high school.
JAMES PERCY
Yeah. Vertical math is hard. But I did take an Introduction to Computer Science class the freshman year of college.
And I'm sure these days, people take that stuff in high school, or in middle school, or earlier.
PROFESSOR DAN GARCIA
Sure, but sometimes not, sometimes not. There are people who don't get it, and we're here for them. We teach a class.
My BJC class is exactly that space to introduce to people in college if that's their first time. We let people into the [INAUDIBLE]..
JAMES PERCY
And so I was learning how to code in C freshman year of college. And I remember half the people in that class had been doing it for years.
And so I felt woefully behind compared to at least half the class. But I worked really, really hard at it, many late nights in the lab. And I realized it was really fun.
I really enjoyed it, and I was pretty good at it.
Once I kind of got my head around the basic constructs of how does the C program work, and what is a pointer, and stuff like that, and what's a function call, I really enjoyed it.
PROFESSOR DAN GARCIA
You feel powerful. Something about C, something about being low level, like I could do anything. I can move bits around in any way I want to, right? Different than Python, you're 10 miles up.
Somehow below it, magical, right? But being low level, we love that. Yeah.
JAMES PERCY
And so that turned into, okay, well, this is cool. How does C work? And how does assembly work? And then — PROFESSOR DAN GARCIA: Oh, I love it.
JAMES PERCY: --how do you boot an operating system? And how does networking work, and how do graphics pixels end up on screen?
So I literally just took every class that I could reasonably take about computer science.
And I took operating systems, and networking systems, and computer graphics, and compilers, and learned about how assembly works, and risk architectures, and stuff like that. And really just absorbed as much as I could.
And I discovered going through that, kind of sophomore and junior year, that I was really interested in computer graphics.
And so that's kind of what I specialized in, in terms of, at Princeton, even if you're an engineer, you kind of do a senior project, or two senior projects.
And so that's what I focused my senior projects on. And I learned about ray tracing. I did some advanced ray tracing algorithms.
And the cool thing I liked about it was — PROFESSOR DAN GARCIA: But still from the software point. Still from the software point of view.
JAMES PERCY
Still software.
PROFESSOR DAN GARCIA
We have CS184, our graphics classes, but it's always software. We're never getting into the hardware behind it, ever.
JAMES PERCY
It was all software programming, and OpenCL, and application level.
But what I really liked about it, and I'll get to the hardware part in a minute — what I really liked about it was you drew pictures at the end.
And the pictures looked really cool. And for me, it was easier for me to think about, hey, here's this model.
And I want to come up with an algorithm, and it generates a picture at the end of the day. It wasn't just bits moving across a network.
PROFESSOR DAN GARCIA
It's art.
JAMES PERCY
It's art.
PROFESSOR DAN GARCIA
In a way, right? It's beautiful.
JAMES PERCY
And that was part of our assignments. We actually had to do assignments using the algorithms that we'd learned, but it had to be artistic.
It had to look cool, or it had to be creative.
PROFESSOR DAN GARCIA
That's beautiful. Did they give you more points for looking cooler? I've never seen that, where they actually give — JAMES PERCY: Yeah, we actually got — PROFESSOR DAN GARCIA: Really?
More points for — wow.
JAMES PERCY
There was a contest at each lab assignment for the best looking image, or picture, or whatever it was. So there was bonus points for being creative. And so I really liked that, too.
And so my advisor at the time was a guy named Thomas Funkhouser. I think he [INAUDIBLE].
PROFESSOR DAN GARCIA
Ah, Tom. That's my buddy.
JAMES PERCY
Yeah, you know Tom?
PROFESSOR DAN GARCIA
We were grad students together. That's my buddy.
JAMES PERCY
Are you serious? Oh, wow.
PROFESSOR DAN GARCIA
Seriously.
JAMES PERCY
Small world.
PROFESSOR DAN GARCIA
You're a young guy, but we were sitting together. He was awesome. Are you kidding me?
JAMES PERCY
Well, I probably wouldn't be here if it wasn't for Professor Funkhouser.
PROFESSOR DAN GARCIA
Look at the small world. Look at this world. That's amazing.
JAMES PERCY
Because he hooked me up with a guy named Mark Siegel, who knew a guy named Bob Drebin, who was working at ATI at the time. And said, hey, I know this guy James.
He's a CS guy. He really likes graphics. Hey, you should check it out.
And the reason I tell you guys that is, it's, never underestimate the value of the contacts that you're going to make at university.
Because it could be Dan knows a guy, who knows a guy about a job that you might be really excited in. And that was basically what got me in into ATI.
And I basically got a job doing verification of graphics hardware, having really no idea what that meant.
All I knew was that I wanted to work on graphics somehow, because I was really interested in graphics.
ATI, at the time, was one of the few companies left making what we call discrete graphics cards, meaning GPUs, or chips that are just dedicated to accelerating graphics algorithms.
And I was like, hey, that sounds pretty cool. I'm really interested in this. Let's go find out what this is about.
I had taken, basically, the prerequisites for Electrical Engineering at Princeton, but I was not an Electrical Engineering major.
And I remember showing up the first day of my job and they started telling me about tape-outs, and synthesis, and netlists. And I had no idea what any of those things were.
I'd asked Bob Drebin at the time, hey, what is all this stuff? And they were probably going, oh, my god. Why did we hire this person? Sorry, this is the long version.
I'll try to get to the point.
PROFESSOR DAN GARCIA
No, that's fine. No, go, go, go. It's great.
PROFESSOR BORA NIKOLIC
That's good.
JAMES PERCY
Go ahead, Bora, if you have a question.
PROFESSOR BORA NIKOLIC
Yeah, I think there are a lot of people here who can relate to that in very different ways. Keep going.
JAMES PERCY
And so I was basically a CS software guy in the hardware organization at a hardware company.
And what I brought to the table at that time was that I knew how to code in C and C++, and I knew graphics algorithms.
And so what I started out doing was I actually worked on writing low-level C tests for different blocks inside the hardware. And also, working on some top-level OpenGL-like tests.
And to be honest, it isn't probably truly what we consider verification in the honest sense, 20 years later.
But it was low enough level that I could read the hardware specs and I could understand, okay, this is how this interface to this block works. This is what the inputs and restrictions are.
We compare it to a C model, which I also worked on.
And that was how we pipe vectors through, and when the vector is matched on the output, then you're good, and your block's done.
That's not necessarily how DV gets verification and hardware verification gets done today. But for me, it opened up — PROFESSOR DAN GARCIA: It's a higher level. It's a high level of verification.
It isn't low level verification, it's upper level verification, right?
JAMES PERCY
It's higher level, and it's something somebody with a C background, and who understood how some of the math or graphics algorithms work could do. And so that's how I got interested in hardware.
And so I started my first couple of years at ATI, to be honest, were some of my funnest of my career. Because in those two years, I worked on production OpenGL drivers.
I worked on emulation platforms. I worked on diagnostics in the lab. I worked on the C model. I worked on verification.
I worked with some guys at Stanford doing a ray tracing demo on some new hardware that we had that hadn't even shipped yet. So I was doing all sorts of different stuff.
And so I was really just like a sponge learning about all these different things.
And eventually, I kind of decided, hey, this is pretty cool but I think if I'm going to stay in this area I really need to understand hardware more.
And so I became a manager of the verification team. And at the same time, I actually started taking a couple of classes at Stanford on introduction to logic design.
And some of that was kind of a refresh from what I had in undergrad but hadn't looked at in a few years, and some of it was getting a deeper understanding of what it takes to productize hardware, which was far outside my comfort zone.
PROFESSOR BORA NIKOLIC
What was it, 272, 273, or something like that?
JAMES PERCY
I think it was 272. It's been over 10 years, but by name, I think that's what it was. And so I wasn't pursuing a degree or anything.
This was purely just so I could kind of get a better understanding of how hardware worked. And I kind of parlayed that into starting to manage part of the design team.
I started managing GPU projects at ATI, and kind of learning — that was just trial by fire. You learn about, what does it take to actually build a chip?
What are all the things that you have to worry about, not just like, hey, do I have my algorithm correct, but can I actually place and route it?
Can I close timing on it?
PROFESSOR DAN GARCIA
Can I cool it?
JAMES PERCY
Can I cool it? What are the power delivery problems? What technology node, and what transistor mix am I going to use?
I mean, you can go on and on about the amount of things I learned in those couple of years managing projects, and learning from people who had been doing this for years.
And then, AMD actually acquired ATI, I think, back in 2006, I want to say, 2006 or 2007. And I kind of was doing the same thing managing a larger team.
And then I also started working on what we called APUs at the time, meaning they were one of the first types of SOCs to combine CPUs and GPUs on the same chip.
And so I was kind of responsible for the GPU part of that chip.
PROFESSOR DAN GARCIA
What's the A stand for in that? Is it like all inclusive, or something?
JAMES PERCY
What did it stand for?
PROFESSOR BORA NIKOLIC
Application?
PROFESSOR DAN GARCIA
Application?
JAMES PERCY
Application processing unit. Yeah, I think that's right. I think that's what it was. And now, of course, fast-forward 10 or 12 years later, everybody is doing these chips.
Apple has been doing those chips, and we see there's huge energy efficiencies from putting GPU and CPUs on the same die along with a whole bunch of other stuff, that it's like everybody's doing it.
But at the time, it was pretty novel, and pretty exciting, and very hard.
And so I stayed at AMD up until about 2013. AMD was having a rough time, unfortunately, at that period. Things are much better now.
And I came to Apple because Apple was starting to build, or wanted to build up GPU capabilities in-house.
And so at that time, Apple had already been shipping iPhones with their own SOCs, their own CPUs. But they were expanding into GPUs and other areas as well.
And so I said, hey, this feels like a really good opportunity and I'm going to give it a shot.
There's a bunch of people that were already at Apple that I already knew, so I was going to a place that was somewhat comfortable.
And I helped build up the Cupertino GPU team from about four or five people at the time to 70 or so people today.
PROFESSOR DAN GARCIA
Whoa, wow. That's a big team.
JAMES PERCY
And actually, now I've actually just taken over responsibility for all of the, what we call GPU IP development at Apple.
So what that means is, frontend, RTL, microarchitecture, and verification is now my responsibility. So now I'm managing a 200-person team across multiple sites.
PROFESSOR DAN GARCIA
Holy moly. Actually, is there engineering that's not in Cupertino?
JAMES PERCY
Probably, yeah. There's lots of engineering outside of Cupertino.
PROFESSOR DAN GARCIA
Oh, okay.
JAMES PERCY
My boss is actually based in Austin, and so there's a team in Austin that I'm responsible for. There's also GPU physical design. There's architecture validation that I'm not responsible for, that's based in Austin.
We have a site in Orlando. We have a site in the UK. We have, not GPU related, but there's a site in Israel, in Haifa and Herzliya. So we are all over the place.
PROFESSOR BORA NIKOLIC
And we didn't know about the GPU until like three or four years ago, right?
JAMES PERCY
Yeah.
PROFESSOR BORA NIKOLIC
It did not exist.
JAMES PERCY
So we went public, or the first time we said we shipped our own Apple-designed GPU was with the iPhone X in 2017.
And so my team had actually been working on pieces of the GPU prior to that. And we shipped as well, like we shipped the new texture unit.
I forget which iPhone it was, but we'd been shipping pieces of it for quite some time. But we kind of did our first from scratch design with the iPhone X.
And as you guys, if you follow the news — not only are we doing iPhones and iPads, we're now shipping them in the Macs now as well.
So my daughter, my wife has one upstairs. But I have one and it's absolutely amazing the experience of a Mac with Apple silicon.
So if you don't have one, I encourage you to get one.
PROFESSOR DAN GARCIA
And it's still holiday season. That's outstanding. I'm curious, how do you — I'm thinking about a team of 200 people.
Are you in meetings every minute, or do you have subgroups? How are you able to manage such a large group just to — and you're working with people above you.
You're layer, right, so you're talking to people above you, or call you in meetings, and you got to call meetings from people below you.
You can't be meeting every 200 person once a week, so how are you able to manage all of that?
PROFESSOR BORA NIKOLIC
It's pretty hard to get on James' schedule, just so you know.
PROFESSOR DAN GARCIA
Yeah, but the folks don't, students don't realize who we're getting for this semester. He's kind of a big deal. I'm just saying, he's kind of a big deal.
This is exciting. That's great. Thank you for your time.
JAMES PERCY
Yes, I am in meetings almost all day every day, which is a challenge, especially early.
I'm not really a morning person, but a lot of the folks I work with are in the UK and the East Coast.
In terms of managing a team like that — and by the way, this is a pretty recent change.
So I'm still kind of learning how to manage this and how I want to run it.
But you really have to have a very strong staff that you trust and that you believe in, and you can delegate tasks to, and you're confident they're going to get done and come to you when they have problems that they need your help with.
And so Apple's an interesting company, in the sense that, clearly, given its size and the number of things that we're doing, delegation has to be a thing.
But you have to learn how to delegate, but still be in the details of the things that matter.
And so I think people that are mid to senior levels at Apple have become very good at that. Clearly, you can't dive into anything.
There just aren't enough hours in the day and in the week, but you have to be very good about choosing which things you want to dive deeply on and get involved with.
PROFESSOR DAN GARCIA
The important things. The important things, right?
JAMES PERCY
It's a skill I am acquiring.
PROFESSOR DAN GARCIA
It's like a performer, you've got to improve performance. You increase everything where the time is being spent slowest, that's the thing you increase to make the biggest performance.
Here, you spend your hours on things that are really the things that matter the most. You make a decision, what can people just do without your needing to touch?
What do you need to really get in to work on? That's great.
JAMES PERCY
Right.
And you have to train your team or your people to make sure they're bringing the things to you that matter the most and that need your input, and are not bring things to you that they're not important, and don't matter, and are also not — hiding is probably the wrong word, but may not be giving you the visibility of the things that you need you need visibility of.
PROFESSOR DAN GARCIA
Sure. I'm going to ask a weird, COVID-y question. I'm sure you're working from home now, or you actually go to Apple Park? Or how do you manage that?
JAMES PERCY
I am working from home. I have a virtual background because I'm in my guest room because I got kicked out of my home office from my kids.
PROFESSOR DAN GARCIA
[INAUDIBLE] kids, too.
JAMES PERCY
I've been into the office probably five or six times, maybe once a month, I'd say, on average.
There's not a strong need for me to go in unless there's something special that needs to be taken care of.
And generally, unless people want to be in the office or need to be for very specific needs, we're encouraged to stay home.
PROFESSOR DAN GARCIA
Right, that's what I was going to say. Hardware people I think of needing to be in that lab. Whenever we see the videos of, oh, here's the new chip, and I see this lab.
You've got to be in that lab. You can't just do that at home. Yeah, I'm at the kitchen table. I'm trying to make the new GPU.
No, you've got to be in the lab with the sensors.
JAMES PERCY
You'd be amazed actually that even many of the people that work in the labs these days, actually don't need to be there in person.
Clearly we need to have a couple of people in there to set them up. Initially, a lot of the debug could be done remotely these days.
PROFESSOR DAN GARCIA
Get out of here. That is so cool.
JAMES PERCY
It's actually very, very, very impressive. And it's actually, we've had to develop that technology anyway because of the multi-sitedness that we talked about earlier.
We've always had people in Austin debugging chips and systems that are physically located in Cupertino. We don't like shipping pre-prototype material around, or people carrying it around.
PROFESSOR DAN GARCIA
Right, lose them on a bar in San Francisco, remember how many times we saw that?
JAMES PERCY
I remember that. Yeah.
PROFESSOR DAN GARCIA
I remember that, right? Yeah, I just left under the counter. Sorry, it doesn't happen.
JAMES PERCY
So we've had that capability, but we've been leaning on it even more, such that we can do a lot of remote debugging without having people being there in person.
PROFESSOR DAN GARCIA
Wow, it's impressive. That's cool.
PROFESSOR BORA NIKOLIC
Just for everybody, even now research apps that are even clunkier when they come — I mean, they're always clunky. These are not products. We test them mostly remotely.
I mean, somebody goes in and gets it through the smoke test. There is a funny thing about the smoke test, right? When you power it up, will it pop?
PROFESSOR DAN GARCIA
Oh, is that what that's called? I actually didn't know that was what that's called, the smoke test.
PROFESSOR BORA NIKOLIC
Right, it's called the smoke test.
PROFESSOR DAN GARCIA
Put some power to see, does it actually make smoke out of it? And that's bad. Smoke is bad, I'm guessing, right?
PROFESSOR BORA NIKOLIC
Yes, smoke is not good.
PROFESSOR DAN GARCIA
Just so we know. Yeah, yeah. That's beautiful.
PROFESSOR BORA NIKOLIC
So once when you pass the smoke test, and if you manage to boot Linux on it, it's all remote. You get the terminal and you can be anywhere.
With us, it's a bit of a challenge running stuff through campus security. I can just imagine how fun is that with Apple security doing these kind of things.
Well, I want a Docker image. No, you can't have a Docker image, and so on.
PROFESSOR DAN GARCIA
Yeah, I guess that's true. I don't know, was this always the case that you're in such a high security kind of a thing, where your stuff is cutting-edge?
And if that gets out, that's the company, or that's a big hit. Have you always been in that space? Or was it a big deal as you moved up to be in more?
Now it's iris scanners, I mean what's the level of security when you go in, in terms of keeping this IP within the walls of the company?
JAMES PERCY
We are very, very selective and careful about any sort of IP that leaves the walls of Apple pre-production, before it's been released to the world.
And there's special carrier programs that are set up for that, and they're typically only people can carry stuff around once things have been announced but not shipped.
So we did have to make some special provisions, I'll say, for dealing with what that looked like in a COVID, work-from-home world. But we did let people do it.
You guys have probably seen all the stuff we've announced the last three months. That stuff would not have shipped if we did not let people bring it home.
So it was something that had to happen.
PROFESSOR DAN GARCIA
Wow. Right, it makes sense. Yeah, that makes sense. Yeah, you can't ask people to always come back in.
Yeah, that makes sense.
PROFESSOR BORA NIKOLIC
When things are out, things are out.
You can see, I think, EETimes now has on the cover of page, there'll be a series of articles of dissecting M1 chip, and they're going to take every single [INAUDIBLE]..
PROFESSOR DAN GARCIA
Oh, wow. Disassemble it. Yeah. Wow.
PROFESSOR BORA NIKOLIC
They're taken apart Mac mini. No, but this is decapping the chip. And they say [INAUDIBLE].
PROFESSOR DAN GARCIA
How did they do what they did? Yeah, how did they do what they did, right?
JAMES PERCY
Yeah, and we're particularly sensitive about performance benchmarks, and people getting data and results that are not production quality.
So we want to, in addition to just the surprise and delight we want to deliver to our customers, and being delighted by the experience, we also want to make sure that we're only high quality data is getting out there.
PROFESSOR DAN GARCIA
Sure, makes sense.
PROFESSOR BORA NIKOLIC
Yeah, we have a bunch of questions to — if you're interested.
PROFESSOR DAN GARCIA
Yeah, we have some questions. This is great. Thank you for all of this background stuff. Now we'll just do Q&A, if we can.
So this is wonderful.
JAMES PERCY
Sure, absolutely.
PROFESSOR DAN GARCIA
There was just an anonymous question that says, "Do you code or design much anymore, or is it just management?"
JAMES PERCY
No, they don't let me code. They don't want that.
PROFESSOR DAN GARCIA
Oh, no. Just meetings, oh, it's so sad.
JAMES PERCY
Just meetings.
PROFESSOR DAN GARCIA
No, that's hard.
JAMES PERCY
It's actually interesting, back at AMD, up until I left, I at least still had a tree and knew how to build it and run simulations, and run tests.
And coming to Apple, and I came into Apple as a management position, and despite it being a small team, I never had time to really get used to the flow or really learn how things work.
And so not having that background of growing up, actually contributing to the code line, and how projects were actually organized in the code line, I never really had the chance to learn that at Apple.
So I tinker on a little bit on the side myself, and hopefully you'll get to see that on Wednesday if we're able to do that.
PROFESSOR DAN GARCIA
Yeah, that'd be great. Yeah, that'd be wonderful.
JAMES PERCY
I have not touched a single transistor, personally, on anything that we've shipped, unfortunately.
PROFESSOR DAN GARCIA
Sure. Sure, wonderful. This is a great question from Adrian. "What are some advantages of creating your own GPU versus premade GPUs?"
Give me the best case, like the most salient thing of what your thing just blows away, a premade GPU, versus something that you can optimize for your own workload?
JAMES PERCY
Yeah, there's a long answer to that and a short answer. The short answer is performance, and really performance per watt that we deliver to our customers that the end users actually see.
Now there's a longer answer to that that's related to features, and optimizing for the features that we want, and having control over those features that is also true.
But really, at the end of the day, we just felt like Apple wanted to control our destiny when it came to the GPU, and really the experience that our customers had.
And our ability to differentiate between our competitors really relied on us owning the GPU and owning our own design.
And you see that in what we ship today, both in the phone, or the pad, and now the Mac space.
If you compare the MacBook with the M1 to what we were previously shipping — both in the CPU space, by the way, CPU and GPU space — it just blows it away by multiple factors.
So really what we care about at the end of the day is the performance and the experience we deliver to our customers.
PROFESSOR BORA NIKOLIC
By the way, related to that, and tell us if you can't answer this.
There are a few questions in our Piazza about this unified memory architecture and 128-bit bus that I tried to answer. Can you say something about that?
What's a unified memory architecture, and what's up with the 128-bit bus?
JAMES PERCY
I'm not familiar with the questions around the 128-bit bus, so I probably can't answer that.
The unified memory architecture is — well, there's a number of things that are good about it. And it was something that, frankly, AMD was interested in as well.
But typically, if you look at a CPU plus GPU system that have separate memory spaces, you have a lot of copies from the CPU to get data to the GPU so the GPU can use it and access it at high bandwidth and relatively low latency.
But the problem is that those copies back and forth from CPU memory space to GPU memory space are expensive, and can often be the bottleneck of your performance.
And in addition to just having performance issues, there is also power cost to doing that.
So what you find is, in a unified memory architecture where you can avoid those copies, you not only get better performance, you also avoid the energy cost of copying all that data around.
And it's also a simpler programming model for software to use. And so it has multiple benefits when you look at it compared to a non-unified memory architecture.
PROFESSOR BORA NIKOLIC
Yeah, I was pretty close. That will do. I'm not quite sure on the simpler programming.
Don't you have to now be careful about pinning things and stuff like that so that it doesn't escape back to the SSD when it's — JAMES PERCY: You have to be careful about — and I probably need be careful about what I say here, but you certainly need to be careful about race conditions.
Let me put it that way. And so the driver in our software stack has to be aware of avoiding creating race conditions.
But the GPU API is pretty good about being explicit about those. Hopefully that was a somewhat vague answer.
PROFESSOR BORA NIKOLIC
And Dan actually covered some of that in terms of multiple CPUs, I think, in last week. So this is pretty good. I think people can connect the dots now, how does this actually work?
But yeah, this pinning stuff is going to be — it cannot be swapping.
Okay, more questions, I think we got a few more.
PROFESSOR DAN GARCIA
Sure, Benjamin has a good one, which is, "What changes in power management enabled the latest Apple silicon MacBook not to have a fan?"
It's just so amazing that you're able to get all that space back and put a bigger battery in, and all that, without a fan. How did you do that?
I've been burning my lap on Apple laptops for 20 years. How did they do it? That's incredible.
JAMES PERCY
It basically boils down to what I said before about energy, performance efficiency, and energy efficiency.
PROFESSOR DAN GARCIA
Right, but it's not a sacrifice. It's not like, well, let's sacrifice it just to cool it down. You're able to get better performance, but yet cool. It seems like it's mutually exclusive.
You get one of the two, great performance, bad heat, or no heat, but low performance. But how do you do both? That's kind of the question, I think.
JAMES PERCY
I think it comes down to all of the techniques that Apple has built up over multiple years of building low-power SOCs for watches, phones, and iPads, and then applying those techniques to a Mac-class platform have enabled us to do that.
So it's not one single thing that you can point to that said, oh, Apple did this and that was what allowed us to build a fanless MacBook Air.
It's really, and this is what makes it hard, is that it requires a enormously large coordinated effort across software, across architecture, across the design team, with the technology team, with the team that builds the physical cooling solution.
Because there still has to be some sort of dissipation thing in there, even if there isn't a physical fan.
And I left out many of them, but all those different teams have to come to the table with their best foot forward to really optimize around power efficiency.
Because without doing that, then you don't get the type of products that we get. And so it's not a silver bullet, where if you just go do this one architecture change, boom.
You'll be able to deliver the performance that we deliver at the power profiles that we deliver. It's really a lot of effort across a lot of different teams that make it happen.
PROFESSOR DAN GARCIA
And I guess you probably can't answer this so I don't mean to put you on the spot, but it almost feels like, did it come from the top, the guy, Tim said, I want you to make a blank, a MacBook Air, that has no fan?
Or did the engineering, and somehow there was a Eureka moment, where we started to look at the numbers of cooling, and it's like, wait, we think we could do this.
And then it bubbled up to say, hey, make sure you know to advertise it? Because we're trying to build this crazy thing.
So did it come from bottom up, or did it come from the top down? I don't know. You can just say top down or bottom up.
I don't know if that gives away too much.
But it'd be cool if it were bottom up, like, oh my god, some engineer at the lowest level is like, I ran the numbers and we can do this.
And then it bubbled up, and then people got excited about it. Or did it come from a visionary?
JAMES PERCY
I'll be honest with you, Dan, I honestly don't know.
PROFESSOR DAN GARCIA
Right. It's such a big deal. It's such a big deal, that's all.
JAMES PERCY
In fact, I didn't know that the MacBook Air that we just shipped was going to be fanless until the launch video.
PROFESSOR DAN GARCIA
Until it was announced? Come on. Wow.
JAMES PERCY
I didn't. Because the interesting thing is — PROFESSOR DAN GARCIA: Wow, that's a good secret. JAMES PERCY: — I know everything about the chip specs and what we need to build the chips.
And this is an interesting thing about Apple, but whether or not there's a fan on a particular system, there's no real reason for me, as the GPU silicon guy, to know that.
All I need to know is this is the power profile that we're targeting, and how much performance you can deliver at that power target.
And it's really up to the system team who's building the system.
PROFESSOR DAN GARCIA
Look, phones have no fan. My phone has no fan, right? Right, and it has a big old piece of plastic in the back, the silicon thing that doesn't — there's no weight.
Where is the air? Because that's a speaker. Where's this getting cool, right? So they're able to do it for other pieces of hardware, not bigger ones.
That's great.
PROFESSOR BORA NIKOLIC
We're getting fairly low on time because we have a few other things that we need to cover. I do have a couple of questions. There are questions that came in about the management.
I've learned that you do hire quite a few full-time employees, and there are quite a few interns in your group, obviously. This pretty big group needs to come from somewhere.
And I've just learned, I think Jared told me, that they are either gone through or are going through these internship applications. Do you have any tips for interviews?
How do you interview? What happens in these interviews with Apple?
JAMES PERCY
Yeah, excellent. Excellent question and comment. The number one thing with any interview, not just Apple, but I think probably for any interview in any job is be prepared.
And so what I mean by that is, go ahead and ask your recruiter or your manager, hiring manager if you have access to them beforehand, hey, what's the interview going to look like?
The worst thing they can say is, hey, we're not going to tell anything, though I don't think they would say that. But what types of questions should I expect?
What languages should I expect to code in? Know your projects like the back of your hand if they're on your resume.
One of my favorite questions that I like to ask intern and new grad candidates is, tell me about the hardest project that you've ever worked on. Why was it hard?
And what did you learn from it?
Those other people on my team will ask you coding questions, but I want to hear about what kept you up in the lab for two days in a row, and what did you learn from that experience?
The candidates that have not done well with me, frankly, will have some project that they did a couple of years ago on their resume.
And they either can't speak to the details of it, they don't remember, or maybe their friend did most of the project and they weren't very involved. That's not good.
PROFESSOR DAN GARCIA
That's not good.
JAMES PERCY
That shouldn't be on your resume.
PROFESSOR DAN GARCIA
Yeah, we've been telling them that since the beginning. You got to know all of all the projects, exactly.
JAMES PERCY
So be prepared to speak to and talk about anything that's on your resume. Any language that's on your resume, be prepared to code in it.
Any project that's on your resume, be prepared to talk about it and explain it in a lot of detail. Because those are the things that people want to hear about.
And frankly, it's much more interesting as an interviewer to hear about hard projects and things that maybe didn't go that well and that were learning experiences, than just asking, hey, can you go code this C program, or Verilog question, or pick your favorite language?
Sure, there will absolutely be some of that. You should be prepared. But if you're a good candidate, that stuff will come easily.
And it's the projects and the experiences that will really set you apart from other candidates.
The other thing I'll add to that is, also, if you have prior internship experience already, that's great. But same thing, be prepared to talk about it as long as you can.
I respect there'll be some confidentiality issues there. But there is something you came away from that internship that was either really challenging that you liked a lot, or maybe even you didn't like.
You learned that, hey, I don't want to go be a diagnostic person in the lab for whatever reason. Say why.
That really shows us something about the way that you think, that you have opinions about things. Having an opinion about things is great.
You don't necessarily have to agree with your interviewer, but showing that you can actually think independently and you have opinions about something, and you can defend those opinions really goes a long way to show you as somebody that has a lot of potential at Apple.
So that was kind of a long-winded answer, but those were some of my thoughts there.
PROFESSOR DAN GARCIA
Nice. Do we have time for anymore, or was that it, Bora?
PROFESSOR BORA NIKOLIC
Yeah, that's great.
PROFESSOR DAN GARCIA
[INAUDIBLE] PROFESSOR BORA NIKOLIC: Yeah, there's another one. What I've learned also is, our students, apparently those that interviewed for firmware positions aced it. So those have been such — filled, taken in a week.
The 61C student or PAs sucked the air out of that room. [INAUDIBLE] PROFESSOR DAN GARCIA: Right, great job, team.
PROFESSOR BORA NIKOLIC
But you told me that — very good, this C thing is pretty useful, and assembly, too, comes in handy. But you mentioned that your path to what you're doing now came through verification.
We don't teach verification. Do you have a hint? Should people be scared if a verification group approaches them and says, hey, can you come over and work with us on verification?
What should they say?
JAMES PERCY
They should not be scared. Nobody taught me verification either in college. So I basically learned it on the job, I think, as many people do, although not all.
There are verification classes out there at some universities right now. Verification is a fantastic field to get into, and one of the reasons is because good verification engineers are extremely scarce.
And you can use the laws of supply and demand to figure out what that means.
But if it's something that you are interested in and passionate about, you can really have a big, big impact on products.
Because if you've got some critical bug you can't ship with, or you have some horrible software workaround that impacts performance, that's a huge problem and gets a lot of visibility.
In terms of what you need to know to become a verification engineer, you don't have to be a UVM guru. You don't even have to necessarily be a system Verilog guru right away.
All you need to be good at is coding in some reasonably modern language and you will be able to learn UVM, or system Verilog, or OVM, or pick your other favorite verification language.
If you're good at C++, you'll be able to figure out the other things. What you really need is to be able to ask good, hard questions, think through problems.
Challenge what you're given, such that if somebody throws a problem at you, you can ask the right questions to get to the bottom of it.
That's what makes a really good verification engineer, challenging the specs that you're given, like, hey, does this really make sense?
And when we put these two hardware blocks together, are they really going to work? It's having that sort of an inquiring mindset that will make a good verification engineer.
PROFESSOR BORA NIKOLIC
All right. Thank you. Thank you so much.
PROFESSOR DAN GARCIA
Yeah, thanks. Thank you so much. The students, let's give James a hand. Thank you so much for your time together with us, James.
Woo.
PROFESSOR BORA NIKOLIC
[INAUDIBLE] James Percy.
PROFESSOR DAN GARCIA
And by the way, please do come — PROFESSOR BORA NIKOLIC: Thank you, we'll see him again Wednesday.
PROFESSOR DAN GARCIA
Yeah, on Wednesday, come back.
Everybody here, make sure you tell your friends, come on back to this channel at 4 o'clock and James will be giving a live lecture on how GPUs work, exciting stuff, and what they're used for.
Thanks again, James. Take care, my friend. That was great.
JAMES PERCY
See you guys then.
PROFESSOR DAN GARCIA
See you Wednesday.
JAMES PERCY
Take care. Bye.
PROFESSOR DAN GARCIA
All right, perfect. Okay, bye.
PROFESSOR BORA NIKOLIC
Bye.
PROFESSOR DAN GARCIA
All right. So let's do a quick — you've got about eight minutes left. Let's do a Computing in the News.
I found this delightful story, Computing in the News — about a new app that uses computer vision, and certainly a neural net trained on different models of food and caloric content of those models of food to be able to instantly take a picture of your food.
And people do this when they go to fancy restaurants anyway, but to actually extract the caloric content. So I guess you'll serve yourself a plate.
Say, I'm going to eat this plate, and then you take a picture of the plate and it estimates the distance, and the tie, and the size of the scale of things, and then figures out all the calories you're eating.
So that's really important for people who are on limited diets or diabetics.
PROFESSOR BORA NIKOLIC
We'll have to see that figuring out how much oil or fat is in there, maybe.
PROFESSOR DAN GARCIA
Right. I don't want to know.
And certainly you can make things with — it's really hard for curries because you can make, example, saag paneer, which is one of my — it's an Indian dish with spinach and cream.
And you can put a lot of ghee and cream in it, or less. So you make it with water or more creamy.
And that little choice, I don't know how a camera would do it. So I like this idea for big things, but I certainly don't think it's going to be the end all, be all.
Because I can make 10 different recipes, one with basically none of the heavy, fatty stuff, and one with a lot of it.
And it wouldn't be able to tell the difference just because it just looks like a big curry or a big saag paneer.
So I don't know how actually that's going to be, but we'll see. The second one, maybe Bora, you could share some of these things for the second one that I'm highlighting here.
Oops, I'm clicking on it. Click, click. Oh, I did the same one. I did it twice.
There's another one, and I guess I lost the slide. Oh, no, no. Here it is. Yoink.
There it is. All right, sorry. I had two duplicate slides in there. A chip delivers smarter light-powered AI.
And I actually visited — this is Australia's Royal Melbourne Institute of Technology.
This is in Melbourne and I visited them when I gave a talk when I was there in 2007, a million years ago.
It's one of the top institutes of technology in Australia there in Downtown Melbourne. But in theory, they have an AI and a single electronic chip.
Do you know anything about this, or some of that work?
PROFESSOR BORA NIKOLIC
Yeah, I mean, I don't know exactly what it is.
But you guys in your projects two and four, you have figured out a lot of stuff that is happening when running neural net inference.
So generally, what they call AI is neural net inference, or machine learning inference.
And what you'll figure it out, also, is a lot of there is either run in convolutions or matrix multiplications. And you guys became pretty good at that by using standard digital hardware.
PROFESSOR DAN GARCIA
Look, classify numbers. Wait, have you seen that before? Right.
You're welcome, by the way. You're welcome. Right.
PROFESSOR BORA NIKOLIC
No, people have figured out that they can do that on different hardware. You can build specialized hardware which is really good at doing matrix multiplication.
Now the problem there is that hardware generally does not do other things, like for loops, well, or whatever other things are needed there. There are other functions that are in there.
So you need to have some kind of a CPU and a well integrated accelerator.
That hardware accelerator can be made of — I don't know, a light, or some kind of a crazy memory, or anything that — what you'll find out, anything that satisfies Kirchhoff's laws and has no linearity is able to do multiplication, right?
You get the second order non-linearity for free, and it does some sort of a multiplication. Sophia Shao teaches a class.
You guys are not quite ready, but I think there is a big interest in her class, ML hardware, or AI hardware.
Half of the class are guest speakers that come and bring their either conventional, really big matrix multiplication thing, like a PPU that we mentioned in the class, or their funky hardware.
And people from Mythic, I think, came and talked about their analog matrix multiplication. Mythic actually just started shipping right now.
So the jury's out which one of these is going to actually work.
I was told that there are 700 startups trying to make something that is going to do matrix multiplication faster, 70 in the Bay Area.
I mean matrix multiplication, put it this way, neural net inference faster. We're going to do your project two and project four faster than you can.
PROFESSOR DAN GARCIA
Right, that's great. Good stuff. All right. Up to the next one.
I'm going to try something. I guess I can't do it. I can't block most of the bottom. Here's the thing, the next slide topic is final exam information.
We have a bomb to drop. The bomb to drop is, the final exam was scheduled for 12-13, a Sunday, all 24 hours.
The campus sent a big memo to everybody saying that you cannot offer a final exam outside of the final exam schedule. And don't make willy-nilly arbitrary changes. Well, sorry, campus.
This was not an arbitrary change. We had this since the first day of school and we let you all know that, and you planned for it. But they said, well, sorry.
They said back to us, sorry. You have to have it on the day that it was scheduled, and we have to acknowledge that request.
Maybe this is good or maybe this bad.
You now have four more days to study, but instead of having your final exam on a day when there's no other exams, you have your final exam on a day when there are other exams.
And so your final exam is now going to be on Thursday, December 17. And you have 24 hours to do your final exam, as before. And we'll talk about the time limit of that.
But that's what's on the schedule. So you're like, wait, why is 61C saying one day when my final exam schedule says one? We have to honor the final exam schedule.
By the way, in spring '20 we were able to move our final exams out, and it was no trouble.
But I think that's because it was kind of a special semester, and this is a semester that's been online since the beginning. And so maybe it's a little different.
But we said, well, in CS10 we give our — in fact, even CS10, we have one part our final exam on the Sunday, still that Sunday before the final exam week.
But they basically said, no, you have to have a final exam during the time. And so we have to honor what campus said. So this is not our choice.
This is just basically deferring to campus.
PROFESSOR BORA NIKOLIC
There is a clarification question here, right?
The campus memo said that we have to hold the final on the date and in the time window when we are supposed to, which is 8 to 11 AM on Thursday.
PROFESSOR DAN GARCIA
Right. But they also said — PROFESSOR BORA NIKOLIC: However, they gave us freedom.
PROFESSOR DAN GARCIA
They also said, we encourage you to have alternate times. And so that's what allows us to have the 24-hour window. And so they're actually strongly encouraging us to have flexibility.
In fact, we just had a survey for in our EECs faculty meeting this morning.
We had a survey that — maybe it was last week — that said that students want flexibility for their exams.
That is something that's been shown to not — don't just make me proctor and be there exactly at the time.
Give us some flexibility for the time to be able to take a rest and all those things. And so that's what we're doing.
It's what we've been doing all along, so it's exactly the same format. You are going to have a limited time, like we had the midterm.
And in fact, here's the next slide, the big one. Dah-dah-dah-dum. We certainly want to honor and thank the student representatives for their feedback. That's been so wonderful these last couple of weeks.
And they've encouraged us to think about scaling this down to be a three-hour window, and that was actually helpful for us to be able to now honor the campus.
So the two pieces came together to both agree that it makes sense to keep this being a three-hour exam. So we're going to have it on the day.
We're going to offer an exam that will take you a total of — or you'll have no more than three hours total.
You can take it in one sitting if you want to do three hours sitting. You can take it from 8 to 11. So if you say, hey, campus.
Yes, look. See here, I'm honoring you. You asked 8 to 11. I'm taking it.
You can take all three all three spots exactly as a normal exam would be.
You can also do what we suggest, which is take — in any order, take one of them, take the second, take the third, take a break between it, eat lunch between them, fit it in between other exams.
I know it's not the best news for everybody because you were hoping to do it on the Sunday — I was talking about the exams.
But we have to do this on the day of, so we're doing it on, as I said, the 17th. And we broke them into three pieces. Here's the three sections.
There's going to be a quest, which is going to be the quest clobber.
There'll be a question called quest clobber question, a midterm clobber question, which just contain all the stuff from the midterm that isn't on the quest.
And then everything is the final, which is everything past the midterm. That's it, those three pieces.
We realized as we were thinking about the load of this is that more material happened in that midterm window. That was just a ton of stuff in that midterm section, a little less here.
Obviously caches is big, VMs is big, the parallelism is big.
But we're going to put those into smaller questions so that we're going to have — we decided the time to be 90 minutes for the midterm clobber and 60 minutes for the final clobber.
And again, overall that's three hours. And we're going to ask that our TAs are able to ace those exams in a sum of 1.5 hours.
So we're going to hope, and we'll be publishing those stats. CC was very good about collecting those stats for the midterm and then publishing them.
We'll try to do the same thing and share what the median is, what the max is.
And by the way, if the TAs can't do that in half the time, 15, 45, and 30, we're going to cut more material.
We just keep cutting, cutting, cutting until our TAs can get it down to half of that time. And then we'll give you double the time. That's the current plan.
So hopefully you'll be able to finish within those three hours. Yes, Bora?
PROFESSOR BORA NIKOLIC
Yeah, these are currently draft times, proposed times — 30, 90, and 60. Based on how our TAs do, we may change and we may have 30, 80, 70, or something like that.
PROFESSOR DAN GARCIA
Yeah, that's true. Yeah, this is the current design plan.
PROFESSOR BORA NIKOLIC
[INAUDIBLE] PROFESSOR DAN GARCIA: Even as it says, current plan. We don't think we're going to change from a three-section plan. We think that's locked in.
But those times, those three, which is right here, these three times are going to be maybe adjusted depending on what our TAs are telling us in terms of difficulty.
But still, we're trying to cut. That's all we're doing, we're just cutting, cutting, cutting. We have ideas and we're just like, here's this, here's the block of text.
How small can I make it, the setup, before we ask you the question to do? And we'll do that. And by the way, we're going to try to have this graded instantly.
The nice thing about Gradescope — I mean, Gradescope — PrairieLearn, is that if we code the solution into — you won't see it. You'll see nothing different.
You'll take the exam like you normally do. Rather than waiting two months for us to author a really good thing, we're actually pre-authoring the autograder into PrairieLearn.
And we're testing it now in PrairieLearn so that when it's all done, we push a button, it's all graded. And there's no re-grades. It'll just be locked in.
It'll be right or wrong, and we lock it in. And we'll have a lot of small parts. If you think about this, I think it's 60 total points.
I think that's right. So it's just basically a lot of small points, 1, 1, 1, 2, 1, 1, 1, 1, 1, and each one is just right or wrong, and we'll do that.
And by the way, this is an important piece, of the C and RISC-V part, it's going to be a fill-in-the-blank thing.
There's just no time to ask you to go to C, or go to Venus and code something, and bring it back. That's not what we're asking you to do.
It's more the standard kind of exam questions you've had on paper, which is, what is this line? What does it do? What's this line? It's a multiple choice-y thing.
It'll be small pieces of snippet-y code that is, basically, has a single answer. And it is just going to be right or wrong. So that'll be good. No re-grades for that.
It's not going to be, well, I had an answer that's here. It'll be something that there is no other way of doing it other than what we've had.
That's the idea. And each one will be like a one-point question. So if you get it wrong, it's not like it's a 20-point question, you get it right or wrong.
There's only 60 points to go around, and if you look, there's a lot of questions we're going to have.
So it's going to be one, or one point here, and it's going to be one or zero for these small pieces. That's what we're kind of doing.
PROFESSOR BORA NIKOLIC
We are running quite low on time. There is a quick question.
If students are taking a pass, no pass — and there are also questions, I think, on Piazza related to that — do they have to take the final?
It's not a requirement if you think you're passing.
PROFESSOR DAN GARCIA
Right. We do have total course points. We have total course points that Stephen has built, which will give you a sense of your total points.
And by the way, we can't take points away from you. Even if everybody gets an A, we love that, right?
So if you've earned a certain number of total course points above the passing bar, that's it. We're not going to raise the limit for what a passing number is.
There's a passing number. I forget what it is. I don't have it offhand. But if you're above that, there are many classes you don't think — PROFESSOR BORA NIKOLIC: 180, I think.
PROFESSOR DAN GARCIA
Yeah, whatever that is.
PROFESSOR BORA NIKOLIC
Yeah, you need to have 180.
PROFESSOR DAN GARCIA
If you're above that and you're pass, no pass, you're done. You don't have to take the final. So that'll help your own stress and your own time.
PROFESSOR BORA NIKOLIC
I do want to make one more point about pass, no pass. There was a lot of stress about grades, and that's natural this semester.
But just keep in mind that our grade distribution, as Dan mentioned earlier on, we have about 2/3 of a class in the A, A-minus range.
When you consider the pass, no pass, think about that.
PROFESSOR DAN GARCIA
Never happened before. Never happened before, by the way. Never happened before.
The number of students who are going to the finish line with basically A's, and A-minuses, and A-pluses, we've never had the fraction as this semester.
You guys have crushed it, crushed the midterm, crushed the projects. Go ahead, Bora. Sorry.
PROFESSOR BORA NIKOLIC
So yeah, exactly. And in order to get the pass you need a C-minus. Yeah, unfortunately, there are people who are not doing well and not in the passing range.
We are going to give them incompletes if they can fix this, if they can get some part of our course redone and done better. I think we're definitely going to help out with that.
But keep in mind, there is a very small number. A majority of the class is doing well.
So if you're getting an A, just take an A. Don't take a B. That's all that I want to say.
PROFESSOR DAN GARCIA
Right. Yeah, definitely look at your total course points and see how you've done. And if you've done well on the midterm that clobbered your quest, basically, you've got this material.
And I think you could probably get an A if you're doing very well. Don't take a P if you can have an A.
PROFESSOR BORA NIKOLIC
And I've had people completely freaked out saying, oh, my course rank is 800. I'm doing terribly. Yeah, you're in between A-minus and B-plus. I wouldn't call it terrible.
PROFESSOR DAN GARCIA
Right. This is one of those classes where you've done so well as a cohort that you could be way below average but still be getting some kind of A, B thing.
Again, we're not taking grades but we love that. We think we've done a good job if everybody gets an A. I think the midterm was just as hard as every midterm, to be honest.
I think we're trying to do the final the same way. We're not trying to do anything different, so that's the space. So think about that.
So let me just try to give some quick answers. Franklin asked, "Is the final exam strictly inside the time window?" It's inside the 24-hour time.
You have a 24-hour time window on Thursday, December — let me make sure I get the date right — it's Thursday, December 17th. And it's not between 8 and 11.
That's the official Berkeley one. But we're sensitive to time zones and all those things. You've got to make sure it works any time in the 24-hour window.
And you have, again, three sections worth roughly those times. Perfect. Ryota asked, will we be — PROFESSOR BORA NIKOLIC: Those are our alternate times. Those are our alternate times.
They're on the same day.
PROFESSOR DAN GARCIA
Exactly. We gave you alternate times and you make it work within your whole day. That's the idea. And by the way, there's no other day.
That's it, there's no other, like, I can't take it that day. That is the day. Sorry, that's the day, and we gave you 24 hours to make it work within that day.
Will we be allowed to use C? Right now, you are not.
Right now the idea is it's more like — it will still be open notes, but you won't be able to use C or RISC-V. We'll ask you to do something and it's there.
You can pull up your reference material but you cannot use C. Remember, some of the questions before, like in the float, you couldn't use C as a calculator.
Here you can't even use C for the C questions. You've got a C question? Go do it. You can't use C to try to test and see if it's going to work.
You've got to do it as if it were a paper exam, so there's that. Adam's asking about midterm retakes being graded. I think we were close to having that done this week.
I think that's our current plan.
PROFESSOR BORA NIKOLIC
I think it's almost done. Tomorrow, hopefully.
PROFESSOR DAN GARCIA
Yeah, it's almost done.
PROFESSOR BORA NIKOLIC
Sorry that these things do take forever.
PROFESSOR DAN GARCIA
And the midterm grades you saw went out, and the retakes, the midterm retakes are still in that loop. And again, it's the same group of people.
It's the same core elite group of people who are helping us write the final and helping us autograde the last one. So it's leaning on them a lot, and they're very overworked.
They're way overworked this semester. So they're getting to it when they can, and we'll try to get that done. But definitely, look, we're going to push them.
Because the 4th is a pass, no pass deadline.
They need to know that they need to get that done before the 4th so you can make an informed decision about whether you should pass, no pass this course.
So they know that there's a really hard deadline for that, and they're working as hard as they can to get that. Let me look at other things.
Total course points count as an A. Go look at it online. Then you'll see whatever that is. It's on the Policies page. Real coding questions.
There may be a coding question where you have to fill in a lot of blanks.
It could be like, the old final exam RISC-V questions where 10 blanks, and you're writing RISC-V, but you don't have Venus. That might be what we do in this semester.
There might be lines, and lines, and lines of writing C code. You might be doing that, you just don't have C. You have to write it by hand.
So that's a piece of that. We're going to try to do it, and maybe it's multiple choice, maybe it's not. Maybe you're writing it, but that's the space.
And our code can actually do the grading of it automatically. So that's the current way we're thinking about that. Again, you cannot be using C or RISC-V outside of the exam window.
You've got this thing, and we'll be very clear what you can't use. And right now, none of the C or RISC-V questions have you using Venus or C at all.
You've got to do it on paper as an old-school paper thing.
PROFESSOR BORA NIKOLIC
One comment about TCP, the APA points, and this extra bonus point that you're offering for filling out the questionnaire on top of that, whatever it is.
PROFESSOR DAN GARCIA
Oh, yeah. That's right.
PROFESSOR BORA NIKOLIC
That's about the 300 points.
PROFESSOR DAN GARCIA
There's APA points. By the way, you don't know this and it's not in the TCP, but we're going to have APA points, which is an internal number.
I think we talked about this in our first lecture. It's an internal number our TAs are giving you for seeing you do effort, participation, and altruism. That sprinkles on top.
You don't see that until the end. Actually, you never see that number. You just have a final number that's internally what grade you get.
But there's a sprinkled invisible number that you get on top of that. Then there's also extra credit-y points for doing these extra performance projects, the performance part of the projects that are totally volunteer.
You don't have to do it, but if you do it you can get some sprinkled points there.
And on top of that, if 90% of you fill in the survey — nudge, nudge, wink, wink, say no more, say no more — do that survey, course evaluations Berkeley .edu — Bora, do you remember the URL?
I think it's something like that, course-evaluations.berkeley.edu PROFESSOR BORA NIKOLIC: Everybody should have an email [INAUDIBLE].
PROFESSOR DAN GARCIA
Yeah. So you should all be emailed. Check that email, look for course evaluations. By the way, I need students to remind their friends who are not listening to this, they're not reading Piazza.
We're going to try to share this with all of the Piazza. If you fill that in, that's one more point for extra credit on top. All those are on top of it.
All those are on top of TCP that you're getting. And people asking about, will you know the clobber?
If we don't get in by the clobber, I think that Stephen's working to try to get it in. But if you don't get the clobber, you could manually compute it.
You just put it in, and figure out the number, and figure out the thing. We'll try to have some standard deviations out there so you can do that work yourself.
But it'd be better if the software did it for you to get their broad number. So hopefully that's there. I'm good. I don't see any other open questions.
PROFESSOR BORA NIKOLIC
He should be done. I mean He said he was going to get it done before the final.
PROFESSOR DAN GARCIA
Yeah. And they need it before the 4th, really, because they need to know before pass, no pass deadline. Okay? All right, folks.
Again, overall you're doing great. Keep up the great, good, good work. Yes, anonymous attendee, we're going to try to remind people to do that continually.
And by the way, the only thing I can see, right now all I can see is the percentage of students who are in right now who've done it.
So all I can share that — I don't know, oh, Bob and Sue did it, and Ralph and Himay haven't. I don't know that.
And by the way, just so you all know, none of us see — I can't see what you wrote. So to be very frank, it's anonymous.
I can't see what you wrote until my grades are in. So none of us can see what you've written about us. So if you say, Dan was terrible.
Bora was great. Whatever you say, I don't care. I mean, I do care, but I can't see it until all the grades are in. And then it will never affect the grade.
So say what you need to say. Give us some feedback. Please tell us we're great or how we can improve. Put that stuff in, and then that closes.
Then that closes, and by the way, that closes on Sunday night before final exam. This is important. This is another reason maybe why they didn't want to have exam before that.
Because they want to have the evaluations happen Sunday, midnight, and then exams happen, and then grades happen. And then we can see the results of the evaluations that happened a month ago.
So it's kind of this weird thing. So we don't get to see what you said. And we will never like, oh, be vindictive. We were angry.
You said we were bad. We can't do that. Whatever the grades go in, and then we just get the results. But please, honestly, it'd be better to give us improvements.
What can we do? I know a lot of things worked and didn't work this semester. We know about most of those things. But if there are other things, we have a huge feedback form.
If there are other things that didn't get to say, or something happened since the last feedback, please do this. And I believe CC had another form asking about PrairieLearn specifically.
We may end up having a small other survey. Please do fill that out when she does that.
PROFESSOR BORA NIKOLIC
I think we're going to have another survey. That would be [INAUDIBLE].
PROFESSOR DAN GARCIA
We're learning so much in this.
PROFESSOR BORA NIKOLIC
[INAUDIBLE] load this one, this campus survey, heavily with our custom question. [INAUDIBLE] PROFESSOR DAN GARCIA: Right, we can only ask four questions in that, and it's not enough.
It's not enough to ask all the things we want to know about how to improve this class overall.
PROFESSOR BORA NIKOLIC
Yeah, so if can you give us reasonable feedback, suggestions, we will very much be interesting in those.
PROFESSOR DAN GARCIA
Thank you. Thank you. All right. We got to run.
Thank you so much, folks. We'll make sure to share this information on Piazza. See you. Good luck, but please do come on Wednesday.
PROFESSOR BORA NIKOLIC
It's already on Piazza. Yeah.
PROFESSOR DAN GARCIA
We'll see you. Oh, perfect.
PROFESSOR BORA NIKOLIC
Thank you.
PROFESSOR DAN GARCIA
Please do come on Wednesday and come on Friday, both live. We'll see you live again on Wednesday. Thanks, everybody. Take care, folks.
Bye.
PROFESSOR BORA NIKOLIC
Thank you.
PROFESSOR DAN GARCIA
Tell your friends to come, too. Okay, bye.
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.