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
Recording this computer. All right. All right, ladies and gentlemen. Welcome everybody to CS 61C week 13, once a week live sessions thread level parallelism week.
Woo! Great to see you here as always. Wonderful, wonderful, wonderful. Let me go to show you the schedule.
Here's the agenda for today. So again, welcome to week 13. We have a delightful guest, a Berkeley alumnus. Valerie Taylor is here waiting in the green room, waiting to jump in.
So we'll welcome her in a second. Then we have Computing in the News, as we always do. There's a really big announcement from Apple. We'll talk a little bit about that.
We're talking about performance and multicore machines. And boy, that's a — what a great launch that — and we actually had — Valerie, we had Apple folks on Monday.
So it was like the Monday before. So it was like they came on Monday. And they couldn't say, I think there's something exciting happening tomorrow. But we can't tell you.
And then it was pretty exciting.
BORA NIKOLIC
It's funny. We had Anand Shimpi in the session. And he couldn't say anything. But the story went live, what, four or six hours later.
DAN GARCIA
Yeah, yeah. It was a good week. It was a good week for computer architecture, if you're a fan of that stuff.
We have an update on the computer — on the academic dishonesty policy, so some sobering stuff to share. We have the schedule we'll talk about. And then, again, we'll have a ask us anything.
We don't have anything formal to announce about the final. But we had a nice meeting with our students to try to pinch down the time to not have it be a 12-hour final.
So we'll talk a little bit about that. But we don't have the official thing yet. But probably next week we'll have the official word. All right.
Well, it gives me great, great pleasure.
I'm now going to invite out of the green room — add spotlight — to welcome everybody — to introduce everyone to Valerie Taylor, Dr. Valerie Taylor, a Berkeley alumnus, PhD.
She graduated — we just confirmed that Valerie and I overlapped a year when we were both here together. Valerie went to Texas A&M and was the chair — dean also or just a chair?
I remember — I know you had some lofty stuff there, much more above — VALERIE TAYLOR: Right. So I went there as the chair.
And then after I stepped down, I became the senior associate dean.
DAN GARCIA
Yeah, boy. So she was dean for so many years.
And I mentioned on this slide — right here, look at this bullet — she's an incredibly passionate advocate for diversity in high performance computing, which is what the field that she focuses on, and general computing in general.
And Valerie and I have been a part of — Valerie invited me to join her program. Did you found CMD-IT, by the way? Did you — VALERIE TAYLOR: Right.
It was five of us.
DAN GARCIA
Okay. So five wonderful, visionary people founded this wonderful center, Center for Minorities with Disabilities in Information Technology, called CMD-IT.
And Valerie was one of the five multi-headed leads of this, which did so many different initiatives.
And one of the ones that I got involved with was the Academic Career Workshop, which brings folks who are plus or minus two years from graduating and it brings them together.
And we share mentorship. We just basically say, here's what the professorial life is like. Here's what the teaching track world is like. Here's what industry looks like.
Here's what working for the government labs look like.
Just a lot of basic advice to help folks just be very successful as they take their leap into a career, into a career with their PhD.
So they're all PhD folks and just deciding where they're going to go. And we've had folks who are part of the military even come in and talk about that.
So lots of different ways — maybe you can work for Pentagon or Air Force. And maybe you work for a national lab. And Valerie will talk some of that.
So really exciting to have you here, Valerie. We always start with an easy question, which is did you always start knowing you were going to be where you are now?
I mean, some people do. Some kids go to their book, and they draw, on my wedding day — when they're three — on my wedding day, I'm going to walk down with these flowers.
They know the whole thing. Have you always known where you are now? Do you always have that path? Or how did you get to where you are?
Share some of your history.
VALERIE TAYLOR
I'll share my history. I'm happy to do that. So it started with my father. So my father is an electrical engineer — actually, a mathematician.
And so in the house, we always had soldering irons, boards, building circuits. And he and his friends started a telecommunications company and so really exciting.
So I grew up with electronics. And then when I was in high school, I took a programming course. And I know students are going to laugh at this.
And I'm showing my age. I programmed with punch cards. [LAUGHS] DAN GARCIA: 1's and 0's.
BORA NIKOLIC
Me too.
DAN GARCIA
The 1's — BORA NIKOLIC: Me too. 1's.
VALERIE TAYLOR
You had the repeat button. And I went to a Catholic school. So the nuns disabled the repeat button so we could not copy another person's program.
DAN GARCIA
Oh my goodness. So you couldn't do a copy and paste. You mean like no copy and paste? Is that what you meant?
VALERIE TAYLOR
No, no. And a lot of times — DAN GARCIA: Oh my gosh.
VALERIE TAYLOR: --working with Fortran, with punch cards, you may have been off in terms of putting the characters — DAN GARCIA: Right, one column. One column.
VALERIE TAYLOR
One column. So you wanted to scoot over and then hit repeat. But you had to type all over again.
DAN GARCIA
Really? They had to type again? There's value in needing to start from scratch every single time. [LAUGHS] Oh my gosh, I'm so sorry.
VALERIE TAYLOR
Right, and you could easily — DAN GARCIA: And I can't believe you weren't turned off. I can't believe that didn't — like, that experience.
Like, people think, oh, well, should I be in computing or not?
Like, the hill you had to climb for an interest in computing to be able to just do it — not if you're going to be successful, just to do it — was incredible.
Incredible.
VALERIE TAYLOR
Well, at that time, it was funny. For my high school, the computer was with the hospital next door. And we could run our programs on a Tuesday and get the results back Wednesday.
So we could only run our programs once a week. But that's — DAN GARCIA: Once a week? Literally, once a — no, I mean, I've heard the, we had to wait a whole day.
You're telling me you literally had — the whole week was set up for the one big run — batch run — on Tuesday nights to get the results on — unbelievable.
VALERIE TAYLOR
So then — DAN GARCIA: You better get it right the first time. I mean — VALERIE TAYLOR: Right. [LAUGHS] DAN GARCIA: Unbelievable.
BORA NIKOLIC
You bring your necks on Tuesday, and you just run them.
VALERIE TAYLOR
Right.
DAN GARCIA
Was it IBM 360?
VALERIE TAYLOR
That's it. And it was funny because you did flow charts and flow diagrams before you even wrote a line of code.
DAN GARCIA
Right, right. You designed it like crazy. You overdesigned it, actually, just so you'd — VALERIE TAYLOR: You did. That's it.
You made certain. You went through on a piece of paper how things would work — DAN GARCIA: Wow.
VALERIE TAYLOR: --so that that way you tried to get it right when you had the punch cards.
DAN GARCIA
Wow. By the way, students, know that — by the way, back in the day, we used to — people were typing in their PhD theses, 100-page theses, on typewriters.
Do you guys know this at all?
VALERIE TAYLOR
That's it.
DAN GARCIA
Just want you to know that, okay?
BORA NIKOLIC
They probably don't know what a typewriter is. I mean, you have to be honest here.
DAN GARCIA
Imagine a keyboard without a computer where you manually have to key, no.
VALERIE TAYLOR
It had this ribbon that you — DAN GARCIA: Oh my gosh. With the whiteout. Oh my gosh.
VALERIE TAYLOR
But no, here's a funny part I have to share.
DAN GARCIA
Go ahead. Go ahead.
VALERIE TAYLOR
That was when I first went to Purdue, I was so glad they had terminals instead of punch cards. But I had a friend.
He said, no, I need to be able to touch my program. So I'm going to stay with the punch card.
DAN GARCIA
Really? Somebody voluntarily wanted to stay with the old — it's unbelievable. Unbelievable.
VALERIE TAYLOR
But yeah. So things have changed. But I really enjoyed being able to write a set of instructions to actually do different tasks.
And it was just like, oh, wow, really wonderful to do that.
DAN GARCIA
Here's a question. Did you always know you wanted to go into high performance computing?
VALERIE TAYLOR
No.
DAN GARCIA
You liked computing. You were following the path. The rivers were taking you. And you were just taking course after course after course.
But when did you make that — that's a big navigation point, where you say, I want to go into systems and not just that but performance and that kind of thing.
How did you — when did you make that call?
VALERIE TAYLOR
It was when I was an undergrad at Purdue. I took a course with HJ Siegel on parallel processing.
DAN GARCIA
Oh, wow.
VALERIE TAYLOR
And I was like, whoa. So we can have multiple devices working simultaneously. Now, that's — DAN GARCIA: Wow. Wow.
VALERIE TAYLOR
Now, hey. Now, that's wonderful.
DAN GARCIA
What was your language? What was your language that you were using? What were some of the abstractions you were using back then?
VALERIE TAYLOR
So we — [LAUGHS] DAN GARCIA: Go ahead.
BORA NIKOLIC
Oh, Fortran [INAUDIBLE],, yes?
DAN GARCIA
Is that Fortran, right? I'm guessing.
BORA NIKOLIC
[INAUDIBLE] Fortran, right.
VALERIE TAYLOR
Yes, it was — yes. That's why I'm laughing. It was still Fortran.
DAN GARCIA
No, okay.
VALERIE TAYLOR
Yes.
DAN GARCIA
But a parallel version of it. But a parallel version of it.
VALERIE TAYLOR
Right. What you had — at that time, you were using, what was it, WATTFOR? I don't know if you remember that.
DAN GARCIA
Not me. Bora, do you — BORA NIKOLIC: Yeah, no.
DAN GARCIA
Not me. Not me. It was before my day.
VALERIE TAYLOR
You had a WATTFOR compiler. But then you also had Fortran 90 that came out that was doing more, in terms of parallelism, so yeah. And I programmed — what is that?
In grad school, that's where I was programming on Intel Paragon.
DAN GARCIA
Okay.
VALERIE TAYLOR
Yeah. And so Intel had — Intel has been in the HPC market for quite some time.
DAN GARCIA
Sure, sure.
VALERIE TAYLOR
They were in the HPC market with the Intel Paragon. Then they went out of the HPC and came back. But I even — what was it?
And it wasn't parallel, but it was just a big machine. Did some programming on the PDP 11/780.
DAN GARCIA
Sure.
VALERIE TAYLOR
Yeah, because that was at my father's company.
DAN GARCIA
Oh, okay.
BORA NIKOLIC
Oh, digital. Digital.
DAN GARCIA
So you went from a Purdue undergrad, where you touched a little bit of the parallel stuff and you liked it. When you applied to Berkeley, did you say, that's what I want to do?
Or did you apply for a general CS thing, systems, and then decide HPC of the systems angle? Like, when did you — and who did you work with when you were at Berkeley, also?
VALERIE TAYLOR
Okay. So at Purdue, after undergrad, I stayed there for a master's. And the person I worked with was José Fortes. And [AUDIO OUT] doing a lot of work in systolic arrays, which are SIMD.
So systolic arrays are just SIMD machines. And they called them systolic arrays because it models the systole, the systolic system of pumping of the heart.
It's like you're pumping data, because you're just having a single instruction and you're pumping data through the machine.
DAN GARCIA
Oh, interesting.
VALERIE TAYLOR
So every cycle, you're doing these instructions but on different data.
DAN GARCIA
Huh. Huh, okay.
VALERIE TAYLOR
Yeah. So they had these machines. It was a systolic array that came out by — it was NCR, which was the National Cash Register company, came out with systolic arrays.
So we were doing some programming on those and actually looking at how to map applications to those systolic arrays.
The interesting aspect was that the processing elements were 1-bit units.
So we had to break down everything in terms of 1-bit computation. So you would break down your — (LAUGHING) I know. Okay, [INAUDIBLE].
DAN GARCIA
So you're doing it with a float. You're telling me you're adding two floating point numbers by doing something on the bits of them, not just by saying — VALERIE TAYLOR: On the bits.
DAN GARCIA: --float 1 plus float 2 equals float 3?
VALERIE TAYLOR
No, we — DAN GARCIA: You're saying you're literally going into the bit and understanding what each — well, you've got to know what the bits do. I mean — my goodness.
VALERIE TAYLOR
Right. And for [INAUDIBLE],, it was very interesting because the reason why you worked with the bits too is because of the data dependencies with the bit level. (LAUGHING) Yes.
But it was fun. So then when I went to Berkeley, at that time, they were doing work with — it was Sprite operating system.
DAN GARCIA
Sure, Ousterhout [INAUDIBLE],, right?
VALERIE TAYLOR
Right.
DAN GARCIA
I remember that.
VALERIE TAYLOR
And also doing work with RISC.
DAN GARCIA
Yeah, right. That was the early days of RISC. And RAID was also in those days. I mean, Randy was working on RAID.
Dave and team were working on RISC at the same time.
VALERIE TAYLOR
RISC, yeah. Mm-hmm. And John Ousterhout was working on Sprite.
DAN GARCIA
Wow. Wow.
VALERIE TAYLOR
And so what I did, I worked with a professor in EE, because I also enjoyed signal processing. So I worked — DAN GARCIA: You know, Bora is a EE faculty member.
So we're crossing Hearst Avenue in teaching this class, which I think of this class — VALERIE TAYLOR: Woo-hoo. DAN GARCIA: --as really the single class that forms the bridge between the two field.
Who did you work on the EE side?
VALERIE TAYLOR
So it was David Messerschmitt.
DAN GARCIA
Oh, sure. Sure.
BORA NIKOLIC
Right. And David, he's well known for adaptive filters in signal processing.
DAN GARCIA
Wow. So you were actually — that was your application of your parallelism. Or that was the application of the program. Your research was on his — on that project.
VALERIE TAYLOR
Actually — I never go down that straight path.
So when I started working with Dave Messerschmitt, he said, hey, why don't we look at applications outside of electrical engineering computer science? So I started looking at applications with finite element analysis.
So I took a structures class and started doing work in, really, computational science.
DAN GARCIA
Right. I was going to say that. It feels like the heart of computational science.
VALERIE TAYLOR
It does.
DAN GARCIA
The moment you say, let's take what we can do but apply it to a scientific problem, that seems like computational science writ large. That's perfect.
BORA NIKOLIC
And a funny thing about systolic arrays, they were popular back then for trying to — people were trying to map all kinds of applications on systolic arrays.
VALERIE TAYLOR
Right.
BORA NIKOLIC
Had limited success until now. I mean, all these neural processing engines are essentially systolic arrays. And I just — last month, I reviewed — DAN GARCIA: Interesting.
BORA NIKOLIC: — a paper for Journal of Solid-State Circuits that implemented 1-bit systolic arrays. So they went the full circle.
DAN GARCIA
Look at that. Look at that.
VALERIE TAYLOR
[INAUDIBLE] DAN GARCIA: Look at that. What's old is new again.
VALERIE TAYLOR
Right. And my research was in — with finite element, you end up with sparse matrices.
DAN GARCIA
Sure, sure.
VALERIE TAYLOR
So it was looking at architectures for sparse matrix computations, which — yeah. And you're doing [INAUDIBLE].
BORA NIKOLIC
Incredibly popular these days.
DAN GARCIA
So now here's the fun part. When you got to Texas A&M, you get to chart your own research path.
What was the thing that you said I want to spend — get grad students and get a group behind me to work on one problem.
What was the thing that you were most hungry to work on when you left Berkeley?
VALERIE TAYLOR
Well, okay, when I left Berkeley, I actually went to Northwestern.
DAN GARCIA
Oh, okay. I'm sorry. Yeah, all right.
VALERIE TAYLOR
Yeah. So I spent about 11 years at Northwestern. And it was so interesting because at that time, I finished computational science. And so I went to my first supercomputing in '91.
And I was penning the paper. And everybody said, given you're at Northwestern, you have to talk to — and they were like, Rick Stevens. He's head of the mathematics and computer science at Argonne.
And they said, he's a tall guy with long hair. And he does supercomputing. And you go, that describes a lot of people up here. [LAUGHTER] DAN GARCIA: In '91, right?
Right, that was the look.
VALERIE TAYLOR
And you just go — I was like, can you give me a little bit more?
DAN GARCIA
That's funny.
VALERIE TAYLOR
But I finally met Rick. Because I didn't know Argonne, even though I grew up in Chicago.
DAN GARCIA
Sure, sure, sure.
VALERIE TAYLOR
And so people were just like, why don't you come out, give a talk? So I went out, gave a talk.
And it was just wonderful because there were so many people doing work in computational science that I just felt like I found my community. It just was wonderful.
They had computing resources. So at that time, they were getting a J-Machine. So I had a chance to program on the J-Machine with active messages. Yeah.
So it was good.
DAN GARCIA
That's great. And when you got to that community, I'm sure the community was remarkably diverse, and remarkably full of women, and full of people of color and disabilities, right?
I'm sure. Right? Wrong.
VALERIE TAYLOR
Wrong. I know — [LAUGHTER] No.
DAN GARCIA
Right. Right, right, right.
VALERIE TAYLOR
Right. And that — DAN GARCIA: Is that what got you thinking, how do I bring other folks into this family, into this umbrella, into the tent? How do we have a big tent approach?
Was that — VALERIE TAYLOR: Right, because you start to say, well, I don't want to be by myself.
DAN GARCIA
Right.
VALERIE TAYLOR
Right, yes.
DAN GARCIA
Right, across those many dimensions. Right.
VALERIE TAYLOR
That's it. And so I met — who was it? That's how I met Bryant York. I met Roscoe, Roscoe with HPC, and yeah.
DAN GARCIA
Wonderful.
VALERIE TAYLOR
So it's been a great community. Because initially, I started going to the ISCA, the computer architecture conferences. But I was doing application-specific.
And at that time, it was really general purpose because you were right in the heart of Moore's law.
DAN GARCIA
Right, right.
VALERIE TAYLOR
So it's not where anyone was looking at the advantages of application-specific.
But then when I went over to the HPC community, where everything is looking at sparse matrix, sparse matrix computations — one person I talked to, I said, well, given you work with all these science domains and different applications, I said, do you have any examples of dense matrices that result?
Because the sparsity is because it's not the case that if you have a large system, something — one particle or one object would interface with something very far away.
Usually, the forces are not that strong. So usually, you're working within a neighborhood, which gives you the sparse matrix. And he just told me, stop looking. There are none.
Move on. [LAUGHTER] DAN GARCIA: Interesting. So the problems are always localized.
I mean, if you think about it, most of the things that happen, like a climate simulation — if there's a hurricane here, it's not going to affect a hurricane on the far side.
Maybe eventually it will. But things are all local. And that results in a sparse matrix as a result of that. That makes sense.
Yeah, yeah.
VALERIE TAYLOR
Mm-hmm.
DAN GARCIA
Stop looking. [LAUGHTER] That's fascinating. And so then you get to — at some point, you move to Texas A&M. Is that right?
VALERIE TAYLOR
Right.
DAN GARCIA
That's when — you and I more overlap with that. And at that point, you said, let me have both — I don't know how you manage, by the way.
You're one of the busiest people I've ever met.
How do you both manage the leadership of this wonderful group to bring outreach to all these folks as well as maintain — A, be a dean and as well as maintain a research agenda.
How do you have all those fires burning at the same time? I'll ask you again. But how many grad students would you have? And then would they find the problems?
Or would you find the problems and hand it to them? How did you have that conversation, in terms of your own group?
VALERIE TAYLOR
So with my own group, a lot of times, you have senior grad students. And then you bring in a new grad student. And then they're all working together.
So they can see what the senior grad student is working on. And usually — and I always say this — when you're working on a master's, a master's is really good.
And you answer some questions. But usually, a master's will give you more questions than you answer.
Then when you go for a PhD, I always say a PhD, you still have open questions because you've made some assumptions.
But usually, you answer more questions than what you'll have at the end, you know, and — DAN GARCIA: Than you create, yeah.
VALERIE TAYLOR
Right. But a master's is just very fruitful because you'll start down a path, you make a lot of assumptions, and then you realize, and you'll say, wow. Okay, this works for this narrow case.
What happens if I go here, go here? But then you go, I'm at my two years. So I'm ready to go. Bye-bye.
DAN GARCIA
Interesting, interesting. So when you have students who leave with a master's — I'm sure you have; I certainly have as well — do you feel like there's a missed opportunity?
Feel like, oh, if you'd only stayed, there'd have been such good — oh, so close. We were so close.
VALERIE TAYLOR
You do. You do. But then you say, but it is based upon what you want to do.
DAN GARCIA
Right, right. Yeah, sure. And there are actually wonderful career paths for people with just a master's.
I'm just thinking if this is part of the mini advising session for our audience, as you're thinking about what you want to do, there's certainly a ton of opportunities in industry right now.
It's very, very hot. You could also think about a PhD. In advising, I tell them, there's a grad school for everybody. It's not only about the Berkeleys, and the Stanfords, and the MITs.
VALERIE TAYLOR
[INAUDIBLE].
DAN GARCIA
There's a lot of great places out there that would love to have you. Certainly would love to have a Berkeley student.
So it's thinking about if you decided to get a master's or — and funny thing is you can go to master's.
And sometimes you go for a PhD, and you can get a master's on the way to the PhD as like an anchor point.
It's almost like you're rock climbing, and you want to tap in a piece of anchor. I just came back from Yosemite. So I'm thinking about this analogy, so yeah.
You're tapping a piece under the rock in case you fall, in case you decide it isn't working out — life changes, you get married, you have a kid.
And then you have at least the master's to hang it on. Otherwise, you fall back with all that work and you've got nothing to show for it.
So the PhD, always we encourage our master's. But sometimes, you get into master's and you get excited. It's like, wait. I want more.
So you only applied for the master's. And you see, wow, look at all these things I could do. You didn't even realize how much fun it is to do research as a grad student.
And then you decide to jump on — jump more.
VALERIE TAYLOR
Yeah, and that's just what happened. After I got the master's, yeah, you wanted more.
DAN GARCIA
So did you come in for the PhD?
Or did you come in for the master's and decide to get hungry when you got here and want the PhD after that, in terms of your own career?
VALERIE TAYLOR
I actually [AUDIO OUT] the PhD when I went to Berkeley because I finished the master's at — DAN GARCIA: Right, sorry. Yeah, that's right. You mentioned that.
VALERIE TAYLOR
Right. And see, after I finished the master's, I — at first, I only wanted the master's.
But then I ended up working on research instead of the coursework only because I was like, no, courses, that's wonderful, but I'm ready for something a little different.
And I just love the fact that you're given a open problem and there's no one solution.
So I always go, when there's no one solution, I can bring my full self to the problem and answer it in a way that's comfortable for me.
So that's the part I so enjoy versus feeling like there's only one — you have to get this one path right.
So I enjoy the research that gives you that open space to just explore.
DAN GARCIA
Wonderful. Wonderful.
BORA NIKOLIC
We have a question from one of our students, if you don't mind.
DAN GARCIA
Oh, yeah. That's right. We have Q&A. Yeah, yeah, perfect. Yeah, [INAUDIBLE].
BORA NIKOLIC
Ben is asking what technologies do you think will be the key to creating the next generation of supercomputers to advance the field of HPC?
VALERIE TAYLOR
Oh, what technologies — that's a good question.
So I think a couple of things, because right now, you're seeing supercomputers where you have CPUs, multicores, and — combined with GPUs.
But you're also seeing a large number of AI accelerators. So you have the accelerators like Cerebras.
You have the accelerators SambaNova that's with, I think, Kunle and — BORA NIKOLIC: Kunle, yeah.
VALERIE TAYLOR
Yeah, Olukotun at Stanford. You have GRAT, Graphcore.
And so with applications — and I primarily focus on scientific applications — you're starting to see also with those applications where those applications are incorporating AI methods.
And that could be AI methods for surrogate models. And so I think you'll start to see more heterogeneity in the supercomputers.
Then you're also seeing work that's being done in the quantum space.
And if you look at, for example, IBM, IBM came out with a roadmap with the quantum space to say 2023 was the projection for about a thousand qubits.
And so that'll be interesting as well as to what we will explore in the quantum space and having that connect as well with supercomputers.
So I think in another area — I know — and that is you're seeing some work that's being done with FPGAs to do more in terms of domain-specific or application-specific to actually, let's say, work with a particular complex function.
So I think the technologies will be along multiple dimensions, where you'll have different accelerators. Right now, we're seeing a lot in the AI space.
You'll continue with CPUs, GPUs, possibly FPGAs.
And then I think also as we get more in terms of larger scale quantum systems with larger number of qubits, that'll be interesting as well.
So I think it'll be supercomputers will move toward more heterogeneous — DAN GARCIA: It's many flowers blooming phase right now kind of a thing.
VALERIE TAYLOR
Mm-hmm.
DAN GARCIA
Wow. There's more questions, Bora. Do you want to share that? That's great.
BORA NIKOLIC
Yeah. I will. I mean, but we may have to change this class in a few years. So that's kind of worrying me.
DAN GARCIA
We just finished the videos.
BORA NIKOLIC
But if you want me to read the questions — DAN GARCIA: We just finished off the videos. No, no.
BORA NIKOLIC
Do you know somebody who knows how to program these kind of things that you're talking about? Because I don't know anybody.
DAN GARCIA
Right, right.
VALERIE TAYLOR
Right. And here's the part — I think right now you're seeing a lot with programming CPUs and GPUs, where with GPUs, you're seeing a lot with the CUDA programming.
I think Intel is now talking about oneAPI.
DAN GARCIA
Right. I was going to ask about oneAPI, whether there is a way to have a software layer above — an abstraction above everything.
VALERIE TAYLOR
Right, yes. And then you're seeing the work that's being done with the quantum computers in terms of where the quantum work is not where it's connected with supercomputers at this time.
I mean, you have a small number of qubits, in the range of a hundred qubits. So you really want to get to about a thousand qubits until you get to something interesting.
So you're seeing that work that's done with quantum is not done in terms of a connection with supercomputing.
And that's where you've seen a number of centers that were just announced in the quantum space.
And I think it will be a question — and it's a good question to ask — is how to program those systems that's very heterogeneous.
And a question that would come up is, what becomes the compiling that will do the mapping of applications to these particular components?
And I think that's a really good question that you're seeing research that's being done in that space.
DAN GARCIA
Right. Yeah, well, we mentioned actually — that's on my slide.
It says this is a hard problem, and this is an open problem still to — you live in some line — you live in C.
And automatically parallelize your code, automatically map it to all the different devices in a really efficient way. So that's still, again, an open problem.
VALERIE TAYLOR
Right.
BORA NIKOLIC
I'll rephrase the next question a bit. I'll rephrase it this way.
Which kind of previously intractable problems can be solved with better hardware and parallelism?
And they're asking about some of these NP-complete problems that are out there. So which one — what is a big problem that we need to solve?
VALERIE TAYLOR
That's a really good question. That's a very good question.
And I hadn't thought of that question just in terms of NP-complete problems and looking at it in terms of heuristics.
I will comment because today I was attending the supercomputing conference — DAN GARCIA: SC20, yeah.
VALERIE TAYLOR
Right, that's now virtual.
And the one person — Bjorn Stevens from the Max Planck Institute was talking about climate modeling.
And he was talking about the parameters that are needed in terms of doing the climate modeling.
And he talked about this one parameter for which he was saying for such a long time it's not where they had the compute power to be able to solve for this parameter.
But he noted — he said, exascale computing is — you're looking at that in next year, 2021.
So DoE has exascale computers at Argonne as well as Oak Ridge for next year. And so that 10 — DAN GARCIA: You're like this. You're like, [CACKLING].
I cannot wait. Cannot wait to get a hand on that, right? I'm sure.
VALERIE TAYLOR
That's it.
DAN GARCIA
Chomping at the bit.
VALERIE TAYLOR
Yeah, so you're looking at a billion billion operations per second.
So then he started talking about — he said, when we start to look at the scale of the problem that we want, an exascale computer, we're now at the point of having this exascale computer in which we can actually look at solving those problems.
And so he was saying now we're at the point of having some detailed models that we can start to use with climate modeling. So I bring that up.
It's not addressing the question about NP-complete problems but just complex problems in general that need that level of computing to actually solve some of those problems.
And then you'll go the next step. Now that you have the models and the detail of the models to look at all the different interactions, you can start to do what if scenarios.
And that is to say, what if this occurred here? How does that impact climate?
DAN GARCIA
Right. What if we had at least somebody at the front office who's having some climate legislation? Oh, okay. How could that — let's just say.
Let's just say somebody who believed in climate change and was having some legislation. How would that change a number? That kind of thing. That's a good point.
I mean, what I've heard from folks — there's a similar thing I've heard Valerie say that I've heard other places, which is all faster computers let you do is work on larger problems.
Like you were saying, you had little toy things you could do and solve completely and work on that.
But as the computers get faster and faster and you're at the exascale — by the way, ExB, 2 to the 60-something — remember ExB?
2 to the 60th number of floating point operations per second. Now all of a sudden, your input size can be bigger. And you can now do it in reasonable time.
Rather than waiting 50 years, you're now waiting a month, you're waiting a week for that kind of thing. So you can work on higher resolution, larger problems, same kind of thing.
But nothing fundamentally different. The only thing that's fundamentally different from what I've heard is it's quantum. Quantum is fundamentally different in terms of what it's going to give you.
But everything else that's not quantum is just bigger things, a little bit faster, 10 times faster. It's not really cracking a nut that you couldn't crack before, just a larger problem size.
I think that's it.
VALERIE TAYLOR
But I would say that larger problem size is important because, for example, you can take something and the granularity that you may be able to do is something that's 10 kilometers.
And you really need something that's 10 meters instead of 10 kilometers. But you can't do that because it would take months to get the results.
Now you can start to do that type of model. So it is the case you're doing something bigger and faster. But it's also you're doing the analysis that you haven't been able to do.
DAN GARCIA
Right. Yeah, Kathy Yelick — you know Kathy Yelick well, I'm sure. Kathy always shows this video.
No, in CS 10, in our non-majors' BJC — Beauty and Joy of Computing class — she always shows a video of here's what happens if you model hurricanes at 10 kilometer radius.
And you see these blues and this. And here's 1 kilometer, and you can actually see the eddies. Like, there are things that you get revealed. So it is fundamental.
When you go to higher resolution, you can see things you couldn't see before. And because of that, the scientists are like, oh, wow. Look at that.
And they make some new theories based on that. So it is the case that higher resolution does — can be transformation in terms of the analysis you can get of the data itself.
Wonderful.
VALERIE TAYLOR
Right. Yes.
DAN GARCIA
Bora, you want to take the other ones? You want to — BORA NIKOLIC: Yeah, here is another question.
DAN GARCIA
This has been fun. This is great.
BORA NIKOLIC
This is more of a personal question. We have a limited amount of time, so maybe a couple of more.
I feel that an additional four or five years in academia for PhD will be challenging to enjoying the other parts of life, especially after 18 years in K through 12, two years community college, and now two years in Berkeley.
At the same time, I am fascinated in exploring more in EECS.
How do you balance your options to pursuing another four years in academia in grad school versus going into industry immediately and be free of academia?
VALERIE TAYLOR
Well, here's how I balance. I look at the fact that you'll be working 40, 45 years of your life once you get out of school.
So four more years, that's like the 10%.
So with 10%, spending 10% longer in academia can open doors to where that 40 plus years that I'm working, I'm doing what I enjoy.
And that's the part that made the difference. So I'll give you an example.
When I was at Purdue, when I finished the master's, I interviewed after the master's because I thought I wanted to finish and just work.
But I wanted to go to places that were pursuing research because I so enjoyed the open problem of not having this one path. And every place I went, people told me the same thing.
They said, you don't want to come here without a PhD. I was like, really? They were like asking, you know. And I thought, but they're interviewing me.
They go — [LAUGHTER] BORA NIKOLIC: They don't mean it.
DAN GARCIA
You paid for the flight. I'm here. You don't really want me, or what? What is it?
VALERIE TAYLOR
They said — but if you want to actually have more input and actually steer the direction of where you're going, you want to go for the PhD if you want to do the research.
And that's when I decided to go to grad school is that — to continue on to grad school and get the PhD. And in grad school, I had so much fun.
DAN GARCIA
Me too.
VALERIE TAYLOR
I got a chance to interact — so it wasn't where I felt — and grad school is very different from undergrad. In undergrad, you have a few electives.
But you're primarily taking the courses, the required courses. In grad school, you're taking the courses that are of interest to you, which makes a difference.
And then you're exploring problems that are of interest to you as well. And then you have a lot of friends that are in grad school. And we were doing fun things in grad school.
At Berkeley, when I started there, we would go sometimes on a Saturday, hang out at the beach. I went to tie-dye parties because, of course, that's what you do when you're in Berkeley.
And — [LAUGHS] But we had a good time. We went to concerts.
You have a friendship that forms in grad school.
Well, today, I'm still in contact with many of the people I met in grad school because you have this bond where you're doing a lot of work.
But you recognize you need to take breaks. You need a balance. And it's not where you're just working 24/7. You're enjoying yourself as well but have a focus on research.
So it's a really good time in grad school.
BORA NIKOLIC
One kind of related question — or opposite question, say — what motivated you to leave from your professorship and go work at Argonne lab?
Do you still advise grad students is another question. Maybe there are potential candidates over here.
DAN GARCIA
[LAUGHS] VALERIE TAYLOR: Actually, it's very interesting. So when I was at Northwestern, I had a joint appointment with Argonne.
And I organized my teaching schedule such that I taught on Tuesdays and Thursdays. And I was at Argonne on Monday, Wednesday, Fridays.
So I had a office there. I was interacting a great deal.
I thought, for me, I had the best of both worlds of Argonne with the community and also the resources and also teaching, doing research, grad students.
And so then I went to Texas A&M. And I missed Argonne and that interaction. And when I had a sabbatical at Texas A&M, I actually went to Argonne for my sabbatical.
And one — what was it? Argonne invited me back to give a research talk, which I was excited to do. And during my research talk, I was talking with different people.
And Rick asked the question, are you ready to return?
DAN GARCIA
[LAUGHS] VALERIE TAYLOR: You know, I hadn't thought about that. He would say, well — DAN GARCIA: Look at that.
VALERIE TAYLOR
Yeah.
DAN GARCIA
Look at that.
VALERIE TAYLOR
Mm-hmm.
DAN GARCIA
Do you still mentor grad students? Do you still work with grad students as well?
VALERIE TAYLOR
I work with grad students because, for example, I work with a professor at IIT. And we work with her grad students.
Or if I work with someone at UChicago, but it's not where I have a — DAN GARCIA: Not your own. VALERIE TAYLOR: --tenured faculty appointment. And I could pursue it.
But it was just right now, especially being a division director, I focus on Argonne and interact a lot with collaborations there.
BORA NIKOLIC
Yeah, that's a busy job, being a division director.
DAN GARCIA
Folks, do you realize who we've had on these calls? We had the lead from DARPA, the division director here. I mean, this is amazing.
Thank you for taking the time, by the way, from your busy schedule, from all the things you must be doing. My goodness. That's just great.
BORA NIKOLIC
Yeah.
VALERIE TAYLOR
It's a good time right now. [LAUGHS] DAN GARCIA: That's true.
VALERIE TAYLOR
Exciting times.
DAN GARCIA
Maybe it's because holidays coming up. Maybe it's a little quieter kind of a thing. That's true. Yeah, that helps a little bit.
VALERIE TAYLOR
Right. And in supercomputing, it's this week.
DAN GARCIA
Oh, that's right. Yeah, the conference. So the whole lab is going to the thing.
VALERIE TAYLOR
That's it.
DAN GARCIA
That's true.
VALERIE TAYLOR
You usually have supercomputing. And then you have to have it followed by Thanksgiving so you can get that break.
DAN GARCIA
That makes sense. Yeah, makes sense.
VALERIE TAYLOR
Right, and I think — DAN GARCIA: Sure. Sure, sure, sure.
VALERIE TAYLOR
Mm-hmm. [LAUGHS] DAN GARCIA: Bora, what do you think? One or two more?
BORA NIKOLIC
Yeah, I think we are out of time. We [INAUDIBLE].
DAN GARCIA
Maybe you'll take me out — take us out, Valerie, with one thought, which is our students are now doing — as I mentioned before, they're doing SIMD parallelism, thread level parallelism, data level parallelism with MapReduce and Spark.
Give advice, tell us what thoughts you have for our group of students going through the stuff that you do. But they're touching a little bit of it.
They're not really going deep into any — touching here, touching here, touching there, small projects here. What's the advice for a student who likes this stuff? What's your advice for people using this stuff?
Here's an example. Is it maybe use a functional approach because a functional approach means that you don't have to worry about the order of operations? Live in a functional language.
What is some advice you have for our students going through the work now?
VALERIE TAYLOR
Oh. Okay, here's some general advice. So I'm not going to give — because the general advice is to be curious and to try different approaches.
So for example, a lot of things that we do with different applications, we do a lot with MPI with OpenMP.
And the OpenMP gives us the thread level on a given node. And you also combine it with some CUDA and MPI between. But be curious and explore.
And see what happens under different scenarios because that's the way to really get an understanding of what's going on is to try different things.
And it's software. And if things don't work, it's okay.
DAN GARCIA
[LAUGHS] Yeah, it's not chemistry.
VALERIE TAYLOR
That's it.
[LAUGHS] And it's okay. And it's — DAN GARCIA: Wonderful.
VALERIE TAYLOR
Right. So take that opportunity to explore things and know it's okay if things don't work. You can try different things.
Start off with something small that's working, get an understanding, explore different pathways, see what happens if you change some parameters around.
DAN GARCIA
That's great. They actually are all — we're launching them to project 4, which is optimization, try to make this code faster. And we're not telling them how to do it.
So we're giving little hints in here. But again, that's great advice to them. Play with different things. Try this one.
If that doesn't work, try the other one. Maybe you can loop unroll. Maybe you can do something here. Maybe play with more OpenMP and optimize your loop a little bit.
Think about what that is. That's great advice.
VALERIE TAYLOR
Right, right. And you can do blocking — DAN GARCIA: Right, hash blocking.
VALERIE TAYLOR
Right. It's so many — here's the question, too.
Do students get access to different profiling tools as well that may be available and give you more insights into what's going on and maybe hardware counters?
DAN GARCIA
Where are you losing your time? Exactly right. Yeah, yeah. Wonderful.
VALERIE TAYLOR
Right.
BORA NIKOLIC
Well, we haven't covered those in class. But they're very, very useful.
VALERIE TAYLOR
They are, yeah.
DAN GARCIA
Wonderful, wonderful. Well, folks, let's — audience, let's give Valerie a hand. Thank you so much for joining us, Valerie. It was great to see you.
Thank you for taking the time, Valerie. Woo! Amazing. So great.
I love these. This has been so good to see all these old friends and meet some new ones. It's just been — thank you again for the time, Valerie.
This is great.
BORA NIKOLIC
Thank you.
DAN GARCIA
Our students love it.
VALERIE TAYLOR
A pleasure. Thank you for the invitation and enjoy.
DAN GARCIA
Yay. Thanks again.
BORA NIKOLIC
Thank you.
DAN GARCIA
Thanks again for joining us, Valerie. Perfect.
VALERIE TAYLOR
Okay. Bye-bye.
DAN GARCIA
All right. Take care. Bye-bye. Perfect.
All right. So we are up to the next slide, which is a quickie Computing in the News. I've got six minutes left — five. Let's cover this.
You all saw — hopefully, you saw that — and that's actually — I saw a couple responses to a long thread about this — Apple's new hardware release of their own silicon.
We kind of knew that it was coming because they talked about it at WWDC. But it actually hit. And we're very excited to see some of the results that they're talking about.
Eight core GPU, eight core CPU. We're learning about that this week. I mean, this is the TLP week, where we talk a little bit about how that works.
16 core neural engine — all these things on chip. I saw memory on chip. I was amazed by that, by the way, if you saw that. Memory's not in Sacramento anymore.
If memory's on chip, that's a whole other conversation of what happens. It's like, well, caches are on chip, but not memory. It looked like memory was going to be on chip.
It's amazing.
BORA NIKOLIC
Memory is in the package. Memory is in the package. So I think this is like — that's the biggest advantage over here. Memory is not in Sacramento.
If you're in Berkeley, it's like they built a freeway now to Costco in Richmond. [LAUGHTER] So memory is in Richmond, accessible to freeway, yeah.
DAN GARCIA
Yeah, it makes sense. I love the analogy. But it also means — that's going to make — if someone says, Dan, how do you make a computer any faster?
I say, buy more memory. Have the Sacramento trip something that you don't just have to pay all the time. That's a way to fix that. So they've certainly hit that.
And here's the slide. Dun dun dun dum. This is the fastest Apple computers. They looked at the whole lineup of what they have.
And the three M1s — I just want to say the three M1s, you can see — you can't see my screen.
The three M1s in the top three bars are all the new M1 machines. These are, by the way, not even the highest end machines. They normally have a higher end MacBook Pro.
They have the Mac Pro. Can't wait to see what that comes out when they put the M1 chips in there.
But I wanted to point out the red curve, which is the fourth fastest computer on the Apple lineup, is the M1 chip running in emulation mode of the MacBook Air.
Do you realize that? The core i9 is slower than the M1 chip emulating Intel. Unbelievable machine. So this one is — BORA NIKOLIC: But this will be fantasy on real world applications.
I think it's running things that look — DAN GARCIA: That's some benchmark. I don't know what benchmark this is. I don't what the benchmark is.
BORA NIKOLIC
Yeah, this is going to spend — this will be very interesting.
DAN GARCIA
I'm excited. Okay, so that's exciting. Let's get heavy for a second. Let's take a step back.
Bora, do you want to take the lead on this one? You're the one who mostly been working with this. But we can share and tag team this one too.
BORA NIKOLIC
So unfortunately — I mean, during the semester, we have been doing some code comparisons, submission comparisons in this class.
And we are finding a substantial number of cases that look suspicious, put it this way. Some of them are obvious.
There was an obvious plagiarism that happened. So we need to maintain integrity of this class.
Nominally, what we said on our course web page is that there'll be some severe penalties.
And what we said that in this scenario we are going to — in the worst case scenario, we are going to give you an F and send you to the Center of Student Conduct to resolve the situation.
We don't — I mean, we know that this is stressful and these things happen. And we would like to offer a modification to this policy.
And we are offering students who made a mistake to come forward and confess. We don't want to coerce anybody into confessions.
We are offering here if you did something wrong and you know that you've done something wrong, you're going to have a reduced penalty by admitting that you've done something wrong.
That will save us time. And that'll help us a lot.
But if you admit it and you're one of those people that we have on our list of suspected academic dishonesty, we will just give you 50% negative points on that exam or on a project.
We're not going to look at the assignments this semester. There is no point to doing that. But we're looking at projects and the exams.
So instead of getting 0, we are assigning negative 50% penalty because of just submitting nothing and don't put anybody — don't giving anybody extra work to chase that down.
And this will result in non-reportable warning at C of C. That is not something that is visible to anybody. It is just there.
If the offense gets repeated, there is evidence of that so it's known that it is not the first time. There's more severe consequences if this gets repeated.
But the first offense is nothing, stays sealed in there.
DAN GARCIA
This is part of Berkeley's two strike policy, basically. So they say everybody gets to make one mistake. And so this would be if it's your first, it's just your first mistake.
And that information about that academic dishonesty goes into a folder. And no one can look at it. That's the important piece. There's a second slide here.
So we're going to have a form available to fill — and we do not want to coerce anybody. So stop listening if you've been walking a straight line.
But if it is, it's giving you a chance. We will not automatically fail you on a project if you come forward with that.
BORA NIKOLIC
So I actually went and looked through some of these cases. Some of them are unfortunately completely obvious. And we can just assign an F and send this to the Center of Student conduct.
But we are not going to — at this moment, we are allowing anybody to come forward and take a reduced penalty. Things happen.
Now, if you don't do that, we will have to look through the code and establish facts. It is not just going to be some software tool looking at things.
It will be humans that are going to determine whether there has been a case of plagiarism. And it takes time.
So we have to assess at that point more severe penalties.
They are going to — in some cases, those cases that are obvious, we don't really need to have a hearing.
We can just forward this straight to the Center of Student Conduct and assign an F in the class. And we don't have to spend too much time.
There are some cases where I'd like to hear your opinion and would invite you to tell us how is this possible. What are the chances that this kind of a coincidence happens?
There are no further pleas at that time. This is just basically determining what is the penalty for plagiarism.
There are some — but again, we don't want to coerce anybody.
What we would like you to — if you are not sure — if you know that you blatantly copied something, just tell us.
If you collaborated a little bit or a little bit over-collaborated on something — and again, this does not apply to exams. In exams, there is no collaboration.
But on a project, it is possible that you have talked to friends. If you did something that — DAN GARCIA: You had a bug.
And they helped you a little bit with the bug to debug. And next thing you know, they helped you a little bit too much. And now your code looks like theirs.
But it wasn't digital. But there might be a little bit where it's too close because of how they were helping you and how they helped you fix your problem.
And now it looks like their code in how you fixed your problem. These are all squishy. These are all squishy here, these lines.
BORA NIKOLIC
Yeah. In most of these cases, it may happen that you have crossed the line. But if it is something that it is not clear, feel free to ask us.
You don't have to admit anything. Ask us, and we can help you figure out whether this was right or not quite right, not in the spirit of academic collaboration.
So this is going to go on Piazza after the class. And there will be a link over there with the Google Form to be filled out. Be specific.
DAN GARCIA
Yeah. And that's the line of communication.
BORA NIKOLIC
Again, if you didn't do — DAN GARCIA: That's the cleanest way because we've got a thousand students. If we just have email — I don't want it to get lost.
So just make sure nothing gets lost. So go through that form. And it'll all be there on a sheet. We'll be able to find that, yeah.
Go ahead, Bora.
BORA NIKOLIC
If you did nothing wrong, just forget about this.
DAN GARCIA
Forget this. Yeah, ignore all this stuff.
BORA NIKOLIC
You're free. Go. No need to deal with this.
DAN GARCIA
Yep. There's a quickie — this is actually from the Center for Student Conduct. And so what you're seeing here — nobody can see my cursor.
But in the top right — BORA NIKOLIC: Yeah, it's a little hard.
DAN GARCIA: --it says something suspected and the meet with students or — so this instructor does not meet — on the right, it goes straight to basically the disposition form.
It goes on to there. If you meet with students, then you can assume responsibility and do this. Otherwise, your student does not assume responsibility.
So the meet with students, this is the part that we're having the form for. This is the kind of thing — this works great for a 10-person class.
This does not work great for a thousand-person class to have all those meetings. So we're putting this form to basically let you say the things you would have said face-to-face.
But you're saying in a written form that we can review and have that, because oftentimes, nobody ever records these face-to-face meetings. And so some information is lost.
So essentially, this first top left bullet that says "instructor meets with students," this is the chance to have that conversation.
So whatever you would have said to us during your meeting, you're writing down on the form. And that's what we're taking and making a decision.
And in that form, you're either assuming responsibility, which you follow the left block, or do not assume responsibility, the middle one.
And it might be something where you need to talk about it some more. I'm not doing either one. I just want to tell you what happened.
But I don't know if that's actually crossing the line. That's the part that's the third bar. So I'm saying I still need to have a conversation about that.
So there's a third option, where you say, this is what happened. I don't think I cheated. I mean, I don't think I crossed the line, but I may have.
So that's the part that's the middle that a human will come in and have some conversation with you and figure out what that is. Oh, I lost you.
I lost the audio. I lost audio. Yeah, go ahead.
BORA NIKOLIC
Please respond to Piazza privately, to Piazza post, if you're like, hey, this is what happened. Should I fill this out? And we'll respond.
But again, if you do not respond, if you don't admit any of this — we'll give you till Thursday to respond to this.
If you don't, then we have to come to you with the evidence. In some cases, we are going to just notify you. In some other cases, we're going to invite you for a conversation.
DAN GARCIA
All right. Anyway, it's tough stuff. But we really — we take this seriously.
BORA NIKOLIC
Yeah, this is not something that is fun.
DAN GARCIA
We take this seriously.
BORA NIKOLIC
But it happens. It happens.
DAN GARCIA
It happens every year.
BORA NIKOLIC
And we have to review the integrity of this class.
DAN GARCIA
Yep, definitely. All right. Let's take a breath. Let's do a quickie thing, we are — schedule really fast.
We are in week 13. We have three weeks of class left. There's an RRR week and then an exam. So we're looking at four weeks from the exam.
I believe next week we'll be able to have a conversation about what the final will look like. But we're trying to have a final that looks like the old school.
That's our spirit. We actually go into the design session for the final to make it look like the old school exam.
The old school paper exam was a three-hour session where our TAs could take it in an hour and a half. But you had three hours total. That's what we're looking at a design space.
So we're hoping to get that kind of thing with PrairieLearn.
So we're going to be in PrairieLearn and ask the kind of questions we used to ask in the olden days in which if there's any code, it's a very small line of code.
Very small, probably not even asking you to code in RISC-V. Probably not asking you to code C. You're filling in a line in a box not using that.
So you're not going to be — we're looking at the current design to not have you code C or code RISC-V on your own.
It might be write the C code that does this or fill in three lines of this. But it will not be with the benefit of Venus. That's what we're currently looking at right now.
And so we'll figure out where the partial credit can be for all those things. But that's the current space we have for that.
So this week is TLP — Thread Level Parallelism — part 2 and part 3. On Friday, we've got MapReduce and Spark. This is all the parallelism week. It's wonderful.
It's like Shark Week at the Disney Channel — the Discovery Channel. This is parallelism week on the 61C channel. And then next Monday is the last lecture on parallelism, data centers, and cloud computing.
Then we have two days of a Thanksgiving recharge and rest. And I wrote here that project 4 is due on 12/2, which is the Wednesday after Thanksgiving.
I said we moved that to give you some time after Thanksgiving to come on back and finish up what you need to do.
And if you still have leftover slip days, you can use those. So this, in theory, could even push six days past that. That's what we've got.
BORA NIKOLIC
Officially three days for Thanksgiving, actually, Wednesday through Friday.
DAN GARCIA
Yep, that's right. Correct, three days of Thanksgiving — Wednesday, Thursday, Friday. Exactly, exactly. All right.
Almost done. And actually, that was my last slide. That was my last slide. And we're already eight minutes over.
Okay. So I'm looking at questions. Questions — I enjoyed writing my own code. Well, you might be able to write some code but just very small, very small amounts.
But we're not — we're looking — our design space is not to ask you to do that. I'm not officially declaring it yet.
But our design space is to not have you — it will look more like the old school 61C finals. And all those are available on HK and you can look at it.
And we'll try to give partial credits whenever we can. We'll do that there.
Let's see. Can we cite and use other papers and use their cache blocking method? I think that'd be a wonderful idea. Do research on how to do this better.
And as long as you refer to it, that'd be great. I think that's a very reasonable thing to do.
BORA NIKOLIC
Yeah, that's an absolutely brilliant idea.
DAN GARCIA
Love it.
BORA NIKOLIC
Just do it.
DAN GARCIA
Brilliant idea, yeah.
BORA NIKOLIC
Just don't copy somebody else's code because — DAN GARCIA: Exactly. BORA NIKOLIC: --then there is a possibility that somebody else has found that code. And then our — DAN GARCIA: Exactly, exactly.
BORA NIKOLIC: --similarities software finds all of you.
DAN GARCIA
Exactly. Midterm scores were delayed partly because — if you know this, but many of you who took the midterm retake happened this last weekend. And there was a ton of time.
We have basically an inner circle team that's working on exams. And they're the folks working on the autograders. They put together a very early rough draft for autograders that was able to inform.
We didn't want to have to make everybody retake the midterm. We had a list of, I think, 200 or 300 initially.
We pruned it down to — I think it was down to less than a hundred, Bora, if you remember, a hundred-ish plus or minus a little bit.
BORA NIKOLIC
Yeah, it was 90-something, yeah.
DAN GARCIA
Yeah, it was very nice. So we used that first autograder for that. But they were all spending time on this.
And they all spend time on helping answer questions on this — our inner circle team, our exam team. So they were helping with the exam, making the midterm retake happen.
We are just about ready to release the midterm grades. So we're sorry for all this delay. Thank you for your patience. Certainly, thank you for your patience on all this.
We want to do it right. It's much better to wait and do it right than to do it early wrong.
And then there's all the flame, blah, blah, blah, and then make the scores go down. We don't want that to happen.
We want it to be here's at least the minimum scores and your score won't go down. So that's the idea, the goal for the midterm autograders that we're running this stuff through.
So that's really quite important and to do it right. And that's why it's taking so long.
There's one question. Which of the assignments had these issues, had some of the cheating issues? And I think, Bora, we've seen cheating on all 1, 2, and 3 so far.
Is that right?
BORA NIKOLIC
Yeah, we have seen them in different places. I mean, people will find out.
DAN GARCIA
Yeah. Yeah, it looks bad. And people are giving some feedback on the coding session. They thought the midterm retake was really good.
So I'm happy to hear that. That's great. And here we go. So ETA, we're trying to get midterm scores back this week.
That's our — fingers crossed. Again, it's better to be right than fast. What's the three, Bora? You can either be cheap or — no.
Do it right, do it — BORA NIKOLIC: Forgot that.
DAN GARCIA
No, there's three things. You can only have two of them. What's the three — god, what is the thing?
Do it right, do it cheap — anyway, you can't do it right, and cheap, and correct. And so the idea is better to do it slow and correct.
I forget. I'll get it. Two seconds after I cut, I'll remember what it is. But we're going try to do this right and don't want to have to go through many iterations of it.
So we're really trying to do this. We don't to make it — fast, right, and cheap. You can only have two of those three. Exactly.
Thank you.
BORA NIKOLIC
Yeah, I mean, we don't want redos. We want to do it and be done.
DAN GARCIA
Right. And project 3B results — I don't know, Bora. Do we have a ETA on that one and prizes? That's the next thing to work on.
BORA NIKOLIC
We are done. We are done.
DAN GARCIA
Yeah, people were finishing up until last week.
BORA NIKOLIC
We officially will post those now.
DAN GARCIA
Yeah, we should be able to post those soon.
BORA NIKOLIC
Yeah, this week. I think there was still a few people who were left, had some circumstances that extended it. I think maybe even tomorrow or something like that.
So there was the things that happened. We are trying to release those scores this week.
DAN GARCIA
Good stuff. Again, thanks for all your patience. And we do hope this last three weeks isn't as crazy. I know you're working project 4.
And one project always is the dominant. But hopefully, you're learning a lot from it. This is a fun project for a lot of people.
We will ask at the end what you preferred, which was your favorite project, to rank them.
And a lot of people every year — years past say project 4 is their favorite, people who like to do this parallelism stuff. So please do jump in on that.
And we do find, by the way, the labs will help. Remember, labs are no longer required. They're out there. I mean, they're required in the sense that the material is still testable.
But we're not going to ask you to just show up for the check-off. They're called check-ins now. But we do find that these labs are really good to do before you do the projects.
You will learn about and do it in lab to do the project. So please do do the labs. But again, there's no rush on time.
Just get them done before the projects and you'll certainly find doing the projects much more fun and much, much easier to do. All right, folks. Thank you so much.
Bora, good to see you as always, my friend. We'll see you next week.
BORA NIKOLIC
All right. Thank you, Dan. Thank you, all.
DAN GARCIA
And thanks, everybody. We'll see you, folks. Take care, everybody. Do well.
And enjoy — BORA NIKOLIC: All right. Take care.
DAN GARCIA
We'll have one more visit before Thanksgiving. So we'll see you guys there. All right. Thanks, folks.
Bye.
BORA NIKOLIC
Thank you. 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.