YT YouTube2Text

YouTube2Text → Transcripts → [CS61C FA20] Weekly Lecture 13.LIVE - TLP

[CS61C FA20] Weekly Lecture 13.LIVE - TLP — Transcript

by CS 61C Departmental · 11,844 words · 1,834 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Recording this computer.
0:02All right. All right, ladies and
0:03gentlemen, welcome everybody to CS 61C
0:06week 13 once a week live sessions thread
0:08level parallelism week. Woo!
0:11Great to see you here as always.
0:13Wonderful, wonderful, wonderful.
0:15Wonderful. Let me go to show you the
0:17schedule. Here's the agenda for today.
0:20So again, welcome to week 13. We have a
0:22delightful guest, a Berkeley alumnus,
0:24Valerie Taylor is here waiting in the
0:26green room waiting to jump in. So we'll
0:29welcome her in a second. And then we
0:30have computing the news as we always do.
0:32There's a really big announcement from
0:33Apple. We'll talk a little bit about
0:34that. We're talking about performance
0:35and multi-core machines and boy, that's
0:37a what a great what a great launch that
0:40and we had and we actually had Valerie
0:41had Apple folks on Monday. So like the
0:43Monday before so it was like they came
0:44on Monday and they couldn't say I think
0:46there's something exciting happening to
0:47tomorrow but we can't tell you and then
0:49it was pretty exciting. So It's funny we
0:51we had Anand
0:53Selby in in the in the session and he
0:56couldn't say anything and but his story,
0:58you know, went live for four uh
1:01six hours later. Yeah, yeah, it was very
1:03it was a good week. It was a good week
1:04for our computer architecture if you're
1:05a fan of that stuff. Um we're going we
1:08have an update on the computer on the
1:09academic dishonesty policy so some
1:11sobering stuff to share. We have the
1:12schedule we'll talk about and then
1:14again, we'll have a ask us ask us
1:15anything. Uh we don't have anything
1:17formal to announce about the final but
1:18we do have we had a nice meeting with
1:20our students to try to pinch down the
1:22time to not have it be a 12-hour final.
1:23So we'll talk a little bit about that
1:24but we don't have the official thing
1:26yet. But probably with next week we'll
1:27have the official word.
1:29All right. Well, it gives me great,
1:31great pleasure. I'm now going to I'm now
1:32going to invite out of the green room at
1:35spotlight to welcome everybody to
1:37introduce everyone to Valerie Taylor,
1:38Dr. Valerie Taylor, a Berkeley alumnus
1:41PhD.
1:42Uh she graduated and we just confirmed
1:44that Valerie and I overlapped a year
1:45when we were both here together.
1:48Uh Valerie went to uh Texas A&M uh and
1:51was the d- the chair a dean also or just
1:54a chair? I remember I know you had some
1:56lofty stuff there. Much more above
1:59Right, so I went there as the chair and
2:01then after I stepped down I became the
2:04senior associate dean. Yeah, boy. So,
2:06she was dean for so many years and I
2:08mentioned on this on this slide right
2:09here. Look at this bullet. She was she's
2:11like an incredibly passionate advocate
2:13for our diversity in in high performance
2:14computing, which is what the field that
2:16she focuses on and general computing in
2:18general.
2:19And Val and I have been part of Val
2:21invited me to join her program. Did you
2:24found Commended by the way? Did you
2:26There were five of us.
2:28Okay. So, five wonderful visionary
2:30people founded this wonderful center,
2:32Center for Minorities with Disabilities
2:33in Information Technology, called
2:34Commended. And Valerie was the one of
2:37the five kind of multi-headed leads of
2:39this, which did so many different
2:41initiatives. And one of the one that I
2:42got involved with was the academic
2:44career workshop, which brings folks who
2:47are plus or minus two years from
2:49graduating and it brings them together
2:52and we just we share mentorship. We just
2:54basically say, "Here's what the
2:55professorial life is like. Here's what
2:57the teaching track world is like. Here's
2:58what industry looks like. Here's what
3:00Here's what working for government labs
3:02look like." Just a lot of basic advice
3:04to help folks just be very successful as
3:06they could take their take their leap
3:08into the into a career into a career
3:09with their PhD. So, they're all PhD
3:11folks and just deciding how to where
3:13they going to go. And you know, we've
3:14had folks who were part of the military
3:15even come in and talk about that. So,
3:17lots of different ways maybe you can
3:18work for Pentagon or Air Force and maybe
3:20you work for a national lab and Val will
3:22talk some of that. So, really exciting
3:23to have you here, Val. We always start
3:25with an easy question, which is Did you
3:27always start knowing you were going to
3:29be where you are now? Did you I mean
3:30like some people do. You know, some kids
3:32go to their book and they draw On my
3:35wedding day, you know, when they're
3:36three, on my wedding day I'm going to
3:37walk down with these flowers. They know
3:39the whole thing. Have you always known
3:41where where you are now? Did you always
3:43have that path or like how did you How
3:45did you get to where you are? Can share
3:46some of your history. Okay, I I'll share
3:48my history. I'm happy to do that. So,
3:50um, it started with my father. So, my
3:53father is an electrical engineer,
3:55actually a
3:56mathematician.
3:58And so, in the house, we always had
4:01soldering irons, boards, you know,
4:03building circuits. And he and his
4:05friends started a telecommunications
4:08company.
4:09And so, you know, really exciting. So, I
4:12grew up with electronics.
4:15And then, when I was in high school, I
4:17took a programming course.
4:19And I know students are going to laugh
4:21at this. So, um, and I'm showing my age.
4:24I programmed with punch cards.
4:28Ones and twos.
4:30Ones and twos. Right, but
4:32wait,
4:33you had the repeat button.
4:35But, you know, and I went to a Catholic
4:38school. So, the nuns disabled the repeat
4:41button, so we could not copy another
4:44person's
4:46program.
4:47So, you couldn't do copy and paste. You
4:48mean like no copy and paste. Is that
4:49what You can't do that. No. No.
4:52You know, and a lot of times, you know,
4:53you're working with Fortran with punch
4:55cards, you may have been off in terms of
4:57Right. One column. One column. So, you
5:00wanted to scoot over, and then, you
5:02know, hit repeat, but you had to type
5:05all of it in. You had to retype. Really?
5:06They had to type again. There's some
5:07There's something that's value in
5:09needing to start from scratch every
5:11single time.
5:13Oh my gosh, I'm so sorry. Oh.
5:15I can't believe you weren't turned off.
5:17I can't believe that didn't like that
5:18experience. Like people think, oh,
5:20I
5:21should I be computing or not? Like the
5:22the hill you had to climb for an
5:23interest in computing to be, you know,
5:25to be able to just do it. Not forget to
5:26be successful, just to do it. It was
5:28incredible. Incredible.
5:30know, at that time, it was funny. We
5:32had, for my high school, the computer
5:34was with the hospital next door. And we
5:38would be able We could run our programs
5:40on a Tuesday and get the results back
5:42Wednesday. So, we could only run our
5:44programs
5:46once a week. BUT, THAT'S WHAT WHAT ONCE
5:48A WEEK? Literally once a No, I I I mean
5:51I've I've heard the we had to wait a
5:53whole day. You're telling me you
5:54literally had the whole week was set up
5:56for the one big run batch run on Tuesday
5:58nights to get the result on
6:00unbelievable. But then You better get
6:02You better get it right the first time.
6:04I mean
6:06unbelievable.
6:07Bring your decks on on Tuesday and you
6:09just run them. Was it IBM 360?
6:12Yes, and you know, and it was funny cuz
6:14you you did flow charts and flow
6:16diagrams before you even wrote a line of
6:19code.
6:19Oh, right right. You designed it like
6:21crazy. You over designed it actually
6:22just so you You made certain you went
6:24through on a piece of paper, you know,
6:26how things would work so that that way
6:28you tried to get it right when you had
6:30the punch cards. Wow. But by the way,
6:32students students know that By the way,
6:34back in the day we used to People were
6:36typing in their PhD theses, 100 pages,
6:39on typewriters. Do you guys know this at
6:41all? Just want you to know that, okay?
6:42And they probably don't know what the
6:44typewriter is. I mean, you have to be
6:46honest here. Imagine a keyboard
6:49without
6:51computer where you manually the key No.
6:54It had this ribbon, you know, that Oh my
6:56gosh, with the with the whiteout Oh my
6:59gosh. Here's the funny part I have to
7:01share with
7:02Go ahead.
7:02That was when I first went to Purdue, I
7:04was so glad, you know, they had
7:05terminals instead of punch cards. But I
7:08had a friend, he said, "No, I need to be
7:10able to touch my program. So I'm going
7:13to stay with the punch cards.
7:15Really? Somebody voluntarily wanted to
7:17stay with the old
7:20It's like unbelievable.
7:23But yeah.
7:24So things have changed, but I I just I
7:26really enjoyed being able to write a set
7:30of instructions
7:31to actually do different tasks. And it
7:35was just like, "Oh, wow." And
7:38you know, really wonderful. Here's a
7:39question. Did you always know you wanted
7:41to go into high performance computing?
7:43Like if you kind of you kind of you like
7:44computing, you were following the path
7:46that you know the rivers were taking you
7:47and you were just taking course after
7:48course after course. But, when did you
7:50make that you know that's a that's a big
7:51navigation point where you say I want to
7:53go into systems and not just that but
7:54performance and you know that kind of
7:56thing. How did you When did you make
7:57that call?
7:58It was when I was in in undergrad at
8:00Purdue. I took a course with H J Siegel
8:04on parallel processing.
8:06And I was like, "Whoa! So, we can have
8:08multiple devices working
8:10simultaneously." Now that's Now hey Now
8:13that's wonderful.
8:14What What was your language? What was
8:15your language that you were using? What
8:16were some of the abstractions you were
8:17using back then? So, we
8:20Go Go ahead. You go.
8:22Parallel Fortran on me, I guess. It's a
8:24Fortran, right? I'm guessing. Was it?
8:26Yeah, that's why I'm laughing. It was
8:28still Fortran. No, okay. Yes.
8:31But a But a parallel version of it. But
8:32a parallel version of it. Right. Well,
8:34you had at that time you were using what
8:36was it? WATFIV? I don't know if you
8:38remember that. Not me. Boy,
8:41you you
8:42Not me. Not me. This is before my day.
8:44You had WATFIV compiler.
8:46But then you also had like Fortran 90
8:49that came out that was
8:51doing more in terms of parallelism. So,
8:53yeah and I programmed What is that? In
8:56grad school, that's where I was
8:57programming on Intel Paragon.
9:01Okay. Yeah. And so, Intel had Intel has
9:05been in the HPC market for quite some
9:07time. They were in the HPC market with
9:11the Intel Paragon, then they went out of
9:13the HPC and and you know came back. But
9:17I I even What was it? And it wasn't
9:20parallel, but it was just a big machine.
9:23Did some programming on a PDP 11780.
9:27Sure. Yeah.
9:28So, cuz that was at my father's company.
9:31Oh, okay. Oh, digital.
9:33Digital.
9:35So, you went from you went from a Purdue
9:36undergrad where you kind of touched a
9:38little bit of the parallel stuff and you
9:40liked it. When you applied to Berkeley,
9:41did you say that's what I want to do or
9:43did you apply for kind of a general CS
9:45thing systems and then decide HPC of the
9:48systems angle? Like when did you kind of
9:50and who did you work with when you were
9:51Berkeley also? Okay, so at Purdue after
9:55undergrad, I stayed there for a
9:56master's.
9:57The person I worked with was Jose
10:00Fortes.
10:02We're doing a lot of work in systolic
10:04arrays. Which are SIMD. So, systolic
10:08arrays are just SIMD machines and they
10:11called them systolic arrays because it
10:14models the systole, the systolic system
10:17of pumping of the heart. Mhm. It's like
10:19you're pumping data because you're just
10:21having a single instruction and you're
10:23pumping data through the machine. Oh.
10:25So, every cycle you're you're doing
10:28these instructions but on different
10:31data.
10:32Huh. Huh. And so, yeah, so they had
10:34these machines, um,
10:37it was a systolic array that came out by
10:40it was NCR, which was the National Cash
10:43Register company came out with systolic
10:46arrays. So, we were doing some
10:47programming on those and actually
10:50looking at how to map applications to
10:53those systolic arrays. The interesting
10:56aspect was that, um,
10:59the the processing elements were one-bit
11:03units.
11:04So, we had to break down everything in
11:07terms of
11:08one-bit computation. So, you would break
11:11down your
11:14Oh, so if you So, so you're doing it
11:17with a float. You're telling me you're
11:18adding two floating-point numbers by
11:19doing something on the bits of them, not
11:21just by saying float float one plus
11:23float two equals float three. You're
11:25you're saying you're literally going
11:26into the bit and understanding what each
11:28Well, you got to know what the bits do.
11:29I I my goodness.
11:31And so, it it was very interesting
11:33because you the reason why you work with
11:35the bits, too, is because of the data
11:37dependencies with the bit level. Mhm.
11:40And so,
11:41yes. So, but it was fun. So, then when I
11:44went to Berkeley, at that time, they
11:47were doing work with um
11:50it was Sprite operating system. Sure.
11:53Ousterhout. A whole bunch of Sprites,
11:55right? Right. And also, um
11:59doing work with RISC. Mhm. Yeah, right.
12:01Those were the early days of RISC. Those
12:03were the early days of RISC. And RAID
12:04was also in those days. I mean, Randy
12:05was working on RAID. Dave and team were
12:07working on RISC at the same time. Yeah.
12:09Yeah. Mhm. And John was Ousterhout was
12:12working on Sprite. Wow.
12:14but what I did, I worked with a
12:17professor in EE, um because I also
12:20enjoyed signal processing. Okay. You
12:22know, Bo Bo is a double E faculty
12:24member. So, we're we're crossing the
12:25Hearst Avenue in teaching this class,
12:27which I think of this class is really
12:28the the single class that forms the
12:30bridge between the two fields. Who who
12:31did you work on the double E side? So,
12:33it was David Messerschmitt. Oh, sure.
12:36Sure.
12:36Right. And David is he's well-known for
12:38adaptive filters in signal processing.
12:42So, you were actually that was your
12:43application of your parallelism, or that
12:44was the application of the of the pro
12:46program you were you were your your your
12:47research was on on his on on that that
12:50project. Actually, so here's I I you
12:53know, I I never go down that straight
12:54path.
12:57So, when I started working with Dave
12:58Messerschmitt, he said, "Hey, why don't
13:01we look at applications outside of
13:03electrical engineering computer
13:05science?"
13:06So, I started looking at applications in
13:09with finite element analysis. So, I took
13:12a structures class and, you know,
13:15and started doing work in really
13:17computational science. Right. I was
13:19going to say that. That's it. It feels
13:20like the harder computational science.
13:21The moment you say, "Let's take what we
13:23can do, but apply it to a scientific
13:24problem." That seems like computational
13:26science at large. That's perfect.
13:28Funny thing about, you know, systolic
13:30arrays, they were popular back then for
13:32trying to People are trying to map all
13:34kinds of applications on systolic
13:37arrays.
13:38Right. You know, had limited success
13:40until now. I mean, all these neural
13:42processing engines are essentially
13:43systolic arrays. And I just last month,
13:46I reviewed a paper for Journal of
13:47Solid-State Circuits that
13:49implemented one-bit systolic arrays. So,
13:52they went to the
13:53Look at that. Look at that.
13:56And you know, Old is new again. Right.
13:58And my research was in, you know, with
14:00finite element, you end up with sparse
14:02matrices.
14:04So,
14:05yes. So, it was looking at architectures
14:07for sparse matrix computation. Right.
14:10Right. Right.
14:11And you know, Incredibly popular these
14:13days.
14:15So, so now here's the fun part. When you
14:17got to Texas A&M, you get to chart your
14:19own research path. What was like the
14:21thing that you said, "I wanted, you
14:22know, I want to spend get grad students
14:24and get a a group behind me to work on
14:26one problem." What was the thing that
14:27you were like most hungry to work on
14:29when you left Berkeley?
14:31Well, okay. When I left Berkeley, I
14:32actually went to Northwestern. Oh, okay.
14:35Sorry. Yep. All right. So, I spent about
14:3811 years at Northwestern, and it was so
14:40interesting
14:41because at that time, you know, I
14:43finished computational science. And and
14:46so, I went to my first supercomputing in
14:49'91. And I I was sending a paper, and
14:52everybody said, "You know, given you're
14:54at Northwestern, you have to talk to"
14:56and it, you know, they were like, "Rick
14:58Stevens. He's head of the mathematics
15:00and computer science at Argonne." Mhm.
15:03And they said he's a tall guy with long
15:05hair. You know, you go to
15:06supercomputing, and you kind of go,
15:08"That kind of describes a lot of people
15:11there."
15:12In in '91, right? That was the right?
15:14That was the look. I WAS LIKE, "OKAY,
15:15YOU KNOW, I WAS JUST LIKE, CAN YOU GIVE
15:17me a little BIT MORE?"
15:20THAT'S FUNNY. BUT I FINALLY MET RICK,
15:23you know, cuz I didn't know Argon even
15:25though I grew up in Chicago. Sure, sure,
15:27sure. So people just like why don't you
15:28come out give a talk? So I went out gave
15:31a talk and it was just wonderful because
15:34there were so many people doing work in
15:36computational science that I just felt
15:39like I found my community. Mhm.
15:42It It just was wonderful. They had
15:44computing resources. So at that time
15:47they they were getting a J machine. So I
15:50had a chance to program on the J
15:51machine, you know, with active messages
15:54and
15:55Yeah, so it
15:56great. And And when you And when you got
15:57to that community, I'm sure the
15:58community was remarkably diverse and
16:00remarkably full of women and full of
16:02people of color and disabilities, right?
16:04I'm sure, RIGHT? RIGHT? WRONG.
16:12RIGHT, RIGHT, RIGHT, RIGHT, RIGHT. SO IS
16:13THAT IS THAT WHAT got you thinking like
16:15how do I bring other folks into this
16:16family, into this umbrella, into the
16:18into the tent? How do we have a big big
16:19tent of Right. Cuz you start to say,
16:21well, I don't want to be by myself.
16:23Right. Right. Right. Across as many
16:26dimensions, right?
16:27That's it. And so I met Who is it?
16:29That's how I met Bryant York. I met
16:31Roscoe. You know, Roscoe with HPC and
16:35yeah. So it's been a It's been a great
16:38community cuz it initially I started
16:40going to the ISCA, the computer
16:43architecture conferences.
16:45But But I was doing application
16:47specific. And at that time it was really
16:50general purpose because you were right
16:52in the heart of Moore's law.
16:54Right. Right. It's not where anyone was
16:57looking at the advantages of application
17:00specific. But then when I went over to
17:02the HPC community where you know,
17:05everything is looking at sparse matrix,
17:08you know, sparse matrix computation. You
17:11know, one person I talked to, I said,
17:13well,
17:14given you work with all these science,
17:16you know, domains and different
17:18applications. I said, do you have any
17:20examples of dense matrices that result?
17:24You know, cuz the sparsity is because
17:26it's not the case that if you have a
17:27large system,
17:29something, you know, one particle or one
17:33object will interface with something
17:35very far away. Usually, the forces are
17:38not that strong. So, usually, you're
17:39working within a neighborhood, which
17:42gives you the sparse matrix. And he just
17:44told me, "Stop looking. There are none.
17:47Move on."
17:49Interesting. So, the So, so the problems
17:51are always localized. I mean, you think
17:52about If you think about most of the
17:53things that are happening, like, you
17:55know, a climate simulation. You know, if
17:57if there's a hurricane here, it's not
17:58going to affect a hurricane on the far
18:00side. Maybe eventually it will, but
18:01things are all local. I mean, and that
18:03and that results in a sparse matrix as a
18:05result of that. That makes sense. Yeah.
18:06Yeah. Mhm.
18:07Stop looking.
18:12That's fascinating. And so then then you
18:14get to At at some point you moved to
18:15Texas A&M. Is that right? That's when
18:17you and I more more overlap with that.
18:19And at that point you said, "Let me have
18:21a have a both I don't know how you
18:23manage, by the way. I You're like one of
18:24the busiest people I I've ever met. How
18:26you both manage a leadership of this
18:28wonderful group to bring outreach to all
18:30these folks, as well as maintain a be a
18:33dean, and as well as maintain a research
18:34agenda?" Like, how do you How do you
18:36have all those fires burning at the same
18:38time? Like, how many I I was going to
18:40ask how many grad students would you
18:41have? And then would you Would they find
18:43the problems or would you find the
18:44problems and hand it to them? How did
18:45you have that conversation in terms of
18:47your own group? Uh So, with my own
18:49group, uh a lot of times, you know, you
18:52have senior grad students, and then you
18:54bring in a new grad student. And then
18:56they're all working together, so they
18:58can see what the senior grad student is
19:00working on. And usually, and I always
19:03say this, when you're working on a
19:05master's, a master's is really good, but
19:09and you answer some questions, but
19:12usually a master's will give you more
19:14questions that you than you answer. Mhm.
19:18Then, when you go for a PhD, I always
19:21say a PhD, you still have open questions
19:24cuz you made some assumptions.
19:26But, usually you answer more questions
19:31than what you'll have at the end, you
19:33know, in the
19:34yeah. Right. But, a master's is just
19:37very fruitful cuz you'll start down the
19:39path, you make a lot of assumptions, and
19:41then you realize
19:43and you'll say, "Wow. Okay, this works
19:45for this narrow case. What happens if I
19:48go here or go here?" But, then you go,
19:50"I'm at my 2 years, so I'm ready to go.
19:53Bye-bye."
19:54Interesting. Interesting. So, when you
19:56have when you have students who leave
19:57with the master's, I'm sure you have I I
19:59certainly have as well. Do you feel like
20:00there's a missed opportunity? Feel like,
20:02"Oh, if you'd only stayed, there would
20:03have been such good bro Oh, so close. We
20:06were so close."
20:07You do. But, then you say, "But, it is
20:10based upon what you want to do." Right.
20:12Right. Yeah, sure. And there are there
20:13are actually you know, a wonderful
20:15career path for people with just a
20:16master's. I'm just thinking if this is
20:17part of a mini advising session for our
20:19from our audience, as you're thinking
20:20about what you want to do, there's
20:22certainly a ton of opportunities in
20:23industry right now. It's very, very hot.
20:25You could also think about a PhD. I as
20:27in advising, I tell them, "There's a
20:28grad school for everybody. It's not only
20:30about the Berkeley's and the Stanford's
20:32and the MIT's and there's a lot of great
20:34places out there that that would love to
20:37have you." Certainly love to have a
20:38Berkeley student. So, it's it's it's
20:40thinking about if you decided to get a
20:41master's or And funny thing is, you can
20:43go for a master's and sometimes you go
20:45for a PhD and you can get a master's on
20:47the way to the PhD as kind of like an
20:49anchor point. It's almost like you're
20:50rock climbing and you want to tap in a
20:52piece of anchor. I just came from back
20:53from Yosemite, so I'm kind of thinking
20:55about this analogy here. You tap in a
20:57piece under the rock in case you fall,
20:59in case you decide it isn't working out.
21:01Life changes, you get married, you have
21:02a kid, and then you have at least the
21:04master's to hang it on. Otherwise, you
21:05fall back with all that work and you got
21:07nothing to show for it. So, the PhD
21:09always we encourage our master
21:11But sometimes you get into masters and
21:12you get excited. It's like, wait, I want
21:13more. So, you only applied for the
21:15masters and you say like, wow, look all
21:16these things I could do. You didn't even
21:18realize how much fun it is to do
21:19research as a grad student and then you
21:20decide to jump on jump jump jump more.
21:23And that's what happened when after I
21:25got the masters, yeah, you want more.
21:28So, you did you come in for the PhD? Did
21:30you come in for the masters and decide
21:31to get hungry when you got here and want
21:33the PhD after that in terms of your
21:34career? PhD when I went to Berkeley.
21:37Because I finished the masters at um
21:39Right, I sorry, right. That's right.
21:40That's right. Yeah, yeah, you mentioned
21:41Yeah, you mentioned that. You mentioned
21:42that. I I said after I finished the
21:43masters, I at first I only wanted the
21:45masters.
21:47And but then I ended up working on
21:49research instead of the course work only
21:51because I I was like, you know, courses,
21:53that's wonderful, but I'm ready for
21:54something a little different. And I just
21:57love the fact that you're given a open
21:59problem and there's no one solution.
22:03So, I always go when there's no one
22:05solution, I can bring, you know, my full
22:09self to the problem
22:11and answer it in a way that's
22:13comfortable for me.
22:15So, that's the part I I still enjoy
22:17versus feeling like, you know, there's
22:19only one you have to get this one path
22:22right. So, I I enjoy the research that
22:25gives you that open space to just
22:27explore.
22:27Wonderful.
22:28Wonderful.
22:29We have a question from one of our
22:30students. Oh yeah, so we have we have we
22:31have Q&A. Yeah, yeah, perfect.
22:33Ben is asking, what technologies do you
22:36think will be the key to creating the
22:39next generation of supercomputers to
22:41advance the field of HPC?
22:44Oh, what technologies? That's a good
22:46question. So, I think a couple of things
22:49because right now you're seeing
22:51supercomputers where you have CPUs, you
22:54know, a multi-core
22:56and combined with GPUs.
22:59But you're also seeing um a large number
23:03of AI accelerators. And that is so you
23:08have the accelerator
23:10like Cerebras.
23:12You have the accelerators, you know,
23:15SambaNova that's with I think Kunle and
23:19Kunle, yeah.
23:20Yeah, Olukotun at Stanford. You have
23:22Graphcore.
23:25And so with applications and I primarily
23:28focus on scientific applications, you're
23:31starting to see also with those
23:33applications where those applications
23:36are incorporating AI methods. And that
23:38could be AI methods for surrogate
23:40models.
23:41And so I think you'll start to see more,
23:46you know, heterogeneity in the
23:48supercomputers.
23:49Then you're also seeing work that's
23:52being done in the quantum space.
23:55And if you look at for example IBM,
23:58IBM came out with a roadmap with a
24:01quantum space to say, you know, 2023 was
24:05the projection for about 1,000 cubits.
24:09And so that'll be interesting as well as
24:13to, you know, what we will explore
24:16in the quantum space and having that
24:18connect as well with supercomputers. So
24:22I think, you know, in another area
24:25and that is you're seeing some work
24:27that's being done with FPGAs to do more
24:30in terms of domain specific or
24:33application specific to actually, let's
24:36say, work with a particular complex
24:39function.
24:41So I think the technologies will be
24:43along multiple dimensions where you'll
24:46have different accelerators.
24:48Right now we're seeing a lot in the AI
24:51space. You'll continue with CPUs, GPUs,
24:55possibly
24:57FPGAs. And then I think also as we get
25:01more in terms of
25:02larger scale quantum systems with larger
25:06number of qubits
25:07that'll be interesting as well. So, I
25:09think it'll be supercomputers will move
25:12toward more heterogeneous.
25:14Mhm.
25:16I think many flowers blooming is phase
25:19right now, right? Kind of a thing.
25:21Well, there's more questions. What do
25:22you want to you want to
25:23Yeah, I Well, I mean but we we may have
25:26to change this class in a few years. So,
25:27that's kind of worrying me.
25:30We just finished the video. We just
25:33finished off the video.
25:35No, no. Do you do you know somebody who
25:37knows how to program these kind of
25:38things that you're talking about?
25:40Because I don't know anybody. Right.
25:42Right.
25:43Right. So, here's the part I think right
25:45now you're seeing a lot um,
25:48you know, with programming CPUs and
25:50GPUs. You know, where with GPUs you're
25:53seeing a lot with the CUDA programming.
25:56I think Intel is now talking about one
25:59API. Mhm. Right. I was going to ask you
26:02about one API, whether there is a way to
26:04kind of have a software layer above an
26:05abstraction above everything, right?
26:07Right. Yes. And then you're seeing the
26:09work that's being done with the quantum
26:12computers
26:13um, in terms of where the quantum work
26:17is not where it's connected with
26:18supercomputers at this time. I mean, you
26:21have a small number of qubits
26:24in the range of
26:26100 qubits. So, you really want to get
26:28to about 1,000 qubits, you know, until
26:31you get to something interesting. So,
26:33you're seeing that work that's done with
26:35quantum is not done in terms of a
26:37connection with supercomputing. You And
26:40that's where you're seeing a number of
26:43centers that were just announced in the
26:45quantum space.
26:47And I think, you know, it will be a
26:50question and this a good, you know,
26:52question to ask is how to program um
26:56those systems that's very, you know,
26:58heterogeneous. And the question that
27:00would come up is, you know, what becomes
27:03the compiler
27:04that would do the mapping of
27:06applications to these particular
27:09components. And I think that's a really
27:11good question that you're seeing
27:13research that's being done in that
27:15space. Right. Yeah, well, we mentioned
27:17actually that's a from my slide is like
27:19it says it's a hard problem and this is
27:21an open problem still to be a you live
27:23in some language you live in C and
27:25automatically paralyze your code,
27:27automatically map it to all the
27:28different devices to to in a in a really
27:30efficient way. So, that that's that's
27:32still going to be an open problem. Let
27:33me rephrase the next question a bit. Can
27:37I would I would rephrase it this way.
27:39Which kind of
27:41um previously intractable problems can
27:45be solved with a better hardware and
27:48parallelism?
27:49Um
27:50and you know, they're asking about some
27:52of these NP-complete
27:54uh problems that are out there. So,
27:57which one what is the what's the big
27:59problem that we need to solve?
28:01That's a really good question.
28:04That's a very good question. And you
28:06know, I I you know, I
28:09I hadn't thought of that question in
28:10just in terms of NP-complete problems
28:13and and looking at it in terms of
28:15heuristics. I will comment um
28:18because today I was attending the
28:20supercomputing conference. Mhm. Okay.
28:23SC20, yeah. Right. You know, that's now
28:25virtual.
28:27And the one person um
28:31Bjarne Stroustrup um
28:34from the
28:35um Max Planck Institute was talking
28:38about climate modeling.
28:40And he was talking about the parameters
28:43that are needed in terms of
28:47doing the climate modeling.
28:49And he talked about this one parameter
28:52for which he was saying for such a long
28:54time is not where they had the compute
28:57power
28:58to be able to, you know, solve for this
29:01parameter.
29:02But he noted, he said, you know,
29:05exascale computing
29:08is you're looking at that in next year,
29:112021. So, DOE has exascale computers at
29:16Argonne as well as Oak Ridge for next
29:20year. And so, that's
29:2210, you know,
29:25I cannot wait. Cannot wait to get a hand
29:27of that, right? I'm sure.
29:29That's it. It's
29:31Yeah, so you're looking at a billion
29:33billion operations per second.
29:36So, then he started talking about he
29:38said, you know, when we start to look at
29:40the scale of the problem that we want,
29:44an exascale computer, we're now at the
29:46point of having this exascale computer,
29:50in which we can actually look at solving
29:52those problems. And so, you know, he was
29:55saying now we're at the point of having
29:58some detailed models
30:00that we can start to use with climate
30:02modeling. So, I bring that up. It's not
30:05where
30:06you know, addressing the question about
30:07NP-complete problems, but just complex
30:10problems in general that need that level
30:14of computing to actually solve some of
30:18those problems. And then you'll go the
30:20next step.
30:22Now that you have the models in the
30:24detail of the models to look at all the
30:26different interactions, you can start to
30:29do what if scenarios.
30:31And that is to say, what if this
30:34occurred here? How does that impact
30:36climate? Right. What if we had at least
30:38somebody at the at the front office
30:39who's having some climate legislation?
30:41Oh, okay. How could that Let's just
30:43Let's just say Let's just say somebody
30:45who believed in climate change and was
30:47having some legislation, how would that
30:48change the number? That kind of thing.
30:50That's a good point. Yeah, I mean, what
30:51I've heard from folks is this is the
30:52same thing I've heard
30:53I've heard in other places, which is all
30:55all faster computers let you do is work
30:57on larger problems. Like you were
30:59saying, you you had little toy things
31:00you could do and solve completely and
31:02work on that. But as these computer as
31:04the computers get faster and faster and
31:05you're at the exascale, by the way, exbi
31:07two to the 60-something. Remember exbi?
31:09Two to the 60 per second. Now all of a
31:11sudden your your input size can be
31:13bigger and you can now do it in
31:14reasonable time. Rather than waiting 50
31:16years, you're now waiting a month,
31:17you're waiting a week, you're for that
31:18kind of thing. So you can work on higher
31:20resolution, larger problems. Same kind
31:21of thing. But nothing fundamentally
31:22different. The only thing that's
31:23fundamentally different what I've heard
31:24is is quantum. Quantum is fundamentally
31:26different in terms of what it's going to
31:27give you, but everything else that's not
31:29quantum is just bigger things, a little
31:30bit of faster, a little bit faster, 10
31:31times faster. It's not really cracking a
31:33nut that you couldn't crack before, just
31:34a larger problem size. I think that's
31:36it. It is But but I would say that
31:38larger problem size is important
31:40because, for example, you can take
31:42something and the granularity that you
31:45may be able to do is something that's 10
31:47km.
31:48And you really need something that's 10,
31:51you know,
31:52um 10 m instead of
31:54Right. in, you know, kilometers. But you
31:57can't do that because it would take, you
31:59know, months to get the results. Now you
32:02can start to do that type of model. So
32:04it is the case you're doing something
32:06bigger and faster, but it's also you're
32:09doing the analysis that you haven't been
32:11able to do. Right. Yeah, Kathy Yelick,
32:14you know Kathy Yelick well, I'm sure.
32:15Kathy always shows this video. No, she
32:17in the CS 10, in our non-majors BJC BJC
32:19computing class, she always shows a
32:20video of here's what happens if you
32:22model hurricanes at 10 km radius. And
32:25you see these blues and this and then
32:27here's 1 km and you can actually see the
32:28eddies. Like there are things that you
32:30get revealed. So it is fundamental that
32:32when you when you go to higher
32:33resolution, you can see things you
32:34couldn't see before and because of that
32:36the scientists are like, "Oh, wow, look
32:38at that." And they make some new
32:39theories based on that. So, it is the
32:40case that high resolution does can be
32:42transformation in terms of the analysis
32:44you can get of the data itself.
32:45Wonderful.
32:46Right.
32:47Yes.
32:49Boy, you want to take the other ones? Do
32:50you want to Yeah, here is another
32:51question. Um this is more of a kind of a
32:53personal question. Um
32:55We We have a limit to the amount of
32:57time, so maybe a couple of more. Um I
32:59feel that an additional four or five
33:01years in academia for PhD
33:04will be challenging to enjoying the
33:06parts of life, especially after 18 years
33:08in K through 12, uh two years community
33:11college, and now two years in Berkeley.
33:14At the same time, I'm fascinated in
33:16exploring more in ECS. How did you
33:19balance your options to pursuing another
33:22four years in academia in grad school
33:24versus going into industry immediately
33:26and be free of academia?
33:29Well, here's how I balance. I look at
33:32the fact that you'll be working
33:3540 45 years of your life once you get
33:39out of school.
33:41So,
33:43four more years, that's you know, and I
33:46do That's like that 10%. So, with 10%
33:51spending 10%
33:53longer
33:55in academia can open doors to where I'm
34:00that 40-plus years that I'm working, I'm
34:03doing what I enjoy.
34:06And And that's the part that made the
34:07difference. So, I'll give you an
34:08example.
34:10When I was at Purdue and I finished the
34:12master's, I interviewed after the
34:14master's cuz I thought I wanted to
34:16finish, okay? And And just work.
34:19The But I wanted to go to places that
34:22were pursuing research
34:24because I so enjoyed the open problem,
34:27you know, not having this one path.
34:30And
34:31every place I went, people told me the
34:33same thing. They said, "You don't want
34:35to come here
34:36um without a PhD." I was like, "Really?"
34:40They were like, I said, you know, and I
34:41thought, "But they're interviewing me."
34:44They go
34:47They told me you paid for the flight.
34:49I'm here.
34:50What do you mean you don't really want
34:51me or what? What is it? They said, "But
34:53if you if you want to actually have more
34:58input and actually steer the direction
35:01of where you're going,
35:03you want to go for the PhD if you want
35:05to do the research."
35:07And so, you know, and that's when I
35:10decided to go to grad school.
35:13Is that to continue on to grad school
35:15and get the PhD.
35:17And in grad school, I had so much fun.
35:23You too. I got a chance to interact. So,
35:25it wasn't it wasn't where I felt In grad
35:28school it's very different from
35:29undergrad.
35:31In undergrad, you you have a few
35:33electives, but you're primarily taking
35:36the courses the required courses. In
35:38grad school, you're taking the courses
35:40that are interest to you.
35:43Which makes a difference. And then
35:46you're exploring problems that are of
35:49interest to you as well.
35:51And then you have a lot of friends that
35:54are in grad school and we were doing fun
35:57things um in grad school. In Berkeley,
36:00when I started there, we would go, you
36:02know, sometimes on a Saturday hang out
36:04at the beach. We had, you know, I went
36:07to tie-dye parties cuz of course that's
36:10what you do when you're in Berkeley.
36:15But we we had a good time. We went to
36:17concerts. So, you you get, you know,
36:20there you have a friendship that forms
36:23in grad school.
36:24Where today I'm still in contact with,
36:28you know, many of the people I met in
36:29grad school because you have this bond
36:32where you're doing a lot of work, but
36:34you recognize you need to take breaks.
36:36You need a balance.
36:37And it's not where you're just working
36:3924/7.
36:41You're enjoying yourself as well, but
36:44have a focus on research.
36:47So, it's a it's a really good time in
36:49grad school.
36:51One kind of related question
36:53or
36:56opposite question, say, what motivated
36:58you to leave from your professorship and
37:01go work at Argonne Lab?
37:03Um do you still advise grad students? Is
37:06another question. You know, maybe
37:07they're potential candidates over here.
37:11Actually,
37:13um it's it's very interesting. So, when
37:14I was at Northwestern, I had a joint
37:16appointment with Argonne. And I
37:20organized my teaching schedule such that
37:23I taught on Tuesdays and Thursdays, and
37:26I was at Argonne on Monday, Wednesday,
37:28Fridays. So, I had a office there.
37:33I was interacting a great deal. It you
37:36know, I I thought for me I had the best
37:39of of both worlds of Argonne with the
37:42the community and also the resources and
37:44also teaching, doing research, grad
37:47students.
37:48And so, then I went to Texas A&M, and I
37:52missed Argonne and and that interaction.
37:55And when I had a sabbatical
37:58at Texas A&M, I actually went to Argonne
38:01for my sabbatical.
38:03And one it was what was it? Argonne
38:06invited me back to give a research talk,
38:10which I was excited to do. And um
38:14and during my research talk, you know, I
38:16was talking with different people, and
38:18Rick asked the question, you know, are
38:20you ready to return?
38:24I I hadn't thought about that. He would
38:27say, "Well,
38:28Yeah. Look at that. Mhm.
38:32Do you still Do you still mentor grad
38:34students? Do you still work with grad
38:35students as well? I I work with grad
38:37students um because for example, I work
38:40with um professor at IIT and we work
38:43with her grad students or if I work with
38:46someone at U Chicago, but it's not where
38:49I have a team faculty appointment.
38:52And you know, I could pursue it, but it
38:56was just right now, especially being a
38:58division director, I I focus on on
39:02Argonne.
39:03And and interact a lot with
39:05collaborations there.
39:07Yeah, that's a busy job being a division
39:09director. Um
39:11folks, do you realize who we've had on
39:13these calls? We have We have the lead
39:14from lead from DARPA, the division
39:16director here. I mean, this is amazing.
39:18Thank for Thank you for taking the time
39:20by the way from your busy schedule for
39:21all the things you must be doing. My
39:22goodness.
39:23That's just great.
39:24Yeah. It's a good time right now.
39:26That's true. Maybe it's Maybe it's
39:29holidays coming up. Maybe it's a little
39:30quieter kind of a thing. That's true.
39:32Yeah, that helps a little Right. And you
39:33know, in in supercomputing is this week.
39:35Oh, that's right. Yeah, the conference.
39:37So the whole the whole the whole the
39:38whole lab is going to anything. You
39:39know, it's
39:41You usually have supercomputing and then
39:43you have to have it followed by
39:44Thanksgiving so you can get that break.
39:46That makes sense. Yeah, that makes
39:47sense. Sure. Sure. Sure. Sure. Sure.
39:49Mhm.
39:52Boy, what do you think? You want to one
39:54or two more or we Yeah,
39:55Yeah, I think we are out of time.
39:58Look, maybe you'll take take me out take
40:00us out with one thought which is our
40:02students are now doing uh you know, as I
40:04mentioned before, they're doing
40:06uh SIMD parallelism, thread level
40:08parallelism, data level parallelism with
40:09MapReduce and Spark. Tell us give
40:12advice. Tell us what thoughts you have
40:13for our group of students kind of going
40:15through the stuff that you do, uh but
40:16they're touching a little bit of that.
40:17They're not really going deep into any
40:18particular They're touching here,
40:19touching here, touching there, small
40:21projects here. Uh what's the I mean,
40:23advice for student who get who likes
40:25this stuff? What's your advice for
40:27people using this stuff? What's the, you
40:29know, maybe here's an example. Is it
40:30maybe use a functional approach because
40:32a functional approach means that you
40:34don't have to worry about the order of
40:35operations, live in a functional
40:36language. What What is some advice you
40:38have for our students going through the
40:39work now?
40:41Oh, okay. Here's some general advice.
40:44So, that's what I give you, you know,
40:46because
40:47the general advice is to be curious.
40:51And to draw to try different approaches.
40:54So, for example, a lot of things that we
40:57do with different applications, we do a
40:59lot with MPI with OpenMP.
41:02Um you know,
41:04and and the OpenMP gives us the thread
41:06level on on a given node, and and we do
41:10some, you know, and you also combine it
41:11with some CUDA
41:13and MPI between, but be curious and
41:17explore. And um see what happens under
41:21different scenarios
41:23because that's the way to really get an
41:26understanding of what's going on is to
41:29try different things.
41:31And you know,
41:33it's software.
41:35And if things don't work, it's okay.
41:39It's not chemistry.
41:40That's it. Yeah.
41:47It It's okay. And it's
41:49Right. So, take that opportunity to
41:51explore things and know it's okay if
41:53things don't work. You can try different
41:56things, start off with something small
41:59that's working, get an understanding,
42:01explore different pathways, see what
42:03happens if you change some parameters
42:06around.
42:07That's great. They actually are all
42:09We're launching them to a to to project
42:12four, which is an optimization try to
42:14make this code faster and we're not
42:15telling them how to do it. So we're
42:17giving you little hints in here but like
42:18again that that's a great advice to
42:19them. Play with different things. Try
42:21this one. If that doesn't work try the
42:22other one. Maybe you can loop on roll.
42:23Maybe you can do something here. Maybe
42:25you may play with more MPN optimize your
42:27loop a little bit. Think about what that
42:28is. That's a great advice. Right. And
42:29you can do blocking, you know Right.
42:32Cash blocking. Right. You know it's so
42:35many Here's a question too. Do you also
42:38um
42:39the students get access to different
42:41profiling
42:43um tools as well.
42:45Right. That may be available and and
42:47give you, you know, more insights into
42:50what's going on and maybe hardware
42:51counters and Where where are you losing
42:53your time? Exactly right. Yeah. Yeah.
42:54Right. We haven't covered those in class
42:56but they're very very useful. They are.
43:00Wonderful. Wonderful. Well folks let's
43:02audience let's give Valerie a hand.
43:03Thank you so much for joining us
43:04Valerie. It's great to see you. Thank
43:05you for taking the time Valerie.
43:09Amazing. So great. I love these. This
43:11has been so good to see all these old
43:13friends and meet some new ones. It's
43:14just been Thank you again for the time
43:16Valerie. This is great. Absolutely love
43:18it.
43:18you for the invitation and enjoy. Yay.
43:21Thanks again to
43:23Thanks again for joining us Valerie.
43:24Perfect.
43:26All right. Take care. Bye-bye.
43:28Perfect.
43:29All right. So we are up to the next
43:32slide which is a quickie computing the
43:35news. I've got six minutes left. We've
43:36got five. Let's cover this. You all saw
43:38hopefully you saw that And that's
43:40actually there's a saw a couple of
43:41responses to a long thread about this.
43:43Apple's new hardware release of their
43:45own silicon. We knew we kind of knew
43:47that it was coming cuz they talked about
43:48it at WWDC but it actually hit and we're
43:50very excited to see some of the results
43:52that they're talking about. Um eight
43:54core GPU eight core CPU. We're learning
43:56about that this week. I mean this is a
43:58TLP week where we talk a little bit
43:59about how that works. 16 core neural
44:01engine. All these things on on chip. I
44:04saw memory on chip. I was amazed by that
44:06but if if if if you saw that. Memory is
44:08not in in Sacramento anymore. If memory
44:09is on chip, that's a whole another
44:10conversation of what happens. Right, so
44:13you got caches are on chip, but not
44:14memory. So look at what memory is going
44:16to be on chip.
44:17Memory is in the package. Memory is in
44:19the package, so I think that this is
44:21like um
44:22I I that's the biggest advantage over
44:24here. So uh
44:26It's It's not memory is not in
44:27Sacramento. If you're in Berkeley, it's
44:29like they built a freeway now to keep
44:31Costco in Richmond.
44:35But and Richmond is accessible to
44:38freeway, yeah. Yeah, makes sense. I love
44:40the analogy, but it also mean that's
44:41going to make it If someone says, "Dan,
44:42how do you make your computer run any
44:43faster?" I say, "Buy more memory." Like
44:45have that have the Sacramento trip
44:46something that you don't just have to
44:47pay all the time. That's a way to fix
44:49that. So they've certainly hit that. And
44:50here's a here's a slide. Dun da da dum.
44:53This is the fastest uh Apple computers.
44:56They looked at the whole lineup of what
44:57they have and the the three M1s I just
44:59want to say the three M1s you can see
45:01you can't see my screen. Uh the three
45:03M1s in the top three bars are all the
45:05new M1 machines. These are by the way
45:06not even the highest end machines. They
45:08normally have a higher end, you know,
45:09MacBook Pro. They have the Mac Pro, but
45:11we can't wait to see what that comes out
45:13when they put the M1 chips in there. But
45:14I want to point out the red curve, which
45:16is the fourth fastest computer on the
45:18Apple lineup
45:20is is the M1 chip running in simu-
45:23emulation mode
45:25of the MacBook Air. Do you realize that?
45:27The Core i9 is slower than the M1 chip
45:30emulating Intel
45:34unbelievable machine. So this one is a
45:36ship.
45:37fun
45:37fun to see in real world applications. I
45:39think it's you know, that running things
45:41that look uh
45:42That's some benchmark. I don't know what
45:43benchmark this is. I don't know what
45:44benchmark this Yeah yeah this is going
45:45to spend this will be this will be very
45:48interesting. I'm excited. Okay, so
45:50that's exciting. Uh let's get heavy in
45:52the first second. Let's let's take a
45:53step back. Uh what do you want to take
45:55the lead on this one? You're the one who
45:56have been mostly been working with this,
45:57but we can we can share an entire team
45:59with this too.
45:59um
46:00so unfortunately I mean all over during
46:03the semester we have been doing some
46:07code comparisons,
46:08um submission comparisons in this class,
46:11and we are finding um
46:13substantial number of cases that look uh
46:17suspicious.
46:19Uh put it this way.
46:20Um some of them are obvious.
46:24Um there is there was an obvious
46:26plagiarism
46:27uh that happened. Um so, we need to
46:30maintain the integrity of this class. Um
46:34um
46:35Nominally, what we said on our core
46:38course web page is that uh there'll be
46:41some severe penalties, and you know,
46:43what we said that in in this scenario,
46:46we are going to um you know, the in the
46:49worst-case scenario, we are going to
46:50give you an F and send you to the
46:52uh center of student conduct um to
46:56resolve the situation.
46:59Uh we don't
47:01I mean, we know that this is stressful,
47:04and these things happen, and we would
47:05like to
47:08um offer a modification to this policy,
47:12uh and we are offering um
47:16students who, you know, made a mistake
47:18to come forward
47:20and um
47:21confess. We don't want to coerce anybody
47:23into confessions. We are offering here,
47:26if you did something wrong, and you know
47:28that you've done something wrong, um
47:30you're going to have a reduced uh
47:32penalty
47:33um by admitting that you're done
47:36um something wrong.
47:38That'll save us time,
47:40um
47:41and that'll help us um a lot. But, you
47:44know, if you admit it, and we find you
47:45know, you're one of those people that we
47:47have on our list of um
47:50uh suspected
47:52um
47:54academic dishonesty, uh we will just
47:56give you 50% negative points on that
48:00on that exam or on a project. We're not
48:03going to look at the assignments this
48:04semester. There is no point of doing
48:06that.
48:07Um but you're going to take We're
48:08looking at projects and and the exams.
48:11So,
48:12uh
48:1350% uh
48:15So, instead of getting
48:17zero, we are assigning negative 50%
48:19penalty um because you could have just
48:22submitted nothing and don't put anybody
48:26um in in you know, don't giving anybody
48:28extra work to to trace that down.
48:30Um
48:32and this this will result in a
48:35non-reportable warning at CSC. That does
48:39That is not something that is visible to
48:40anybody. It is just there if the offense
48:43is
48:44gets repeated. There is a evidence of
48:46that so that this So, it's known that it
48:49is not the first time. There's more
48:52severe consequences if this gets
48:54repeated. first offense does is nothing.
48:58Um stay sealed
49:01in there. Um This is part of the
49:03Berkeley's two-strike policy, basically.
49:05So, they say everybody gets to make one
49:06mistake. And so, this this would be if
49:08you're if you're first it's just your
49:09first mistake and that information about
49:11that that academic dishonesty goes into
49:14a folder and no one can look at it.
49:15That's the important piece. There's a
49:16second slide here. So, we're going to
49:18have a a form available to fill out and
49:20we do not want to coerce anybody. So,
49:22stop listening if if you're if you've
49:24been walking a straight line. But if it
49:26is, it's giving you a chance. We will
49:27not automatically fail you on on on a
49:29project if you if you
49:31come forward with that. Um
49:33And so, I I I want to add So, I I I
49:36actually went and looked through some of
49:38these cases. Some of them are
49:41you know, unfortunately completely
49:43obvious and we can just assign an F and
49:47send this to the center of student
49:48conduct. But we're not going to, you
49:50know, at this moment we're
49:51allowing anybody to come forward and
49:54take a reduced penalty. Things happen.
49:57Now, if you don't do that, we will have
50:00to
50:02look through the code and establish
50:04facts.
50:05You know, it is not just going to be,
50:07you know, some software tool looking at
50:09things. It'll be humans that are going
50:11to determine whether there has been a
50:15a case of plagiarism.
50:17And
50:18it takes time. Um so, we have to assess
50:22at that point more severe penalties.
50:25Um they are going to in some cases, if
50:30you know, the those cases that
50:32um
50:34are obvious, we don't really need to um
50:38have a hearing. We can just forward this
50:41straight to the office of student
50:43conduct and assign an F in the class and
50:46you know, we don't have to spend too
50:47much time with that.
50:49There are some cases where we'd like to
50:50hear your opinion and we'd invite you
50:53to tell us, you know, how, you know, is
50:57this
50:58possible? You know, what are the chances
51:00that you know, this kind of coincidence
51:02happens?
51:03Um
51:05Um there are no further pleas at that
51:08time. You know, this is just basically
51:10determining
51:11what is the penalty for for plagiarism.
51:14Um
51:17There are some but we again, we don't
51:19want to curse anybody. We what we would
51:21like you to if you're not sure, if you
51:24know that you blatantly copied
51:26something, just tell us.
51:29Um
51:29If you collaborated a little bit or a
51:32little bit over collaborated on
51:34something. And again, this does not
51:36apply to exams. In exams, there is no
51:38collaboration. But on a project, it is
51:40possible that you've talked to to
51:41friends. If you did something
51:43and they helped you a little bit with
51:44the bug to debug and next thing you
51:46know, they helped you a little bit too
51:48much and now your code looks like like
51:50theirs, but it wasn't digital, but there
51:52was might be a little bit where it's too
51:53close because of how they were helping
51:54you and how they helped you fix your
51:56problem and now it looks kind of like
51:57their code and how you fixed your
51:58problem. All these are all squishy.
52:00These are all squishy here, these lines.
52:02Yeah, in in in most of these cases, it
52:06may happen that you have
52:09crossed the line, but if it is something
52:10that you're it is not clear, feel free
52:13to ask us. Don't you know, you don't
52:14have to admit anything. Ask us and we
52:17you know, we we can help you figure out
52:18whether this was right or not quite
52:20right.
52:22Not, you know, in in in in in the spirit
52:24of academic collaboration.
52:26So,
52:27this is going to go on Piazza after the
52:30the class and there'll be a link over
52:32there with
52:34Google Form to be filled out. Be
52:36specific.
52:37Yeah, and that's the that's the lot of
52:39communication. That's the that's the
52:41cleanest way cuz we've got a thousand
52:42students. If we just have email, I don't
52:43want to get lost. I just make sure
52:45nothing gets lost. So, go through that
52:46form and it'll be there on a on a sheet.
52:48We'll be able to find that, yeah. Go
52:49ahead, Bo. If you did nothing wrong,
52:51just forget about this. You're free to
52:53go. No no no no need to deal with this.
52:55Yep.
52:57As a quickie, this this is actually from
52:59the Center for Student Conduct.
53:00And so, what's what you're seeing here,
53:02I can't you nobody can see my my my
53:04cursor, but in the top right it says
53:06something suspected
53:08and then the meet with students or
53:10so the suspected does not meet on the
53:12right, it goes straight to the the
53:14basically the the disposition form it
53:16goes on to there.
53:17If you meet with students, then you can
53:19assume responsibility and do this.
53:21Otherwise,
53:22your student does not assume
53:23responsibility. So, the meet with
53:25students, this is the part that we're
53:26having the form for. This is the kind of
53:28thing this works great for a 10-person
53:29class. This does not work great for a
53:31thousand-person class
53:33to have all those meetings. So, we're
53:34putting this form to kind of basically
53:36let you say the things you would have
53:37said face to face, but you're saying it
53:38you know, in a written form that we can
53:40review and have that. So, cuz often
53:42times, nobody's ever records these
53:43face-to-face meetings and so some
53:45information is lost. So, essentially,
53:46this first top left bullet that says,
53:49"Instructor meets with students." This
53:50is the chance to have that conversation.
53:52So, whatever you would have said to us
53:53during your meeting, you're writing down
53:54on the form, and that's what we take you
53:55and we're making a decision. And in that
53:57form, you're either assuming
53:58responsibility, which you follow the
53:59left block, or do not assume
54:01responsibility, the middle one. And it
54:03might be something where you need to
54:04talk about it some more. You know, I'm
54:05not I'm I'm not doing either one. I just
54:06want to tell you what happened, but I'm
54:08not I don't know if that's actually
54:09crossing the line. That's the part
54:10that's kind of like a third box. So, I'm
54:11saying, "I still need to kind of have a
54:13conversation about that." So, there's a
54:14third option where you say, "This is
54:16what happened. I don't think I cheated.
54:17I mean, I don't think I crossed the
54:19line, but I may have." So, that's the
54:20part that's in the middle that will
54:21engage that will that a human will come
54:22in and have some conversation with you
54:24and figure out what that is.
54:26Oh, I lost you. I lost the audio. I lost
54:28audio. Please
54:30respond to Piazza privately to Piazza
54:33post if you like, "Hey, you know, this
54:35is what happened.
54:36Should fill this out." And we'll we'll
54:38we'll we'll respond.
54:40Uh but again, if you do not respond, if
54:42you don't admit any of this, you will
54:45give you until Thursday to to respond to
54:47this. If you don't, then we have to
54:50come to you
54:51with uh
54:54the evidence. In some cases, you know,
54:55we we are going to just notify you.
54:58Uh in some other cases, we're going to
55:00invite you for a conversation.
55:03All right. Anyway, it's tough stuff, but
55:04but but we don't we we know we really
55:06take this seriously.
55:07that this one it happens. It happens It
55:10happens every year.
55:12Yep, definitely.
55:14All right, let's take a breath. Let's do
55:15a quickie thing. We are scheduled really
55:17fast. We are in week 13. We have three
55:20weeks of class left. There's an RRR week
55:22and then an exam. So, we're looking at
55:23four weeks from the exam. I believe next
55:26week we'll be able to have a
55:27conversation about what the final will
55:29look like. Um but we're trying to have a
55:30final that looks like the old school.
55:32That's our spirit. We're actually going
55:34to the design session for the final to
55:35make it look at the old school exam. The
55:36old school paper exam was a 3-hour uh
55:38session where our TAs could take it in
55:40an hour and a half, but you had 3 hours
55:42total. That's what we're looking at a
55:44design space. So, we're hoping to get
55:46that kind of thing with PrairieLearn.
55:47So, we're going to be in PrairieLearn
55:48and ask the kind of questions we used to
55:50ask in the olden days. Um in which if
55:52there's any code, it's a very small line
55:54of code, very small. Probably not even
55:56asking you to code in RISC-V, probably
55:58not asking you to code C. You're You're
56:00filling in a line in a box. We're not
56:01using that. So, you're not going to be
56:03We're looking at the current design and
56:05not have you code C or code RISC-V on
56:07your own. It might be write the C code
56:09that does this or fill in three lines of
56:11this, but it will not be with the
56:12benefit of Venus. That's what we're
56:13currently looking at right now. Um
56:17And so, we'll we'll figure out where the
56:18partial credit can be for all those
56:19things, but that's the current space we
56:21have for that. Um So, this week is TLP,
56:24thread-level parallelism part two and
56:25part three. On Friday, we've got
56:27MapReduce and Spark. This is all the
56:29parallelism week. It's wonderful. Uh
56:31it's like Shark Week at the Disney
56:32Channel and or or the Discovery Channel.
56:34This is parallelism week on the 61C
56:36channel. And then next Monday is the
56:38last lecture on parallelism, data
56:40centers, and cloud computing. Then we
56:42have 2 days of a Thanksgiving recharge
56:44and rest. And I wrote here that project
56:46four is due on 12/2, which is the
56:47Wednesday after Thanksgiving. I said we
56:49moved that to give you some time after
56:51Thanksgiving to come on back and finish
56:52up what you need to do. And if you still
56:54have leftover slip days, you can use
56:56those so that you know, this in theory
56:57could even push 6 days past that.
57:00That's what we've got.
57:01Actually, three days off for
57:02Thanksgiving, actually.
57:03Yep, that's right.
57:04Friday. Correct. Three days of
57:06Thanksgiving, Wednesday, Thursday,
57:06Friday. Exactly, exactly. All right,
57:08almost done. And I think actually that
57:10was my last slide. That was my last
57:11slide, and we're already 8 minutes over.
57:13Okay, so So, I'm looking here at
57:14questions. Questions. Uh
57:17enjoyed writing my own code. Well, you
57:19might be able to write some code, but
57:21just very small, very small amounts, but
57:22we're not We're looking at our design
57:23space is not to ask you to do that. This
57:25I'm not officially declaring it yet, but
57:27our design space is to not have you to
57:29look more like the old school 61C
57:31finals, and those all those are
57:32available on HK and you should look at
57:33it.
57:34And we'll try to give partial credit
57:35whenever we can. We'll do that there.
57:37Um
57:40Uh let's see. Can we cite and use other
57:42papers and use their cash blocking
57:44method? I think that that I'd be a
57:45wonderful idea, right? Do research on
57:47how to do this better and you know, as
57:49long as you refer to it, that'd be
57:50great. I think that's a very reasonable
57:52thing to do. Um Yeah, that's a
57:54absolutely brilliant idea. Just do it.
57:56idea. Yeah, cool. Just don't copy
57:58somebody else's code because Exactly.
58:01You know, the then you know, then there
58:03is a possibility that somebody else else
58:05has found that code and then our
58:07similarities software finds all of you.
58:11Exactly.
58:12Midterm scores were delayed partly
58:13because if if you know this but many of
58:15you who took the
58:16the midterm retake happened this last
58:18weekend and there was a ton of time. We
58:20have a basically inner circle team
58:21that's working on exams and they're the
58:23folks working on the auto graders. They
58:25put together very early rough draft auto
58:27graders that was able to inform. We
58:29didn't want to have to make everybody
58:30retake the midterm. We had a list of I
58:32think two or 300 initially. We kind of
58:34pruned it down to I think it was down to
58:35less than 100 or if you remember the
58:37100-ish. Yeah, plus or minus a little
58:39bit. 90-something, yeah. Yeah, it was
58:40very nice. So we we used those that
58:42first auto grader for that, but they
58:44were all spending time on this and they
58:45all spent time on helping helping, you
58:47know, answer questions on this our inner
58:48circle team exam team. So they were
58:50helping with the exam getting making the
58:52midterm retake happen. We're just about
58:54ready to release the midterm grades. So
58:56we're sorry for all this delay. Thank
58:58you for your patience. Certainly, thank
59:00you for your patience on all this. We
59:01wanted to do it right. It's much better
59:03to wait and do it right than to do it
59:04early wrong and then there's all the
59:05flame and blah blah blah and then make
59:07the scores go down. We don't want that
59:08to happen. We want to be Here's at least
59:09the minimum minimum scores and your
59:11score won't go down. So that's the idea
59:13a goal for the midterm auto graders that
59:16we're running this stuff through. So
59:17that's really quite important and to do
59:18it right and that's why it's taking so
59:19long.
59:21Um
59:22Which There's one question, which of the
59:24assignments had these issues? Had some
59:26of the cheating issues? And I think
59:27probably we've seen cheating in all one,
59:28two, and three so far. Is that right?
59:31Yeah, we have seen them in in different
59:33places. Well, well, I mean, people will
59:36will find out. Yeah, yeah, it looks bad.
59:41And people giving me some feedback on
59:43the coding session that they're doing
59:44they they thought the midterm retake was
59:46really good. So, I'm happy to hear that.
59:47That's great.
59:50Here we go. So, ETA we're trying to get
59:51midterm scores back this week. That's
59:53our fingers crossed. Again, it's better
59:54to do be right than fast.
59:57What was the What's the three boards?
59:59You can either be
1:00:00cheap or no,
1:00:02do it right, do it No, there's three
1:00:04things you can only have two of them.
1:00:05What's the three card? What are the
1:00:06thing?
1:00:07Do it right.
1:00:09Do it cheap.
1:00:10Anyway, you can't do it right, cheap,
1:00:12and correct. And so, the idea is better
1:00:13to do it slow and correct. I forget I'll
1:00:15get I'll get it in 2 seconds after I get
1:00:16that I'll remember what it is. But,
1:00:18we're going to try to do this right and
1:00:19don't want to have to go through many
1:00:20iterations of it. So, we're really
1:00:21trying to do that. We don't want to make
1:00:23We don't want to make it fast, right,
1:00:25and cheap. You can only have two of
1:00:26those three. So, exactly. Thank you.
1:00:27Yeah.
1:00:28Yeah, I mean, we we don't want to, you
1:00:31know, we don't want redos. We want to do
1:00:33it and and be done.
1:00:36Right. And project 3B results, I don't
1:00:38know, Boar, do we have a ETA on that one
1:00:40and the prizes? That's the next thing to
1:00:42work on.
1:00:43We are done.
1:00:44Um we are done with
1:00:45were finishing up until last week. Yeah,
1:00:47we should be able to post pretty soon.
1:00:48this week. I think there was there was
1:00:51still a few people who were
1:00:53left and some second instances that
1:00:56extended it. I think my you know, maybe
1:00:58even tomorrow or something like that.
1:01:00So, there was
1:01:01other things that happened. We we are we
1:01:03are trying to release the scores this
1:01:05week.
1:01:06Good stuff.
1:01:07Again, thanks for all your patience. And
1:01:08we do hope this last three weeks isn't
1:01:10as crazy. I know you're working project
1:01:11four and one project always is the
1:01:12dominant, but hopefully you're learning
1:01:13a lot from it. This is a fun project for
1:01:15a lot of people. We will ask at the end
1:01:17what your preferred, which is your
1:01:18favorite projects to kind of rank them.
1:01:19And so, a lot of people every year, you
1:01:20know, years past say project four is
1:01:21their favorite. People who like to do
1:01:23these parallelism stuff. So, please do
1:01:24jump in on that. And we do find, by the
1:01:26way, the labs will help. Remember, labs
1:01:28are no longer required. They're out
1:01:30there. I mean, they're required in the
1:01:30sense that the material's still
1:01:31testable, but we're not going to ask you
1:01:33to just show us up for, you know, show
1:01:35up for the for the check for the
1:01:37check-off. They're called check-ins now.
1:01:39Um but we we we do find that these labs
1:01:41are really good to do before you do the
1:01:42projects. You will learn about it and do
1:01:44it in lab to do the project. So, please
1:01:46do do the labs, but again, there's no
1:01:47rush on time. Just get them done before
1:01:49the projects and you'll certainly find
1:01:50doing the projects much more fun and
1:01:51much much easier to do.
1:01:53All right, folks. Thank you so much. Bo,
1:01:55it's good to see you, as always, my
1:01:56friend. We'll see you next week. Thank
1:01:58you, Dan.
1:01:58And thanks, everybody. We'll see folks.
1:02:00Take care, everybody. Do well. Enjoy and
1:02:02enjoy We'll see you before We'll have a
1:02:03one more visit before Thanksgiving. So,
1:02:05we'll see you guys there. All right,
1:02:05thanks, folks. Thank you. Bye.
1:02:07Bye.

About this transcript

This page contains the full transcript of [CS61C FA20] Weekly Lecture 13.LIVE - TLP by CS 61C Departmental, generated from the public captions YouTube serves with the video. The transcript has 11,844 words across 1,834 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.