YT YouTube2Text

YouTube2Text → Transcripts → [CS61C FA20] Weekly Lecture 15.LIVE - Dependability

[CS61C FA20] Weekly Lecture 15.LIVE - Dependability — Transcript

by CS 61C Departmental · 13,454 words · 2,144 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00yeah all right ladies and gentlemen
0:03welcome to
0:04week 15 last monday live
0:07session 61c
0:11great to see you all here delightful
0:13delightful do know this
0:15i'm going to say it early and often like
0:17voting in chicago or
0:18early and often make sure you come on
0:20wednesday
0:21for a live session for james our awesome
0:24guest who's here is going to give a live
0:26session on gpus come here
0:27same bad time same bad channel and also
0:30make sure you come on friday for our
0:31last live lecture so
0:33this is three live live live it's as if
0:35we weren't even in a virtual world
0:37and we're just having live sessions
0:38monday was a friday at the same time
0:39okay four to five
0:40come the same channel same link we'll
0:42see you there perfect by the way borah
0:44we need to make the actual webinar let's
0:45not forget to do that
0:48we may have forgotten to do that so
0:49let's make sure we do that
0:51all right yeah also the award you know
0:53preliminary awards yes
0:55and the awards happen remember the words
0:56are gonna happen on friday so we're
0:58excited about that all right
0:59uh today's agenda we always do the
1:00agenda here uh welcome to week four
1:02fifteen where can we fourteen
1:03cross set up week can you can you fix
1:06that and post bob yeah sure
1:07week fifteen uh our guest is james percy
1:10from apple computer
1:12we're going to have a couple of computer
1:13in the news pieces we'll talk about
1:14important final exam information we're
1:16going to drop
1:16we're going to drop a mic on some of
1:18that stuff and then we'll we're going to
1:19do a review of the schedule and then
1:20they'll have to ask us anything and
1:22we'll have some time there but let's
1:23jump into our
1:24wonderful guest james so
1:27our guest is james percy from apple
1:29computer i'm going to now add you to the
1:31spotlight so you're on stage
1:33welcome james great to see you here
1:35james has been involved in computer
1:37engineering
1:38in many many roles i think since
1:40undergrad at princeton i'm thinking back
1:42to the time um he was an engineering
1:44manager you can see here at ati
1:46that he went to amd as the director of
1:48engineering uh and then
1:50joined apple uh since 2013 has been the
1:53gpu leader
1:54so he's been one of the key key like
1:56various
1:57as you get up in the top of the of the
1:59pyramid
2:00that the the group meetings become
2:02smaller and smaller and changes in
2:03smaller and smaller group meetings with
2:04more and more important people
2:06i'm sure you're meeting with tim cook
2:07every other week or something like that
2:09i'm just very
2:10crazy um but anyway it's delight james
2:12it's been great to have you we've been
2:13working together this last semester
2:15uh thinking about this and he also has
2:16been a guest at 61c
2:18uh uh in other cup a year ago so thank
2:21you again for
2:22returning to 61c uh the purpose of this
2:24is just to meet you and get to know you
2:26and talk about
2:27kind of your path and advice you have
2:28for students and who are thinking about
2:30similar compu
2:31careers or you know thinking about your
2:32time back in as a school so first of all
2:34i always start with
2:35how did you get to where you are how did
2:37you how did you even
2:39did were you all i'll just throw a
2:40couple questions out there did you
2:41always know you wanted to be what you
2:42are now
2:43and and if not how did you get to where
2:45you are now so kind of tell us your path
2:46tell us your career path
2:48sure uh i'll i'll uh try to go
2:51you can go far but go as far back as you
2:53want to you know in second grade
2:55mrs hawkins hackenhauser always told me
2:58i'd be a great engineer so go ahead
3:00anything you want to do
3:02so going back way back um i was always
3:05interested
3:05in in computers um going back as far
3:09i forget how old i was probably fourth
3:11or fifth grade um
3:12which would have been in uh in the 80s
3:14believe it or not
3:15um interested in in computers since i
3:18was a kid
3:19um we used to play computer games back
3:21then you'd have to run stuff on dos
3:23and you'd have to set it up to manage
3:25your memory and stuff like that
3:26um and so that got me very um very
3:29interested in like how do computers work
3:31and how the operating systems work
3:33um and and back in those days you really
3:35had to kind of hack your way through
3:36just to get games to run um
3:40and so that was kind of like my early
3:42interest in
3:43in computers and um that kind of turned
3:46into me becoming a computer science
3:48major
3:49in in college at princeton um and i went
3:51into princeton
3:52knowing i wanted to be in engineering um
3:54i liked math
3:56a lot like physics a lot i discovered
3:59that
3:59um that like real math meaning studying
4:02math in the university is really hard
4:04like really really hard um and despite
4:07being very good at it i was not that
4:08good at it
4:09um and so my story that's my story too i
4:12was like a superstar in high school
4:13wait this math is a lot harder than it
4:15was yeah
4:16critical math is is hard right um and so
4:19uh but i did take a
4:20a um an introduction to computer science
4:23class the freshman year of college and
4:25i'm sure these days people take that
4:27stuff in high school or in middle school
4:28or earlier sure
4:29but sometimes some does not sometimes
4:31not you know there are people who don't
4:32get it and we
4:32we're here for them we teach a class
4:34might be my beach my bjc class is
4:36exactly
4:37that space to introduce to people in
4:38college if that's their first time you
4:40know we let people into
4:43freshman year of college and i remember
4:45half the people in that class had been
4:46doing it for years and so i was
4:48i was i was felt woefully behind
4:50compared to at least half the class but
4:52um i worked really really hard at it
4:54many late nights in the labs and i
4:56realized it was
4:57it was really fun um i really enjoyed it
5:00and i was pretty good at it i was able
5:01to once i kind of got my heads around
5:03the basic constructs of like
5:05how does a c program work and what is a
5:07pointer and stuff like that and what's a
5:08function call
5:10um i really enjoyed it and and if you
5:12feel powerful
5:13like something about c like something
5:15about being low level like i can do
5:16anything i can move bits around any way
5:18i want to right
5:19different than python you're 10 miles up
5:21somehow it's below it
5:22magical right but really being low level
5:24we love that yeah and
5:26and so that turned into okay well this
5:27is cool how does how does c work and
5:29like how does assembly work
5:31and then how do you love it how do you
5:32boot an operating system and how does
5:34networking work and how do graphics
5:36pixels end up on screen so i
5:38i literally just took all every class
5:40that i could
5:41reasonably take um about computer
5:43science and i took operating systems and
5:45networking systems and computer graphics
5:47and compilers
5:48and then and learned about how assembly
5:50works and risk architectures and stuff
5:53like that
5:53um and really just absorbed as much as i
5:55could and i discovered going through
5:57that kind of sophomore junior year that
5:58i was really interested in pure graphics
6:00um and so that's what i that's kind of
6:02what i specialized in in terms of
6:04a princeton you do even if you're an
6:05engineer and you kind of do a senior
6:07project or two senior projects and so
6:09that's what i i focused my
6:11um uh my senior projects on and i did i
6:14learned about ray tracing i did some
6:16kind of advanced ray tracing algorithms
6:18um and the cool thing i liked about it
6:20it's still from the software point still
6:21from the software
6:23i mean we have you know we have cs124
6:25our graphics class but it's always
6:26software we're never getting into the
6:27hardware behind
6:28ever it's all kind of software
6:30programming and opengl and kind of
6:32application
6:32level um but what i really liked about
6:35it and i'll get to the hardware part in
6:36a minute
6:36um but i really liked about it was you
6:38drew pictures at the end
6:40and and the pictures looked really cool
6:42and for me
6:43it was it was easier for me to think
6:46about
6:46like hey here's this model and i want to
6:49i want to
6:50come up with an algorithm and it
6:51generates a picture at the end of the
6:52day it wasn't just bits moving across
6:54the network
6:55um actually kind of it's our part in a
6:57way that was right
6:58that was part of our assignments we
6:59actually had to do assignments
7:02using the algorithms that we had learned
7:03but it had to be artistic um it had this
7:05beautiful
7:05or it had to be did they give you more
7:07points for for looking cooler like
7:09that's actually i've never seen that
7:10where they actually give you that really
7:11there were more points for
7:14wow
7:16for like the best looking image or
7:19picture
7:19or yeah yeah yeah or whatever it was um
7:21so there was bonus points for being kind
7:23of creative
7:23and so i really like that too um and so
7:27my advisor at the time this guy named
7:28thomas funkhouser uh i think he's uh
7:30that's all that's my that's my buddy
7:33yeah you know
7:33you know so we we were grad students
7:35together that's my buddy
7:38seriously you're a young guy but
7:41we were like sitting together he was
7:43awesome are you kidding me i wouldn't
7:44be here if it wasn't for for
7:46professionals look at the small world
7:48look at this world he hooked me up with
7:50a guy named
7:51mark siegel who knew a guy named bob
7:53drevin uh who was working at ati at the
7:55time and said hey i know this guy james
7:57he's a cs guy
7:58he really likes graphics hey you should
7:59check it out yeah and the reason i tell
8:01you guys that is
8:02is it's it's never underestimate the
8:05value of the contacts that you're going
8:07to make at
8:08university because it can never it could
8:10be
8:12dan knows a guy who knows a guy who
8:14about a job that you might be really
8:15excited in and that
8:16that was basically what got me in into
8:18ati
8:19and i basically got a job doing
8:21verification
8:22of of graphics hardware having really no
8:25idea what that meant
8:26like all i knew was that i wanted to
8:28work on i wanted to work on graphics
8:30somehow because i was really interested
8:32in graphics um ati
8:34at the time was one of the few companies
8:36left making what we call discrete
8:38graphics cards meaning gpus or chips
8:40that are just dedicated
8:41to accelerating graphics algorithms and
8:44i was like hey that sounds pretty cool
8:45i'm really interested in this let's go
8:47go find out what what this is about
8:50um i had taken basically the
8:52prerequisites for electrical engineering
8:54at princeton but i was not an electrical
8:56engineering
8:57major and i remember showing up the
9:00first day of my job
9:01and they started telling me about like
9:03tape outs and synthesis
9:05and net lists and i was i had no idea
9:08what any of those things were i asked
9:10and asked
9:10about driving at the time hey what is
9:12all this stuff and they were probably oh
9:13my god like why did we hire this person
9:16um sorry this is the long version i'll
9:18try no that's fine no go go go go go go
9:21that's great that's good um
9:24it's very yeah i think there are a lot
9:26of people here who can relate to that
9:29yeah right in many different ways keep
9:31going so i was
9:32basically a cs software guy kind of like
9:35in the hardware organization at a
9:37hardware company
9:38um and what i brought to the table at
9:40that time was that i knew how to code
9:42the cnc plus plus and i knew
9:43graphics algorithms and so what i
9:44started out doing was
9:46i actually worked on writing kind of low
9:48level c tests
9:50um for different blocks inside the
9:52hardware and also kind of working on
9:54some kind of top-level opengl like tests
9:56um and it to be honest it isn't probably
9:59truly what we consider verification
10:01in in the honest sense 20 years later
10:04but it was low enough level that i could
10:05i could read the hardware specs and i
10:07could understand okay this is how this
10:08interface so this block works
10:10this is what the inputs and restrictions
10:11are we compare it to a c model which i
10:14also worked on and that was how we
10:16pipe vectors through and when the vector
10:18is match on the output then you're good
10:19you're block's done
10:21that's not necessarily how dv gets the
10:22verification hardware verification gets
10:24done
10:24today but for me it opened up it's a
10:27higher level it's a high level
10:28verification
10:29it isn't low level upper level
10:30verification it's higher level and it's
10:32something that
10:32with somebody with a c background and
10:34who understood how some of the math or
10:36graphics algorithms worked
10:37could could could do and so um
10:40that's how i got interested in in
10:42hardware and so i started my first
10:44couple years of ati to be honest with
10:45some
10:46my like funnest of my career because
10:49in those two years i worked on
10:50production opengl drivers
10:52i worked on emulation platforms i worked
10:54on diagnostics
10:55in the lab i worked on the c model i
10:58worked
10:58on um i worked on verification i worked
11:01with some guys at stanford out of ray
11:02tracing doing a raytracing demo on some
11:04new hardware that we had that hadn't
11:06even shipped yet
11:07um so i mean i was doing all sorts of
11:09different stuff um
11:10and so i was really just like a sponge
11:12learning about all these different
11:14things and eventually i kind of decided
11:16hey this is pretty cool but i think if
11:18i'm going to stay in this area like i
11:19really need to understand hardware more
11:21um and so i i became a manager of the
11:24verification team and at the same time i
11:26actually started taking a couple classes
11:28at stanford
11:29on kind of introduction to logic design
11:31and some of that was kind of a refresh
11:32from what i had at undergrad that hadn't
11:34looked at in a few years
11:35and somebody was kind of getting a
11:36deeper understanding of like what it
11:38takes to productize
11:39hardware which was far outside my my
11:42comfort zone
11:43um was it what was it 272 273
11:46or i think it was 272 if i remember
11:49it's been over 10 years but by name i
11:51think that was
11:53that's what it was um and so i didn't
11:55get a
11:56i didn't wasn't pursuing a degree or
11:57anything this was purely just so i could
11:59kind of get a better understanding of
12:00how
12:01how hardware worked um and i kind of
12:03parlayed that
12:04into starting to manage like uh part of
12:07the design team
12:08i started managing gpu projects at ati
12:12um and kind of learning uh that was just
12:15trial by fire you kind of learn about
12:16what does it take to actually build
12:18uh build a chip like what are all the
12:20things you have to worry about
12:21not just like hey do i have my algorithm
12:23correct but is it can i actually place
12:25and route it
12:26um can can i close timing on it
12:29can i how do i can i cool it can i cool
12:32it
12:33what are the power delivery problems
12:35what technology node what transistor mix
12:37am i going to use
12:38i mean i can go on and on about the
12:40amount of things i learned in those
12:41couple of years managing
12:42projects and learning from people who
12:44had i've been doing this for
12:46for years and then amd
12:49actually acquired ati i think back in
12:512006 i want to say 2006 or seven
12:54um and i kind of was doing the same
12:56thing um managing a larger team that i
12:59also started working what we called apus
13:01at the time meaning they were the one of
13:02the first types of socs
13:04to combine cpus and gpus on on the same
13:07on the same chip and so i was kind of
13:09responsible for the
13:10the gpu part of of that chip what's the
13:13eight what's the a stand for that is
13:15that like
13:16what does it stand for uh replication
13:20location processing unit yeah i think
13:22yeah i think that's what it was
13:24um and now of course fast forward 10 or
13:2712
13:27years later um everybody is doing these
13:29chips uh
13:30apple has been doing those chips and as
13:32and and
13:33we see there's huge energy efficiencies
13:36um
13:36from from putting gpu and cpus on the
13:39same die along with a whole bunch of
13:40other stuff
13:41um that it's like everybody's doing it
13:44but at the time it was
13:45it was pretty novel and pretty pretty
13:47exciting and very hard
13:50um and so uh so i stayed so i stayed at
13:53amd up until about 2013.
13:55um amd was having a rough time
13:57unfortunately at that that period
13:59things are much better now um and i came
14:01to apple because apple was starting to
14:03build
14:03a uh or wanted to build up gpu
14:06capabilities
14:06in in-house and so um at that time
14:10apple had already been shipping iphones
14:12with their own socs their own cpus
14:14but they were kind of expanding into
14:16into gpus and other areas as well
14:19and so i said hey this feels like a
14:21really good opportunity
14:22um and i'm gonna i'm gonna i'm gonna
14:24give it a shot there's a ton of a bunch
14:26of people that were already at apple
14:27that i already knew so i was going to a
14:29place that was somewhat comfortable
14:31um and uh i helped build up the
14:34cupertino gpu team uh from about
14:38like four or five people at a time to to
14:4070 or so people today oh
14:42wow um and actually now i i've actually
14:45just taken over responsibility for
14:47all of the what we call gpu ip
14:49development
14:50at apple so what that means is is like
14:53kind of front end rtl micro architecture
14:56uh and verification um is now
14:59uh my responsibility so now i'm managing
15:01a 200 person team across multiple sites
15:05is there engineering that's not on nada
15:07not in cupertino
15:09oh yeah there's lots of engineering
15:10outside of cupertino oh okay
15:12so in austin my boss is actually based
15:14in austin um
15:15oh and so there's uh there's a team in
15:17austin that i'm responsible for there's
15:19also gpu physical design um there's kind
15:22of architecture validation that's that
15:24i'm not responsible for that's based in
15:26austin
15:27we have a site in orlando we have a site
15:29in the uk
15:30we have um not gpu related but there's a
15:32site in in israel
15:34uh in haifa and hurstalia um so we're we
15:37are all over the place
15:41and we didn't know about the gpu until
15:43like three or four years ago right
15:46yes so the we went public
15:49or we the first time we said we shipped
15:51our own apple design gpu was with
15:54the iphone 10 in 2017 um and so my team
15:57had actually been working on kind of
15:58pieces of the gpu
16:00prior to that um but it wasn't and then
16:02and we shipped
16:03as well like we shipped the new texture
16:05unit um i forgot i forget which
16:07iphone it was but we've been shipping
16:09pieces of it for quite some time but we
16:11kind of did our first
16:12from design from scratch design uh with
16:15the with the iphone 10
16:16and and as you guys pro if you follow
16:18the news
16:19um not only are we doing iphones and
16:21ipads we're now shipping in
16:23in the max now as well so i i my
16:26daughter
16:27my wife has one upstairs but i i have
16:29one and it's it's absolutely amazing the
16:31experience of
16:32a mac with apple silicon so if you don't
16:34have one i encourage you to get one
16:37and it's still the holiday season it's
16:38still
16:40that's outstanding um i i'm curious
16:44how how how do you
16:47just adjust that i'm looking about a
16:48team of 200 people
16:50how do you how are you in meetings every
16:52minute or do you have sub
16:54groups how are you able to manage such a
16:55large group just to
16:57get together i mean and you're and
16:58you're working people above you like
17:00you're at a layer right so you're
17:01talking to people above you or calling
17:02your meetings and you have to call
17:02meetings for people below you
17:04you can't be meeting every 200 p person
17:05once a week so how are you able to
17:06manage all that
17:08yeah it's pretty hard to get on on
17:09james's schedule just yeah
17:11but folks don't students don't realize
17:13who we're getting for this semester
17:14these are
17:15these are big it's kind of a big deal
17:17i'm just saying it's kind of a big deal
17:18this is exciting
17:19that's great um thank you for your time
17:21yes i am uh i am
17:22in meetings almost uh all day every day
17:26which is which is a challenge especially
17:28early i'm not really a morning person
17:29but a lot of the folks i work with are
17:30in uk in the east coast
17:32um so in terms of managing a team like
17:35that and by the way this is a pretty
17:36recent
17:37change so i'm i'm still kind of learning
17:39how to manage this and how i want to run
17:40it
17:41um but it you really have to have a very
17:44strong
17:45staff that you trust and that you
17:47believe in and so
17:48and you can delegate tasks to that you
17:51you're confident they're gonna
17:52they're gonna get done and come to you
17:54when they have problems um that they
17:56need your
17:56need your help with and so um apple's an
18:00interesting company
18:01in the sense that um clearly given its
18:04size and the number of things that we're
18:05doing
18:06delegation has to be a thing um but
18:09you have to learn how to delegate but
18:11still be in the details of the things
18:12that matter
18:13um and so i i think people that are
18:16kind of mid to senior levels at apple
18:19have
18:19have become very good at that um clearly
18:22you can't dive into anything they're
18:24just not enough hours in the day and
18:25then the week
18:26but you have to be very good about
18:28choosing which things you want to dive
18:30deeply on
18:31um and and get involved with the
18:34important things the important things
18:35that's the skill i'm acquiring it's it's
18:38it's like it's like a performance you're
18:39going to increase performance
18:40you increase the thing where the time is
18:41being spent slowest that's the thing you
18:43increase to make the biggest performance
18:44here
18:44you spend your hours and things that are
18:46really the things that matter the most
18:47you make a decision what can people just
18:48do without you needing to touch
18:50what do you need to really get in to to
18:52work on that's great
18:54and you have to you have to kind of
18:55train your your team or your people to
18:57make sure they're bringing the things to
18:58you that matter the most and that need
19:00your input and are
19:01not bringing things to you that are
19:03they're not important and don't matter
19:04and also not
19:05hiding is probably the wrong word but
19:08you know may not be giving you the
19:09visibility of the things that you need
19:10you need visibility of sure sure uh i
19:13mean i ask a weird
19:14coverty question i'm sure you're working
19:16from home now are you actually go to
19:17apple park or how how are you
19:19how do you manage that i am working from
19:21home i have a virtual background because
19:23i know yeah
19:24because i got kicked out of my home
19:25office for my kids
19:28i've been into the office uh probably
19:31five or six times maybe once a month i'd
19:33say on average um
19:34there's not a strong need for me to go
19:36in unless there's something special
19:37needs to be taken
19:38care of and generally we're unless
19:40people want to be in the office or need
19:41to be for very specific needs i mean
19:43that's why i said hardware people i
19:44think of needing to be in the lab like
19:45whenever we see the videos of oh here's
19:47the new chip and i see this lab you got
19:48to be in that lab you can't just do that
19:50at home
19:50yeah my i'm at the kitchen table i'm
19:52trying to make the new gpu no you got to
19:54be in the lab with the sensors and
19:56you'd be amazed actually that even many
19:58of the people
19:59that work in the labs these days um
20:02actually
20:03don't need to be there in person i mean
20:04clearly we need to have a couple people
20:06in there to set them up
20:07but yeah um initially that a lot of the
20:10debug could be done
20:11remotely uh out of here that is so
20:14cool that is so very very impressive and
20:17it's actually
20:17we've had to develop that technology
20:19anyway because of the multi-sidedness
20:21that we talked about earlier so we have
20:23people
20:23we've always had people in austin
20:26debugging
20:27chips and systems that are physically
20:29located in cupertino we don't like
20:30shipping that's
20:31a pre-prototype material around
20:34people lose them lose them on a bar in
20:36san francisco remember those how many
20:37times
20:39i remember that yeah i just left under
20:40the counter sorry you know
20:42it doesn't happen so uh so we've had
20:44that capability but we've been leaning
20:46on it even
20:47even more so such that we can do a lot
20:49of remote debugging without having
20:51people
20:51being there in person wow it's
20:54impressive
20:54just for everybody's info even though
20:56research chips i mean that they're
20:58even you know clunkier uh you know when
21:01they come back i mean they're always
21:02clunky
21:03these are not products we test them
21:05mostly remotely i mean somebody goes in
21:07and
21:07you know gets it through the smoke test
21:09i mean there is a funny thing about the
21:11smoke test uh
21:12right i mean when you power it up will
21:14it pop
21:16is that what that's called i actually
21:17didn't know that was what that's called
21:18the smoke test
21:19put some power doesn't it it does
21:21actually make smoke out of it and that's
21:23bad smoke is bad i'm guessing right
21:26now just so we know yeah so
21:30once when you pass a smoke test and if
21:32you manage to boot linux on it i mean
21:33it's all remote i mean
21:35you get the terminal and you can be
21:36anywhere i mean with us it's a bit of a
21:38challenge
21:39you know running stuff through campus
21:41security i can just imagine fun is that
21:43with apple's uh
21:44security and doing these kind of things
21:47well i want a darker image no you can't
21:49have that image
21:52right yeah i guess that's true i mean
21:55you're probably i don't know
21:56is this always the case that you're in
21:57such a high security kind of a thing
21:59where your stuff is cutting edge and if
22:00that gets out that's the company
22:02or that that's a big hit like have you
22:04always been in that space or was it a
22:05big deal as you kind of moved up
22:07like to be in more like now it's iris
22:09scanners i mean like what is what's the
22:10level of security when you go in in
22:12terms of
22:12keeping this ip you know within within
22:14the walls of the company
22:16we are very uh very
22:19selective and careful about any sort of
22:22ip that gets
22:23that leaves the walls of apple um
22:26pre-production before it's been released
22:28to the world
22:30um so i mean and there's special carrier
22:33programs that are set up for that and
22:34typically only people can can carry
22:37stuff around
22:37once things have been announced but not
22:39not shipped um
22:41so so we did have to make some special
22:43provisions i'll say
22:44for dealing with what that looked like
22:47in a in a coded work from work from home
22:49world right
22:49we did let people do it i mean we
22:51weren't gonna you guys have probably
22:53seen all the stuff we've announced the
22:54the last three months that stuff would
22:56not have shipped if we if we did not let
22:58people bring it home so
22:59wow wow right it makes sense yeah that
23:02makes sense yeah
23:02you can't ask people to always come back
23:04in yeah that makes sense one things
23:06there are
23:06things are out i mean you can see um i
23:09think e
23:10times now has on a cover page just you
23:12know there'll be a series of articles of
23:13dissecting and one
23:15and one chip and they're going to pick
23:17oh wow
23:18just disassemble it yeah yeah yeah i
23:20mean no but this is you know decapping
23:22the chip
23:23and uh uh you know
23:26they said how how did they do what they
23:28did yeah how did they do what they did
23:29right yeah
23:30yeah and we're particularly sensitive
23:33about like performance benchmarks and
23:35people getting getting data and results
23:37that are not
23:38um you know not production quality so we
23:40want to
23:41in addition to just the surprise and
23:43delight we want to deliver to our
23:44customers and being
23:45delighted by the the experience we also
23:48want to make sure that we're only high
23:50quality data is is is getting out there
23:52sure
23:53sure makes sense okay we have a bunch of
23:55questions yeah we have some questions
23:56this is great thank you for all those
23:57background stuff now we just will just
23:58say
23:58you should do q a if you can so
24:00absolutely
24:02um that was just an anonymous question
24:05that says
24:05do you code or design much anymore or is
24:07it just management
24:09no they don't let me code they don't oh
24:11no it's just meetings
24:13so it's just sad oh that's hard
24:17back at amd when up up until i left i at
24:20least
24:21still like kind of had a tree and knew
24:23how to build it and run
24:24run simulations and run tests um
24:28and coming to apple when i came into
24:30apple as a management position and
24:31despite it being a small team
24:33um i was not really i never had time to
24:36really
24:37get used to the flow or really learn how
24:39things work and so
24:41kind of not having that that background
24:43of having to like
24:44you know growing up actually
24:45contributing to the code line and
24:47how projects were actually organized in
24:49the code line i never really had the
24:51chance to learn that at apple so
24:52um you know i tinker on uh a little bit
24:56on the side myself and hopefully you'll
24:57get to see that on wednesday if
25:08unfortunately sure sure wonderful this
25:10is a great question from adrian
25:12what are some advantages of creating
25:13your own gpu versus pre-made gpus like
25:16so
25:16what are they give me give me the best
25:17case like the sale most sailing thing of
25:19like what your thing just blows away a
25:20pre-made gpu
25:21versus something you can optimize for
25:23your own work workload
25:25yeah there's a there's a long answer to
25:27that and a short answer the short answer
25:29is is performance and and really
25:32performance per watt that we deliver to
25:34our customers that the the end users
25:37actually see um now there's a longer
25:40answer that that's involved that's
25:42related to like to features and
25:44optimizing for the features that we
25:45that we want and having control over
25:47those features
25:48um that is that is also true but really
25:51at the end of the day we just felt
25:53like or apple wanted to
25:56control our destiny when it came to the
25:58gpu and and really
26:00the experience that our customers had
26:02and our ability to differentiate
26:04between our competitors um really relied
26:07on us
26:07owning the the gpu and owning our own
26:10our own design
26:11and you see that in what we ship today
26:13both in the in the phone of the pad and
26:15now the mac space
26:16um if you if you compare
26:19the the macbook with the m1
26:22to what we were previously shipping uh
26:25both in the cpu space by the way cpu and
26:27gpu space
26:28it just blows it away by multiple
26:30factors so
26:31um really what we care about at the end
26:33of the day is
26:34um is the uh the performance and
26:38and the experience we deliver to our
26:40customers
26:41by the way related to that and i you
26:43know tell us if you can't answer this
26:45there are a few questions on our piazza
26:47about this unified memory architecture
26:50and 128-bit
26:51that i try to uh answer um
26:55um can you say something about that
26:57what's a unified memory architecture and
26:59what's up with the 228-bit bus
27:02um i'm not familiar or i'm not
27:06familiar with the questions around 128
27:08bit bus so i probably can't answer that
27:12the the unified memory architecture is
27:15is um is it well
27:18there's a number of things that are good
27:19about it and it was something that
27:21frankly amd
27:22was was interested in as well but
27:24typically if you look
27:26at a um cpu post gpu
27:30system that have separate memory spaces
27:32you have a lot of
27:34copies from the cpu to get data to the
27:37gpu so the
27:38gpu can use it and access it at high
27:40bandwidth and low
27:41relatively low latency but the problem
27:44is that those copies back and forth from
27:46cpu memory space to gpu memory space
27:48are expensive and can often be the
27:50bottleneck
27:51of your performance and in addition to
27:54just having performance issues there's
27:55also
27:56power cost to doing that so what you
27:58find is in a unified memory architecture
28:01where you can avoid
28:02those copies you not only get better
28:03performance you also avoid the
28:06the energy cost of copying all that data
28:08around
28:09and it's also a simpler programming
28:11model for for software to use
28:12and so it it has multiple benefits
28:16um when you look at it compared to a a
28:19non-unified memory architecture
28:21was that yeah i was pretty close
28:25i was not i'm not quite sure on
28:28the simpler programming right don't you
28:31have to now be careful about pinning
28:33things
28:33and stuff like that so that it doesn't
28:36escape
28:37uh back to the to the ssd
28:41you have to be careful about
28:44and i i i probably need to be careful
28:46about what i say here but
28:47you certainly need to be careful about
28:49race conditions let me put it that way
28:50um yeah and so the the driver in our
28:54software stack has to be
28:56aware of of avoiding
28:59creating race conditions um
29:02but the api the metal
29:06the the gpu api is pretty good about
29:09being explicit about those that was a
29:11hopefully that was a somewhat
29:12somewhat vague answer and dan actually
29:16covered some of that
29:17you know in terms of multiple cpus i
29:19think in you know last week so this is
29:21this is pretty good i think people can
29:22can
29:23can connect the dots now how does this
29:26actually work
29:27but yeah the spinning stuff is going to
29:30be
29:31yeah it cannot be swapping
29:35um okay more questions
29:38benjamin has a good one which is uh what
29:40changes in power management
29:41enabled the latest apple silicon macbook
29:44not to have a fan
29:45i mean it's just so amazing that you're
29:47able to get get all that space back
29:49and put a bigger battery and all that
29:50without a fan how did how did you do
29:52that that's an incredible
29:53i mean i've been burning my laptop my
29:55lap on apple's laptops for
29:5720 years how do they do it how is that
30:00that's incredible
30:02it it basically boils down to what i
30:04said before about energy
30:05performance efficiency and energy
30:07efficiency um and so
30:09but it's not a sacrifice it's not like
30:10well let's sacrifice it just to cool it
30:11down you're able to get
30:13better performance but yet cool like how
30:14is that that's like seems like it's
30:16mutually exclusive you can't have you
30:17know you get one one of the two
30:20great performance bad heat or low you
30:22know no heat but low performance but
30:24how do you do both that's the kind of
30:25question i think it really comes down i
30:27think it comes down to
30:29all of the techniques that apple has
30:32built up
30:32over multiple years from building
30:36low power socs for watches phones and
30:39ipads
30:39and then applying those techniques to a
30:42kind of mac
30:43class platform have enabled us to do
30:46that so it's not
30:47one single thing that you can point to
30:50that said
30:51oh apple did this and that was what
30:53allowed us to build a fanless
30:54macbook air right it's really and
30:57and this is what makes it makes it hard
30:59is that it requires a
31:01enormously large coordinated effort
31:04across
31:05across software across architecture
31:07across the design team
31:09acro with the technology team
31:12with the team that builds the um the
31:15physical cooling solution because
31:17there's still
31:17has to be some sort of dissipation thing
31:20in there even if there isn't a physical
31:21fan
31:22right um all those different team and i
31:24i'm left out many of them but all those
31:26different teams
31:27have to come to the table with their
31:30best foot forward
31:31to really optimize around power
31:33efficiency um
31:34because without doing that then you you
31:36don't get the type of products that
31:38that we get and so it's not a it's not a
31:41silver bullet where if you just go do
31:43this one
31:44architecture change boom you'll you'll
31:46be able to deliver the performance that
31:48we
31:48that we deliver at the power profiles
31:50that we deliver it's it's really
31:52a lot of effort across a lot of
31:54different teams that make it happen
31:56and i guess you probably can't answer
31:57this so i don't mean to put you on a
31:58spot but it almost feels like
32:00did it come from the top the guy tim
32:02said i want you to make a blank you know
32:04a macbook air that has no fan or
32:07did the engineering and somehow there
32:09was like a eureka moment
32:10where we started look at the numbers of
32:12cool and he's like wait we think we
32:14could do this and then it bubbled up to
32:15say hey
32:16make sure you know to advertise it
32:17because we're trying to build this crazy
32:18thing so did it come
32:19from bottom up or did it come from live
32:21the top down i don't know you just say
32:23top down and bottom up i don't know if
32:24that gives away too much about that
32:25it'd be cool if it would bottom up like
32:27oh my god some engineer at the lowest
32:28level's like
32:29i ran the numbers and right and we can
32:31do this and then it bubbled up and then
32:33people got excited about it or did it
32:34come
32:35from a visionary yeah honestly i
32:37honestly don't know
32:38right such a big deal it's such a big
32:41deal that's okay
32:42in fact i didn't even i didn't know that
32:44the the macbook air that we were going
32:46to
32:46that we just shipped was going to be
32:48fanless until
32:50until it was announced come on
32:54because the interesting thing is that's
32:56a good secret
32:57i i know everything about the chip specs
33:00and what we need to build the chips
33:02yeah but whether or not there's a and
33:04this is an interesting thing about apple
33:06but whether or not there's a fan on a
33:07particular system
33:08there's no there's no real reason for me
33:10to as the as the gp
33:12silicon guy yeah right that's all i need
33:14to know is this is the power profile
33:16that we're targeting for
33:17right like and and how much performance
33:19are you going to deliver at that at that
33:20power target and it's really up to the
33:22system team
33:23phones have no fan my phone has no fan
33:25right so i mean
33:26they're right so and it has a big old
33:28piece of plastic in the back apple you
33:30know the silicon thing that doesn't
33:31there's no way it's where's the air
33:33because you know that's a speaker
33:35where is it where's this getting cool
33:36right i mean so they're able to do it
33:37for other
33:38other pieces of hardware not necessarily
33:39yeah not bigger ones that's great
33:42running fairly low on time because we
33:45have a few other things
33:46that we need to cover i do have a couple
33:48of questions i mean there are questions
33:49that came in about the
33:51management so you're apparently you know
33:53i i i've learned that you do hire quite
33:56a few
33:56uh full-time employees and quite a few
33:58interns in your in your group
34:00obviously these people is pretty group a
34:03pretty big group needs to come from
34:04somewhere
34:05yeah and
34:08i've just learned i think jared told me
34:11that
34:12they are either gone through or are
34:15going through these internship
34:16obligations do you have any
34:18tips for interviews you know how do you
34:20interview you know what what happens in
34:22these interviews with apple
34:24yep excellent excellent question and
34:26comment um
34:29the number one thing with any interview
34:31not just apple but i
34:32i think probably for any any interview
34:34and any job is
34:35be prepared um and and so
34:39what i mean by that is is go ahead and
34:42ask your
34:43ask your recruiter or your manager
34:45hiring manager if you have access to
34:46them beforehand
34:48hey what's what's the interview going to
34:50look like um
34:51the worst thing they can say is hey
34:53we're not going to tell you anything
34:54well i don't think they would say that
34:55right um but like what types of
34:57questions should i
34:58expect what languages should i expect to
35:01code in
35:02um know your projects like the back of
35:05your hand if they're on your resume
35:07my one of my favorite questions um that
35:10i like to ask
35:11um intern and new grad candidates is
35:15tell me about the hardest project that
35:16you've ever worked on why was it hard
35:18and what did you learn from it
35:19um there's lots of there's other people
35:22on my team will ask you coding questions
35:24but i want to hear about like what kept
35:26you up
35:26in the lab for two days in a row and
35:28what you learned from that experience
35:30um the candidates that have not done
35:33well with me frankly or have some
35:35project that they did a couple years ago
35:36on their resume
35:38and they either can't speak to the
35:39details of it they don't remember or
35:41maybe their
35:42their their friend did most of the
35:44project and and they weren't very
35:46involved
35:47um that's that's not good that's not
35:49good
35:50yeah we've been telling we've been we've
35:51been telling them that for the beginning
35:53you got to know all of all projects
35:55exactly so be prepared to speak to
35:58and and talk about anything that's on
36:00your resume any language that's on your
36:02resume be
36:02prepared to code in it um any project
36:05that's on your resume
36:06be prepared to to talk about it and
36:08explain it in a lot of detail
36:11um because those are the things that
36:12that people are or want to want to hear
36:14about and frankly
36:16um it's much more interesting as an
36:18interviewer to hear about
36:20like hard projects and things that maybe
36:22didn't go that well
36:23and that were learning experiences then
36:25then just asking hey can you go
36:27code this list c program or verilog
36:30question or pick your favorite language
36:32sure there will absolutely be some of
36:33that you should be prepared but if
36:35you're a good candidate that stuff will
36:36come easily
36:37and it's the it's the projects and the
36:39experiences that will really set you
36:41apart from
36:42from other uh other other uh candidates
36:46the one the other thing i'll add to that
36:47is also if you have prior internship
36:50experience um already that that's great
36:54um but same thing be prepared to talk
36:56about it as long as you can i
36:58you know respectfully some
36:59confidentiality issues there but
37:02you know if there was something you came
37:03away from that internship that was
37:05either really challenging that you liked
37:06a lot or maybe even you didn't like you
37:08learned that hey i don't want to go
37:10be a diagnostic person in the lab for
37:12whatever reason
37:13like say why um that that really kind of
37:16shows us
37:17something about the way that you think
37:18that you have opinions about things
37:20having an opinion about things is
37:22great you don't necessarily have to
37:23agree with your interviewer but like
37:25showing that you can actually think
37:26independently and you have opinions
37:27about something and you can defend those
37:29opinions
37:30really goes a long way to show you as as
37:32somebody that has a lot of potential
37:34at apple so um so it was kind of a
37:36long-winded answer but those were some
37:38of my
37:38so my thoughts there nice do we have
37:41time for any more or is that important
37:42that's great
37:44there is another one i mean what i've
37:45learned also is
37:47our students apparently those that uh
37:50interviewed for firmware positions aced
37:53it
37:54so that those have been such you know
37:55filled taken in a week i mean that you
37:57know
37:57they sucked with you know 61c students
38:00your
38:01pa sucked out the air out of that room
38:06but you told me that very good you know
38:09the seating is pretty useful
38:12and the assembly too comes in handy
38:17but you mentioned that your path to what
38:19are you doing now
38:20came through verification we don't teach
38:22verification you know
38:23do you have a hint you know should
38:25people be scared if somebody if a
38:26verification group
38:28approaches them and says hey you know
38:29can you come over and work with us on
38:32verification
38:35they should not be scared i nobody
38:37taught me verification either
38:39in um in in college
38:42so uh i basically learned it on the job
38:44i think as as many people do
38:46um although not all there are
38:48verification classes out there at some
38:49universities right now
38:51um verification is a fantastic field to
38:54get into
38:55and one of the reasons is because good
38:57verification engineers are extremely
38:59scarce
39:00and you can use the laws of supply and
39:02demand and figure out that what that
39:04means
39:04um but if it's it's something that you
39:07are interested in and passionate about
39:09um you can really have a big big impact
39:12on
39:12on products because if you've got some
39:14critical bug you can't ship with
39:16or you have some horrible software work
39:17around um that
39:19that impacts performance like that's
39:21that's a huge problem and gets a lot of
39:23a lot of visibility
39:24in terms of what you need to know um to
39:28become a verification engineer you don't
39:29have to be a uvm
39:30guru you don't have to you don't even
39:33have to necessarily be a systematic guru
39:34at
39:36right away all you need to be good at is
39:38is coding in some
39:40reasonably modern language um and you
39:42will be able to learn
39:43uvm or system verilog or ovm or pick
39:46your other favorite verification
39:47language
39:48if you're good at c plus plus you can
39:49figure out the you'll be able to figure
39:51out the other things
39:52um what you really need is to be able to
39:55um ask
39:56ask good hard questions think through
39:58problems
39:59um challenge what you're given such that
40:02like if somebody throws a problem at you
40:04you can ask the right questions to get
40:05to the bottom of it that's what makes a
40:07really good verification engineer
40:08challenging the specs
40:10um that that you're given like hey does
40:12this really make sense
40:13um and is when we took the put these two
40:16hardware blocks together are they really
40:17going to work
40:18um it's having that sort of a uh
40:20inquiring mindset that will make a good
40:22a good verification engineer
40:26all right thank you yeah thanks
40:30thank you so much the students let's
40:31give james a hand thank you so much for
40:32your time together with us james
40:35wonderful and by the way please do
40:38come yeah on wednesday come back
40:40everybody here make sure you tell your
40:42friends come on back to this channel at
40:44four o'clock
40:45and this will be giving a live lecture
40:46on how gpus work exciting stuff and what
40:48they're used for
40:49thanks again james take care my friend
40:51that was great see you wednesday
40:53take care all right perfect okay bye
40:55right
40:57all right so this is do a quick you got
40:59about eight minutes left
41:01let's start competing in the news i
41:02found this delightful story competing in
41:03the news
41:04about uh a new app that uses computer
41:07vision
41:07and certainly a neural net trained on
41:10different different uh
41:11different models of food and clear
41:13content of those models of food
41:14um to be able to instantly take a
41:16picture of your food and people to do
41:18this when they go to fancy restaurants
41:19anyway
41:19but to actually extract the color
41:21content so you i guess you'll serve
41:23yourself a plate
41:24say i'm going to eat this plate and then
41:25you take a picture of the plate and it
41:26estimates the distance and the time the
41:27size of the
41:28scale of things and then figures out all
41:30the calories you're eating so that's
41:31really important for people who are on
41:33the diet to see the diabetes how much
41:36how much
41:36oil or fat is it right right i don't
41:39i don't i don't want to know i don't
41:41want to know and certainly you can make
41:43things with
41:43i mean like it's really hard for curries
41:45because you can make you know you can
41:47make example um
41:49uh uh saag paneer which one this is it's
41:52an indian dish with uh
41:54spinach and cream and you can put a lot
41:56of ghee and cream in it or less if so
41:58you make it kind of with water or more
41:59creamy
42:00and that little choice i don't know how
42:02a camera would do it so i
42:03i like this idea for big things but i
42:05certainly don't think it's going to be
42:06the
42:07end-all be-all because i can make 10
42:08different recipes one with like
42:10basically none of the
42:11heavy fatty stuff and one with a lot of
42:12it and it wouldn't be able to tell them
42:13because it looks like a big curry or a
42:15big
42:15sock pineapple so i mean i i don't know
42:18how how accurate this is going to be but
42:20we'll see
42:20um the second one maybe borah you could
42:23share some of these these things for the
42:24second one that i'm highlighting here
42:25oops i'm clicking on it click click oh i
42:28made it the same one ah
42:29i did it twice uh let me there's another
42:31one and i guess i lost the slide and
42:33maybe oh no here it is
42:34there it is sorry sorry i had two
42:36duplicate slides in there
42:38uh a chip delivers smarter light-powered
42:40ai
42:42and i actually visited i this is the
42:44australia's royal melbourne institute of
42:45technology this is
42:46in in melbourne and i visited them when
42:48i gave a talk when i was there in 2007 a
42:50million years ago
42:51um that's like one of the top you know
42:54issues of
42:54institution institutes of technology in
42:57australia
42:58uh their downtown melbourne but in
43:00theory they've they have an
43:01ai and a single electronic chip do you
43:03know anything about this or
43:04some of that work yeah i mean i i don't
43:07know exactly what it is but
43:08the the you guys in your projects um
43:12uh two and four you have figured out uh
43:15a lot of
43:16of stuff that is happening when running
43:20neural net difference
43:21um so generally what they call uh ai
43:25is neural net inference or machine
43:27learning reference
43:28and what you'll figure it out
43:32also is um a lot of there is either
43:36running convolutions or matrix
43:37multiplications
43:38and you guys become pretty good at that
43:41by using standard digital
43:43classified numbers wait have you seen
43:44that before
43:48you're welcome by the way you're welcome
43:51right so now people have figured out
43:53that they can do that on different
43:55hardware right you can build specialized
43:56hardware which is really good at doing
43:58major multiplications now the problem
44:00there is that hardware
44:02generally does not uh do other things
44:05you know like for loops well or whatever
44:09other things are needed uh there are
44:11other functions that that are in there
44:12so you need to have some kind of a cpu
44:14and a well integrated
44:16uh accelerator that accelerator can be
44:19made of i don't know uh
44:21you know light or some kind of a
44:27crazy memory or any anything that you
44:29know which will find out anything that
44:31runs
44:31that satisfies uh kirkham's laws and has
44:34non-linearity is able to do
44:36multiplication
44:37right i mean you you get the second
44:39order nonlinearity for free and that
44:40does
44:41you know some sort of multiplication um
44:44uh sophia xiao teaches a class uh you
44:47guys are not quite
44:48ready but you know i think you know
44:50there is a big interest in this our uh
44:52class you know
44:53ml hardware um or ai hardware
44:56and they touch on i mean you know there
44:58she has you know half of the class are
44:59gifts guest speakers that
45:01come and bring their either conventional
45:03you know really
45:04big matrix multiplication thing like a
45:06tpu that we
45:07mentioned in the class or their funky
45:10hardware
45:12and you know people who from
45:15mythic i think came and talked about
45:17their analog matrix multiplication
45:19they mythic actually just started
45:21shipping um uh
45:23right now so uh the jury's out which one
45:26of these is going to
45:27actually work there is i was told that
45:30there were 700 startups trying to make
45:32something
45:33that is going to do matrix
45:34multiplication faster wow
45:37wow 7th in the bay area
45:40[Laughter]
45:44i mean matrix multiplication put it this
45:46way uh
45:47neural net inference faster we're going
45:49to do your
45:50your project 2 and project 4 faster than
45:52you can
45:55that's great good stuff good stuff all
45:57right
45:58up to the next one this is the i'm going
46:00to try something i guess i can't do it
46:01i can't block it what's the bob here's
46:02the here's the thing the next
46:04top slide topic is final exam
46:06information we have a bomb to drop
46:08the bomb to drop is the final exam was
46:11scheduled for 12 13 a sunday all 24
46:14hours the campus
46:15sent a big memo to everybody saying that
46:18you cannot offer a final exam
46:20outside of the final exam schedule and
46:21don't make willy-nilly arbitrary changes
46:24well sorry campus this was not an
46:25arbitrary change we had this since the
46:27first day of school and we'd let you all
46:28know that and you planned for it
46:30but they said well you're sorry they
46:31said back to us sorry you have to have
46:33it on the day that it was scheduled
46:35and we have to acknowledge that request
46:37so
46:38you now have maybe this is good is bad
46:40you now have four more days
46:42to study and but instead of having your
46:44final exam on a day where there's no
46:45other exams you have your final exam on
46:47a day when there are other exams
46:49and so your final exam is now going to
46:50be on thursday december 27th
46:52start 17th 17th and you have 24 hours to
46:56do
46:56your final exam as before and we'll talk
46:58about the time limit of that but
47:00but this is that's what's on the
47:01schedule so you're like wait why do you
47:02use why is 6 20 saying one day when the
47:04my final exam schedule says when
47:06we have to honor the final exam schedule
47:08uh by the way
47:09in spring 20 we were able to move our
47:11final exams out and there was no trouble
47:12but but
47:13i think that's because it was kind of a
47:15special semester and this is a semester
47:16that's been online
47:17since the beginning and so maybe that's
47:19a little different but we we we said
47:20well we did it
47:21you know in cs 10 we give our in fact
47:23even csm we have one part of our final
47:24exam on the sunday still
47:26that sunday before the final exam week
47:27but they basically said no you have to
47:28have a final exam during the time and so
47:30we have to honor what campus said so
47:31this is not our choice
47:32this is just basically deferring to
47:34campus there is a
47:36clarification question uh here uh right
47:38i mean the the campus member said that
47:40we have to hold the final
47:42uh on the time at the on the date and in
47:45the time window when we
47:46are supposed to which is 8 to 11 a.m on
47:49thursday
47:50right um but they also said they also
47:52said they gave us
47:54yeah they also said we encourage you to
47:57have
47:57alternate times and so that's what
48:00allows us to have the 24-hour window and
48:01so they're actually strongly encouraging
48:02us to have flexibility in fact although
48:04we just had a survey
48:05borah was uh you know in our uh eek's uh
48:08faculty meeting this morning
48:09we had a survey that maybe was last week
48:12that said
48:12that students want flexibility for their
48:14exams that is something that's been
48:16shown to not you know
48:17don't just have don't just make me
48:18procter and be there exactly the time
48:20give us some flexibility for the time to
48:22be able to take a rest and all those
48:23things and so that's kind of what we're
48:24doing that's what we're doing all along
48:26so it's exactly the same format you are
48:28going to have a limited time like we had
48:30the midterm
48:31uh and in fact here's the next slide the
48:32big one
48:35we certainly want to honor and thank the
48:37student representatives for their
48:38feedback
48:39uh that's been so wonderful this last
48:41couple of weeks and they have encouraged
48:43us to think about scaling this down to
48:44be a three hour window and that was
48:46actually helpful for us to be able to
48:47now honor the campus so they kind of the
48:49two two pieces
48:51came together to both agree that it
48:52makes sense to keep this being a
48:54three-hour exam
48:55so we're going to have on the day we're
48:57going to offer exam that will take you
48:59a total of well you'll have no more than
49:00three hours total you can take it in one
49:02sitting if you want to do three hours
49:04sitting you can take it from eight to
49:05eleven so if you say hey campus
49:07yes i'm see look let's see here i'm
49:08honoring you you asked eight to eleven
49:10i'm taking it you can take all three
49:11out three spots exactly as a normal exam
49:13would be you can also do what we suggest
49:16which is
49:16take in any order take one of them take
49:18the second take the third take a break
49:20between it eat lunch between them
49:21figure them between other exams i know i
49:23i know it's not the best news for
49:25everybody because you were hoping to do
49:26it on the sunday with
49:27aside from other exams but we'd have to
49:29do this on the on the day of so we're
49:30doing
49:31on as i said the 17th um
49:34and we broke into three pieces here's
49:36the three sections there's going to be a
49:38quest which is going to be the quest
49:39clubber there'll be a question called
49:40quest clobber question
49:41a midterm clobber question which has
49:43contained all the stuff from the midterm
49:44that isn't on the quest
49:45and then everything is the final which
49:47is everything past the midterm that's it
49:48those three pieces
49:50we realized as we were thinking about
49:51the load of this is that more material
49:53happened in that midterm window that was
49:54just a ton of stuff
49:56in that midterm section a little less
49:58here you know obviously caches is big vm
50:00is big
50:00the parallels is big but we're going to
50:02put those into smaller questions so that
50:04we're going to have
50:04we're going to decide to the time to be
50:0690 minutes for the midterm clobber and
50:0860 minutes for the final climber
50:10and again overall that's three hours and
50:12we're gonna ask that our tas
50:14are able to ace those exams in in its
50:16sum of 1.5 hours so we're going to hope
50:18and we'll
50:19we'll be publishing those stats where cc
50:21was very good about collecting those
50:22stats for the midterm
50:23and then publishing that we'll try to do
50:24the same thing and share what the mean
50:25is what the max is
50:26and we're gonna and by the way if the
50:28tas can't do that in half the time
50:3015 45 and 30 we're going to cut more
50:32material we just keep cutting cutting
50:33cutting until our tas
50:35can get it down to half of that time and
50:37then we'll give you double the time
50:38that's the current plan so hopefully
50:41within the three hours yes yeah i mean
50:43these
50:44uh these are currently draft times
50:46proposed times 30 90 and 60
50:48based on how our tas do we made sweet
50:51you know we may change we may have 30
50:5380 70 or something like that that's true
50:55yeah this is the current design plan is
50:57the current this is
50:58even as it says current plan we don't
50:59think we're going to change from a
51:00three-section plan we think that's
51:01locked in
51:02but those times those three which are
51:04right here these three times are going
51:05to be
51:05may be adjusted depending on what our
51:07tas are telling us in terms of
51:08difficulty but we're still we're trying
51:09to cut
51:10that's all we're doing we're just
51:11cutting cutting cutting we have ideas
51:12and we're just like
51:13here's this here's the block of text how
51:16small can i make it
51:17you know the setup before we ask you the
51:19question to do that and we'll do that
51:20and
51:21by the way we're going to try to have
51:22this graded instantly
51:24the nice thing about grade scope and the
51:26grade scope prairie learn
51:28is that if we code the solution into you
51:31won't see it you'll do it like
51:32you'll see nothing different you'll take
51:33the exam like you normally do
51:35rather than waiting two months for you
51:37know for us to author a really good
51:39thing
51:39we're actually pre-authoring the the
51:41auto the
51:43auto grader into into prairie learn and
51:45we're testing it now in prairie learn so
51:47that when it's all done we push a button
51:48it's all graded
51:49and it's no it has no uh regrades it'll
51:52just be locked in it'll be right or
51:54wrong and we lock it in we'll have a lot
51:55of small parts
51:56if you think about this it's a i think
51:58it's 60 total points i think that's
51:59right
52:00so it's just basically a lot of small
52:01points one one one two one one one one
52:03and each one is just right or wrong and
52:04we'll do this so
52:05they won't and and by the way this is
52:07this is an important piece
52:09of the c and risk five part it's going
52:11to be a fill in the blanket thing you
52:12will not be asked
52:13to there's just no time to ask you to go
52:15to see
52:16or go to venus and code something and
52:18bring it back that's not what we're
52:19asking you to do
52:20it's more the standard kind of exam
52:22questions you've had on paper which is
52:23what is this line what does it do
52:25what's this line phil multiple choice-y
52:27thing it'll be kind of small pieces of
52:29snippety code that is you know basically
52:31has a single answer
52:32uh and it is just going to be right or
52:33wrong so that will be that'll be good no
52:35regrades for that so i can be well i had
52:36an answer that's here
52:37it'll be something that there is no
52:38other way of doing other than what what
52:40we've had that's the idea and each one
52:42will be like a one-point question so if
52:43you get it wrong
52:43it's not like it's a 20-point question
52:45you get right or wrong there's only 60
52:46points to go out and
52:47look there's a lot of questions we're
52:48going to have so there's gonna be one or
52:49one point here and it's going to be one
52:51or zero for these small pieces that's
52:52what
52:53we're trying to do we're running uh
52:54quite well over time there's a quick
52:56question
52:56um if students are taking uh
53:00past no pass and there are also
53:01questions uh i think on piazza are
53:03related to that
53:05do they have to take the final um
53:09i mean it's not a requirement
53:12um
53:16we do have total course points we have
53:17total course points that stephen has
53:18built which is which will give you a
53:20sense of your total points and by the
53:21way we can't take points away from you
53:22even if everybody gets an a
53:23we love that right so if you've earned a
53:25certain number of total course points
53:26above the passing
53:28bar that's it that we're not going to we
53:30we're not going to lower the
53:32you know raise the the limit for what a
53:33passing number is there's a passing
53:34number i forget what it is i don't have
53:35it offhand
53:36but if you're above that there are many
53:38classes yeah whatever that is
53:40if you're above that and you're past
53:41that pass you're done you don't take the
53:43final so that'll help
53:44your your own stress and your own you're
53:45on there i do want to make uh
53:48one more point about past no pass um i
53:51don't know if you guys i mean
53:52there was a lot of stress about grades
53:54and that's natural this semester but
53:56just keep in mind that our
54:00great distribution is dan mentioned
54:01early on but we have about two
54:03thirds of a class in the a a minus range
54:06when you're considering never happened
54:08before never happened before by the way
54:10never happened before the number of
54:11students who are kind of going to the
54:13finish line
54:13with basically a's and a minuses and a
54:16pluses
54:17we've never had the fraction as this
54:19semester you guys have crushed it
54:20crush the mid-term question projects go
54:22ahead board sorry so yeah yeah exactly
54:24so
54:24you know you know and and in order to
54:26get the pass you need a c minus so there
54:29is
54:29yeah unfortunately there are people who
54:30are not doing well and not in the
54:32passing range we are going to give them
54:33incompletes if they can fix this
54:35if they can you know get some part of
54:38our course um redone and
54:41done better i think you know
54:44we are definitely going to help out with
54:46that but keep in mind there is a very
54:48small number of
54:49majority of the is doing well so you
54:52know if you're getting an a
54:54just take an a don't take a b
54:57that's all right right yeah i mean you
54:59definitely look at your code of course
55:01points and see how you're done
55:02and if you've done well on the on the
55:04midterm that clobbered your quest
55:06basically you're you've got this
55:07material uh
55:09and and i think you could probably get
55:11an a if you're doing very well so don't
55:13give up the a if you
55:14don't take a p if you can have an a i
55:15bet that's if you know yeah and
55:17i've had people you know completely
55:20freaked out saying oh my
55:21course rank is like 800. you know i'm
55:24doing terribly yeah you're
55:25in between a minus and b plus i won't
55:28call it
55:29right i mean this is other classes where
55:31you've done so well as a cohort that you
55:33could be way
55:34below average but still be getting some
55:36kind of a b
55:37thing i mean again we're not taking
55:38grades but we love that we think we've
55:40done a good job if
55:41everybody gets an a not i think the
55:42midterm was just as hard as every
55:44midterm to be honest i think we're
55:45trying to do the final the same way
55:46we're not trying to do anything
55:47different so that's the space so think
55:48about that
55:50uh so let me just get some quick answers
55:52uh franklin asks is the final exam
55:54circle inside the time window
55:55it's inside the 24 hour time you have a
55:5724 hour time window on
55:59this thursday december make sure i get
56:01the date right
56:02thursday december 17th uh and it's not
56:05between 8 8 and 11 that's the official
56:07berkeley one but
56:09we're sensitive to time zones and all
56:10those things you got to make sure it
56:12works any time in the 24 hour window and
56:13you have again three sections worth
56:15roughly those times perfect uh ryota
56:17asks well
56:18alternate times those are our alternate
56:20times exactly the same
56:22exactly we give you alternate times and
56:23you make it work within your whole day
56:24that's the idea
56:25um we would be and that but there's no
56:27other day that's it there's no other
56:28like i can't take it that day that is
56:30the day sorry that's the day and we give
56:31you 24 hours to make it work within the
56:32day
56:33uh we'll be allowed to you see right now
56:35you are not right now where the idea is
56:37it's more like
56:38you will still be open notes but you
56:40won't be able to use c
56:41or or risk five it's just we'll ask you
56:44to do something and
56:45it's there and whatever you can pull up
56:47a reference material but you cannot use
56:49c
56:49like remember some of the questions
56:50before like on the float you couldn't
56:52use c as a calculator
56:53here you can't use and use c for the c
56:55questions you got a c
56:56question go do it you can't use c to try
56:58to test to see if it's going to work
56:59you've got to do it as if it were a
57:00paper exam
57:01um so there's that adam's asking about
57:04midterm retakes being graded i think we
57:06were
57:07close to having that done this week i
57:08think that's where that's our current
57:09that's our current plan
57:10it's almost
57:14these things do take forever um and and
57:17the midterm grades you saw went out and
57:18the retakes the midterm retakes are
57:20still in that loop and
57:21it's again it's the same group of people
57:23it's the same core elite group of people
57:25who are helping us write the final
57:27and helping us grade the la you know
57:28auto grade the last one so it's
57:30leaning on them a lot and they're very
57:31overworked they're way overworked this
57:33semester
57:33so they're you know they're going to
57:34when they can we'll try to get them down
57:36but definitely look we're going to push
57:37them because
57:38fry the fourth is uh pastel past
57:41deadline
57:42you they need to know that they need to
57:43get that done before the fourth so you
57:44can make an informed decision about
57:46whether you should pass or pass the
57:47score so they know that there's a very
57:48hard deadline for that and that
57:50that they're working as hard as they can
57:51to get that
57:53uh let me look at other things um
57:57total points of course points account as
58:00an a go look it online that'll see
58:02whatever that is it's on our it's on the
58:03policies page
58:05um real coding questions there may be a
58:07coding question we have to fill in a lot
58:08of blanks it could be like you know the
58:09old midterm
58:10the old final exam risk five questions
58:12were ten blanks and you're writing risk
58:14five but you don't have venus
58:16that would that might be what we're
58:17doing this semester there might be lines
58:19and lines of lines or writing c
58:20code you might be doing that you just
58:22don't have c you have to write it by
58:23hand so that's a piece of that
58:24we're going to try to do it maybe it's
58:25multiple choice maybe it's not maybe
58:27you're writing it but
58:28that's for the space and our code can
58:29actually do the grading of it
58:30automatically so that's the current way
58:32we're thinking about that again you
58:33cannot be using c or risk five
58:35outside of the exam window you've got
58:36this thing i will be very clear what you
58:38can't use and right now
58:39none of the crisp five questions have
58:40you using venus or c
58:42at all you've got to do it on paper as
58:44an old school paper thing
58:46um one comment about tcp right i mean
58:50the epa points
58:53and this extra bonus point that we're
58:56offering for filling out
58:57the questionnaire go on top of that
59:00whatever it is
59:02um oh yeah that's right yeah yeah there
59:04there there's this
59:05there's a point by the way you don't
59:06know this and it's not in the tcp but
59:08we're gonna have apa points which is an
59:10internal number i think we talked about
59:11our first lecture
59:12it's an internal number our tas are
59:14giving you for seeing you do ever
59:15participation and altruism
59:16that sprinkles on top you don't see that
59:18until the end you actually you never see
59:19that number you just have a final number
59:21that's internally what grade you get
59:22but there's a sprinkle invisible number
59:24that you get on top of that then there's
59:26also
59:26extra credit points for doing these
59:28extra performance uh
59:30projects the performance part of the
59:31projects that are probably volunteer you
59:33don't have to do it but if you do it
59:34and you you know you can get some
59:35sprinkle points there uh and on top of
59:38that if you all if ninety percent of you
59:40fill in the survey nudge nudge wink wink
59:42say no more say no more do that surveys
59:44course evaluations per pdu board party
59:47remember the url i'll take it like that
59:48course dash evaluation for everybody you
59:50should have been emailed
59:52yeah so you should all be emailed check
59:53that email look for course evaluations
59:55if 90
59:55of you by the way i need i need students
59:57to and remind their friends we're not
59:59listening to this we're not reading the
1:00:00piazza
1:00:01we're going to try to share this with
1:00:02all the piazza if you
1:00:04uh if you fill that in that's that's uh
1:00:07one more point uh for extra credit on
1:00:09top all those are on top of it
1:00:11all those are on top of tcp that you're
1:00:13getting and people asking about
1:00:15um will you know the clobber you could i
1:00:17mean if we don't get it by the clobber i
1:00:18think that steven's working to try to
1:00:19get it in but if you don't get the
1:00:20clubber you could manually compute it
1:00:21you just put it in and figure out the
1:00:22number and figure out the thing
1:00:24we'll try to have some standard
1:00:25deviations out there so you can do that
1:00:26work yourself but it'd be better if the
1:00:27software did it for you to get the
1:00:28braille number so
1:00:29hopefully that's there i'm good i don't
1:00:32see any other open questions he said he
1:00:33was going to get it done before the
1:00:34final
1:00:35yeah i mean and they need it before the
1:00:36fourth really because they need to know
1:00:38before
1:00:38for passed on past deadline okay all
1:00:41right folks
1:00:42again overall you're doing great keep up
1:00:43the great good good work uh
1:00:45we're gonna yes yes anonymous attendee
1:00:47we're gonna try to remind people to do
1:00:49that
1:00:49uh continually and by the way the only
1:00:52thing i can see
1:00:53i can right now all i can see is the
1:00:55percentage of students who are in right
1:00:56now who've done it so all i can share
1:00:58that i don't know oh bob and sue did it
1:00:59and
1:01:00ralph and jaime haven't i don't know
1:01:02that all i know is and by the way we
1:01:03don't just so you all know
1:01:05none of us see i can't see what you
1:01:07wrote um
1:01:08so be very honorable very very very
1:01:10frank it's anonymous i i can't see where
1:01:12you wrote until my grades are in
1:01:13so none of us can see what you've
1:01:15written about so you say dan was
1:01:16terrible boar was great
1:01:17whatever you say i don't care i mean i
1:01:18do care but i can't see it
1:01:20until all the grades are in and then
1:01:22it'll never affect the great so say what
1:01:23you need to say give us the feedback
1:01:25please tell us where greater
1:01:26or how we can improve put that stuff in
1:01:28and then that goes that closes
1:01:30then that closes and the final and by
1:01:32the way that closes on sunday night
1:01:33before final exam this is important
1:01:35this is another reason maybe why they
1:01:38they didn't want to have exam before
1:01:40that because
1:01:42they want to have the evaluations happen
1:01:45sunday and midnight and then exams
1:01:47happen and then grades happen
1:01:50and then we can see the results of the
1:01:51evaluations that happened like a month
1:01:52ago
1:01:53so it's kind of this weird thing uh so
1:01:55we don't get to see what you said and we
1:01:57will never like oh
1:01:58be vindictive we were angry you said we
1:01:59were bad and so we tell we can't do that
1:02:01uh whatever the grades go in and then
1:02:02we'll just get the results so please but
1:02:03you please please
1:02:05me it'd be better to give us
1:02:06improvements what can we do i know a lot
1:02:07of things worked and didn't work this
1:02:08semester we know of most of those things
1:02:10but if there are other things we have a
1:02:11huge feedback form there are other
1:02:13things i didn't get say or something
1:02:14happened since the last feedback
1:02:15please do this and i believe cc had
1:02:18another
1:02:18form of asking about prairieland
1:02:21specifically we may end up having a
1:02:22smaller survey please do that
1:02:24we are going to have to do this i think
1:02:25we're going to have another survey i
1:02:26mean that would be
1:02:27you know we're learning so much the this
1:02:30one
1:02:31the this campus survey heavily with our
1:02:34custom question i mean
1:02:36you can only ask four questions in that
1:02:37and it's not it's not enough it's not
1:02:38enough to ask all the things we want to
1:02:40know about how to improve this class
1:02:41overall so yeah so if you can give us a
1:02:43reasonable you know feedback suggestions
1:02:45we will very much
1:02:46be though thank you thank you thank you
1:02:48all right we gotta run
1:02:49thank you so much folks we'll make sure
1:02:51to share this information on piazza
1:02:53see you good luck but please do come on
1:02:54wednesday
1:02:56oh perfect please do come on wednesday
1:02:58and come on friday
1:02:59both live we'll see you live again on
1:03:01wednesday thanks everybody
1:03:03take care folks bye thank you tell your
1:03:05friends to come too
1:03:06okay bye

About this transcript

This page contains the full transcript of [CS61C FA20] Weekly Lecture 15.LIVE - Dependability by CS 61C Departmental, generated from the public captions YouTube serves with the video. The transcript has 13,454 words across 2,144 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.