YT YouTube2Text

YouTube2Text → Transcripts → [CS61C FA20] Weekly Lecture 10.LIVE - Cache

[CS61C FA20] Weekly Lecture 10.LIVE - Cache — Transcript

by CS 61C Departmental · 11,280 words · 1,867 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00great question
0:04yep perfect
0:14wonderful
0:17okay and students are dropping in now
0:20welcome everybody to week 10
0:2561 c well looks like at least some
0:29people have power
0:30yeah that's right yeah
0:34that's true i don't i i
0:37i i'm in my core
0:41office i mean this picture is fake but
0:43behind me
0:45it is fake but it's almost real almost
0:47real
0:48[Music]
0:50i'm actually i made it to campus because
0:52there is no power at my
0:53home we have a limited amount of time
0:57we have something really really really
0:58special um
1:00today so we are going to get going to
1:02that
1:03we have a cto
1:07a well-known person in in industry which
1:10is mark papermaster from mandy
1:12and it will tell us a key story how to
1:15solve the problems and have
1:17fun in the next 20 minutes
1:20and then you feel free to type some
1:22questions
1:25and and uh we have uh um two
1:28uh a corporate fellow uh mike clark and
1:31uh
1:32senior fellow tom bird who can help um
1:35field some questions afterwards um
1:39mark take it over
1:45very good uh dr nukrod thanks for having
1:48me uh join your class today and
1:50uh it will be quick i'm gonna kind of go
1:52through a little material but it
1:54uh really is about uh engineering
1:58and about i've been uh you know old guy
2:00here been
2:0135 plus years in the industry so i'm
2:03just going to jump in and just give you
2:05some
2:05thoughts and uh share with you uh you
2:08know just some of the experiences i've
2:10had
2:10so you know be a little mix of uh of
2:13things here bear with me as i
2:14just get my uh my uh bearings here i'm
2:18addressing the slides here and and zoom
2:21but look
2:22um time goes back by
2:25pretty quickly when you're having fun so
2:27the first thing i'll tell you
2:28is uh as you're in your studies and
2:30you're deciding you know what do you
2:31want to do in industry what do you want
2:32to do in
2:33academia uh is uh
2:36the biggest thing i'd say is it's it's
2:38fun to solve
2:39problems when you're solving problems 35
2:42plus years can go by
2:43pretty quickly because it really is fun
2:46and i've had a chance of those years
2:47i've been
2:48primarily at ibm uh i've given you an
2:51example of ibm and one of apple in one
2:54of
2:55the last nine years so i've been cto in
2:57running technology and engineering at
2:59amd
2:59so a few examples of how you can both
3:02work
3:02really hard and have a lot of fun so i
3:06start you know look at this photo
3:07of uh that i'm showing there it's the
3:10it's an ibm
3:11super computer as part of what's called
3:13the strategic uh
3:16accelerated strategic compute initiative
3:18where
3:19rather than blowing up atomic bombs uh
3:21you know
3:22the the intent was uh to stop that
3:24practice that was going on across the
3:26world and simulate them
3:28and so we created a
3:31uh and you know it's it's a
3:35pretty amazing uh that this was uh
3:39you know 10 teraflops of
3:42of computing so it took a lot of chips
3:44and a lot of uh
3:45systems to to create uh you know this
3:48type of uh
3:50of computing uh back then
3:53and so it uh it was a lot of fun though
3:55because
3:56uh we really had to boost the
3:58performance
4:00and um and really uh think about problem
4:03solving at scale how to how to scale out
4:05so
4:06i was really uh really brought the team
4:08together very very closely
4:10and we solved some tough problems in
4:12fact this
4:13this particular one called ascii white
4:16was the first use of copper interconnect
4:18so it was not only the designing of the
4:21computer chip but how to use new
4:22materials and that's quite common in the
4:25chip design industry you think
4:26well i can just work on one aspect of it
4:29but not true
4:30it's it's really getting you know many
4:32disciplines uh
4:33uh together and um and you know that
4:37uh after some years i at ibm i
4:41took the job running ipod and iphone at
4:44apple and you know that also showed me
4:48uh the incredible power of collaboration
4:51uh because
4:52in fact you look at the new iphone that
4:54just launched a lot of similarities to
4:55this iphone 4 when it came out
4:57it had written display it had glass
5:00front and back it had
5:01uh you know cnc uh around the you know
5:05metal band on the on the entire
5:06periphery
5:07i was the first introduction of facetime
5:09it was you know a
5:11new high-resolution camera so new new
5:13new new
5:14and a new harbor and software stack and
5:17so it really took
5:18a phenomenal uh collaboration on the
5:21goal evelyn focused on what's the end
5:22goal
5:23and working all the way through not just
5:26the internal engineering teams
5:27but just like that last example i gave
5:30you all the way through
5:31uh the supply chain and the
5:33manufacturing side
5:35and so it's one of the things i i
5:37highlight is uh you know solving
5:39uh solving tough problems really means
5:42uh you know coming together and
5:43anticipating uh really what the use case
5:46will be
5:47you know and how you you have uh you
5:50know the an end product that
5:52uh really is great and changes
5:54experience
5:55uh so iphone uh iphone 4 was certainly a
5:58great example there
6:00and uh you know really exciting program
6:03and
6:03you know brings me to uh to zen so
6:06you have tommen and mike uh
6:09on that on this call who are chief
6:11architects uh
6:12mike is our chief cpu architect at the
6:14company and uh they can tell you a lot
6:16about what we've done with zen and we
6:18couldn't be more proud with uh launching
6:19our third generation of zen
6:21uh which is uh we just announced and
6:23then we began
6:24uh shipping it in just another week and
6:26a half uh so very very exciting time for
6:28us
6:29but that was another huge challenge i
6:31mean that was um
6:32you know you've got to call a problem uh
6:36very candidly and our problem was uh we
6:38had let our single thread
6:40performance slip so the team rallied
6:42together and set the the goal
6:44of of getting amd back to performance
6:47leadership
6:48and it took you know really setting that
6:50right goal uh really defining
6:52you know what are the um you know the
6:54knobs to
6:55to get there uh and and uh you know
6:59delivering that and so we did that
7:00in 2017 with the first generation of zen
7:03uh and
7:04uh and i'll talk in a minute uh a little
7:06bit more detail about the zen three
7:08uh that we just went out but you know
7:10again each each one of these stories
7:13uh it's not where one person just went
7:15off in a corner
7:16and could just innovate and come out
7:18with something that created a great
7:20product
7:21really is about teams working together
7:24solving problems together and making a
7:27big difference
7:29and you know from our standpoint at amd
7:31we've got to stay on that
7:32uh because uh you know you look at the
7:35trend of
7:36gpu and cpu performance it's been
7:38doubling about every two plus years
7:40and that must continue the demand for
7:42more computing is insatiable
7:45and we don't get the lift on every
7:47semiconductor process node that we used
7:49to
7:49the transistors really aren't getting
7:51faster some of the circuits aren't
7:52scaling
7:53as you go node to node and frankly
7:56they're more expensive for transistors
7:58so it's a
7:59just a phenomenal time for innovation
8:01because you know anytime someone says oh
8:03because of this fact or that factor
8:05uh you know the performance gains are
8:08going to stop they're just wrong anytime
8:10you hear that
8:11you can just say look at the history
8:12there's always been innovation
8:14that keeps us driving forward uh and and
8:18providing more and more performance uh
8:20and and certainly and
8:22as far as i can see i i see
8:25there's just no end in sight as to how
8:27we're going to stay on this curve
8:30so let's take a minute and just talk
8:31about graphics rendering graphics i mean
8:33it's you know when you when you're
8:35drawing a triangle
8:36uh it turns out there's a phenomenal
8:38amount of of
8:40computing because you're you're
8:42rendering up you know
8:44billions and billions of these of these
8:48triangles
8:48uh to make incredibly um resolute images
8:52i mean the whole idea of any of us are
8:54involved in graphics development
8:56is we want it lifelike we want it we
8:58don't want you to know it was even a
8:59rendered image
9:00uh and i just show you the example of
9:03you know
9:03connecting your studies to um you know
9:07to having that impact
9:08you you you you in your curriculum you
9:11probably had to take the leaner algebra
9:12and
9:13and and a lot of the uh the math math is
9:16the language of what we do as engineers
9:18uh and you you couldn't see that more
9:20across you know all of what we do to
9:22give you that beautiful
9:24visualization that we focus on in our
9:26discrete graphics products
9:28whether anything from a rasterization to
9:31the geometries we're drawing you know
9:33rendering
9:34and right up to and including uh ray
9:36trace
9:37uh which will be i'll show you the
9:38example in just a minute of our latest
9:40in ray tracing
9:41uh incredibly lifelike uh images when
9:44you apply the math
9:46uh and and uh and you combine it with
9:49incredibly high
9:50performance computing so with that i
9:53said i'd show you a little example let
9:54me uh
9:56see if i can get this to uh to roll here
10:11[Music]
10:25so i'll stop it there but you know we
10:27worked very closely with
10:29the teams on uh on terminator and the
10:32new
10:33the newest uh edition of the terminator
10:35series
10:36and what they what was out there just
10:38couldn't get them the kind of
10:40uh you know images they want and i just
10:41show you this in the trailer because of
10:43course it
10:44it shows you how you truly break down
10:46these images
10:47but it's a game changer with the kind of
10:50cpu and gpu performance
10:52uh that we're able to deliver now and
10:53what what you've seen on movies like
10:55this
10:56uh is the whole set's changed it it's i
11:00used to be that you would film on
11:01traditional media and then you would cut
11:04and and you would then do another take
11:05and that this editing process is
11:07post-processing
11:09uh was incredibly laborious and frankly
11:11the directors
11:12you didn't always know they could even
11:14get the right scene
11:16that has just totally changed this is
11:18all done real time the processing is
11:20done real time
11:21uh the editing can be done real time uh
11:24so
11:24the kind of a performance that we've
11:26been able to bring is is
11:28uh truly uh truly changing
11:31how the media and entertainment sector
11:35is going about delivering their product
11:37really exciting
11:40so uh you know just another uh example
11:43of that and
11:44that last example didn't have ray trace
11:46and i didn't have time to show you my
11:47ray trace uh
11:49example but guess what uh because we
11:51have our
11:52our uh our technology in the new game
11:55consoles coming out
11:56uh with microsoft xbox series x and s
12:00and uh playstation five uh their
12:03pre-orders are
12:04out there and i know uh our family is
12:07a big gaming family so i've got a couple
12:09members my family
12:10uh very excited about this because it
12:13does it does
12:14have incredibly realistic imaging uh
12:17including harbor raytrace which is you
12:19know just
12:20uh it's amazing think about it the games
12:22that you played
12:24my guy i think about just eight years
12:26ago like you know uh
12:27you know the uh players would be wearing
12:30skull caps why
12:31because the physics of calculating the
12:34flow
12:35of the hair is the wind blew you know we
12:38couldn't even render it
12:39without it looking very choppy and so
12:41the game producers had
12:43you know again had those skull caps now
12:45of course we model those physics you see
12:47the hair you see flowing rogue if we
12:50see whatever on that uh that imagery
12:53that they're trying to share and now
12:54with
12:54harbor raytrace uh it's just incredibly
12:57uh realistic as the shadows and the
12:59light
13:00uh in fact truly mimic the math of
13:03real light bouncing around in our
13:04day-to-day lives very exciting
13:08uh now shift over to cpu competition i i
13:11this has
13:12been you know just near and dear to me i
13:14started you know way back on risk
13:17processors so many years ago and
13:20it's just been a phenomenal uh to watch
13:23the development and and uh
13:24and been such a pleasure work with a a
13:27brilliant uh amd
13:28uh team the last over nine years uh and
13:31so
13:32you know one of the things that i always
13:35highlight to folks is um you don't you
13:38don't get
13:38lucky in this business of high
13:40performance
13:42you you design your way to high
13:44performance you have to
13:46really spend a lot of time on the high
13:48level design if you look at this uh
13:50zen 3 that that we just uh you know
13:53announced and we just began shipping
13:55shortly here you know i would always
13:57love to just stop by the engineers
13:59office
13:59when they were back at the early
14:02ideation phase we call it the you know
14:04pre-high level design
14:05and on the white board would be idea
14:07after idea that could drive performance
14:10and of course um you know about half of
14:12them don't work
14:13but that's that's what a good design is
14:16about
14:17it's about trying things and accepting
14:19failure accepting that
14:21uh by no means will all of your
14:23approaches and all of your ideas work
14:25uh and throwing out the ones that don't
14:27work and moving on
14:28uh and you know really testing the
14:30feasibility then
14:32then when you really have analyzed that
14:34design and know you have a
14:35winner then move it in to the phase
14:37where you implement you do the logic
14:39circuit design the verification
14:41and the and the uh validation you know
14:44uh
14:45when you're done it just has to work uh
14:47the mass set with what we're at in these
14:49advanced technology notes is over
14:5115 million dollars from assets so that
14:54decision
14:55say hey i think i got it let's go build
14:57this that's a 15 million dollar decision
15:00on top on top of all the years of work
15:03that
15:03led up to completing that design so uh
15:06you know really uh you know can't
15:08emphasize enough
15:09uh the criticality of the design what it
15:12is you're studying and how to be
15:14rigorous
15:15in the application uh to come up with
15:17great designs and
15:18leadership designs
15:21uh so that is what we did with zen3 you
15:24know
15:24if we if there's time later uh uh mike
15:27and tom can uh
15:28answer a more detailed uh question chad
15:30but just give you have a sense
15:31uh you know what did we do we were
15:34coming off of
15:35uh two very impressive uh designs that
15:37drove
15:38uh with zen one a 52 improvement of
15:40instruction per clock
15:42uh then another 15 percent improvement
15:45uh with zen two and so to drive yet
15:49you know another double-digit
15:50performance improvement uh you know it
15:52really was going across
15:54you know a broad range of of levers uh
15:58really you know
15:58uh how you how you can bring memory in
16:01closer to your compute elements critical
16:03uh so um you know improving the latency
16:06was key particularly on
16:08on gaming games tend to have a dominant
16:10thread and that dominant thread
16:12uh when you can lower the latency uh
16:14just uh you know obviously it's spending
16:16less time
16:17uh waiting for the for the data and you
16:20just you
16:20see it you experience that so when
16:24we doubled uh actually the accessible
16:26all three cache
16:28uh and and so we reorganized how the
16:31cores were laid out uh that along with
16:34you know really
16:35a number of uh other elements across the
16:38load store
16:39unit improving that improving our our
16:41prefetch
16:42um you know it's we always think about
16:45it in terms of a balanced computer
16:47so as you widen out uh you know an issue
16:50rate as we did across our floating point
16:52uh and our integer engines uh we had to
16:54make sure
16:55uh that we you know not only did we
16:57bring uh data in closer with lowering
16:59latency but we wanted more efficient
17:02uh branch prediction i'm so really uh
17:05pleased with uh
17:06you know really a zero wait time in in
17:09uh we call zero bubble
17:10uh it's really improving uh how we
17:13predict uh the next construction
17:15uh and and keep a flow of data
17:17processing to our execution units
17:19uh so we're uh we're we can't let up uh
17:22it's a very
17:22competitive industry we're in uh so
17:25we're gonna we're gonna stay on this
17:26very
17:27strong uh pace of performance
17:29improvement
17:30and it's to be energy efficient uh tom
17:33bird would be
17:34uh disappointed if i didn't mention uh
17:36as well how we focus
17:38equally on the physical implementation
17:41because we're power limited in
17:42everything we do and so
17:45what we were able to do with z3 is
17:46pretty amazing because at 19 percent
17:48instruction for clock
17:50i was in the same technology node we had
17:52been in with zen 2
17:53and we did not raise the power
17:56per socket that we had so really quite
17:59an amazing
18:00amazing feat by the team uh and then uh
18:04wrap it up with my last section
18:05because i know we're just have a few
18:07minutes here but super computing
18:09i i've been involved as you saw from
18:11that from that slide i showed at the
18:13beginning of
18:14of the uh the big department of energy
18:17uh a computer
18:18uh and it it uh is
18:21incredible uh you know how we have
18:24driven what uh
18:25like my example before the teraflops
18:28that used to to fill a room
18:30uh into uh you know a single chip so
18:33this is our radeon 5700 navi design
18:37with 5.4 teraflops a peak double
18:40precision
18:40so gpu is is critical for that
18:43rasterization and that content creation
18:46that i showed you earlier
18:47and equally critical to be paired with
18:50cpu
18:51and drive incredibly dense floating
18:54point
18:55computation i mean gpu is the most
18:57flexible and easier to program
18:59a dense floating point and a vector
19:02engine
19:03in the industry and you know we need it
19:06we need it because
19:08you have a traditional super computer
19:11uh needs a high performance computing uh
19:14you know molecular dynamics and fluid
19:16flow analysis
19:17and uh you know think about uh you know
19:20the expiration of
19:21of the earth geography for uh for uh
19:24resources uh those are traditional hbc
19:27that you see on the right
19:28uh and you can see that we're gonna be
19:30breaking one exa flop next year in fact
19:33i'll talk in a moment of uh we're very
19:35proud of our role on that
19:36but what you're seeing also is a
19:38convergence of
19:40traditional super computing for uh htc
19:44with the needs of machine intelligence
19:47training
19:48it's pretty amazing when you look at
19:50what's going on with natural language
19:52generation so
19:53you know we all think resonant 50 and
19:56and you know
19:57image uh uh recognition
20:00and you know getting a correct
20:03interpretation of those images but when
20:05you look at what
20:06uh is going on with open ai and uh nlg
20:10uh it's just phenomenal what's going on
20:12with a natural energy language
20:14generation and you get the kind of
20:15accuracy you see there
20:16what they're really finding is you
20:18actually need super
20:20computer class computing
20:23uh so again i mentioned that rona pays
20:25to an exascale
20:27and uh you know our role with that is
20:30that we are the cpu and the gpu
20:32underneath
20:33what would be two the two
20:35[Music]
20:37top super computers coming out with the
20:39us department of energy
20:40so it's oak ridge national lab a one
20:42point greater than 1.5 x squat machine
20:45which will be
20:46powered up next year and then beyond
20:49that
20:49uh in in 2023 a greater than two
20:53exo-flop
20:53machine and lawrence livemore national
20:56labs not
20:57too far from where you're at so we
21:00are very proud with both our cpu and gpu
21:03to be partnering with
21:04uh with cray with hewlett packard
21:06enterprise who has acquired great
21:08in creating these incredibly efficient
21:12uh and and uh massive scale super
21:14computers
21:16um also at lawrence livermore uh
21:19uh you know labs we have already running
21:22today the corona system
21:24uh it was not named their corona uh
21:26associated with the pandemic it was
21:28ironically already named corona but we
21:31are
21:32powering these supercomputers and they
21:34are indeed uh being applied to pandemic
21:37research uh and uh in fact we've
21:40just donated uh an additional
21:4412 petaflops to help further build up
21:47this machine
21:48as well as donations we're making across
21:50other academic and uh and research
21:53institutions across the world
21:55it is a phenomenally a challenging time
21:57right now and
21:58all of us have to do what we can in
22:00industry
22:01uh we're thinking of what what you know
22:02any uh how any way we can bring
22:04computing
22:05uh to bear and to speed uh these uh type
22:08of
22:09treatment approaches which are being
22:11explored
22:12okay well again i didn't have much time
22:14so i need to rap
22:16um you know look keep a culture of
22:18innovation you play your role
22:20uh you know use things like zoom i know
22:23it's hard when you're not
22:24fully on on campus uh but i can't say
22:27enough about
22:28uh you know leaning into your studies
22:30and and uh
22:31you know really uh just for rap saying
22:34uh it's
22:35so exciting what's going on now there is
22:37no barriers
22:38uh it is all about innovation and
22:40hardware i say it's cooler than ever
22:43because it is even more about system
22:46design bringing
22:47novel uh hardware and software
22:50techniques together it's going to be
22:51heterogeneous it'll be
22:53cpus gpus specialized accelerators
22:57so there is so much challenge ahead of
22:59you and just
23:00always if just be sure you're having fun
23:03and that you're
23:03learning and you're applying new
23:04concepts and with that uh
23:06you're you're gonna have just a
23:08tremendous uh career
23:09ahead of you and whatever you choose so
23:11uh thank you so much for letting me
23:13spend uh
23:14a few minutes with you here today and i
23:15unfortunately have to run
23:17uh but uh we've got some of the team
23:18that can stay with you
23:20thanks so much thank you so much mark
23:24outstanding
23:25outstanding let's let's bring back let's
23:28bring back more into this let me do that
23:30real fast
23:31perfect i'm in somewhat uh there are a
23:34couple of questions
23:35do you have a second for questions or
23:36should we have should we have uh the
23:38others are gonna
23:39figure out the other answers no worries
23:40no worries very good thank you so much
23:42mark had a reporter meeting that's uh
23:46he's a busy guy i think you guys have a
23:50financial report tomorrow or something
23:51like that
23:53[Laughter]
23:56it's kind of important uh these days um
23:59wonderful wonderful uh there are a few
24:02questions over here uh
24:03i mean one was the directly to mark
24:07that yeah we can't answer that
24:11you guys can see these two right yeah
24:12you can see these two i think this is
24:14very close to
24:15i mean maybe we can go ahead and
24:16introduce uh mike and and
24:18and tom first um yeah you want me to go
24:22first tom is that
24:23yeah oh uh i just said on the zen three
24:25yeah once you
24:27guys i was gonna introduce myself yeah
24:30why don't you go ahead yeah mike clark i
24:33graduated from the university of
24:34illinois in
24:351993 i went to work for amd
24:39on the original k5 if anybody
24:42has seen that one it was amd's first
24:46sandy's first attempt you know we were
24:48just a second source
24:50for the intel chips through the 486 well
24:52the k5 was our first
24:54grounds up design um you know
24:58learned a lot from that obviously our
25:00first attempt
25:02you know i like to say you know learn a
25:04lot more from your failures and your
25:05successes and
25:07you know a lot of people may view k-5 as
25:09a failure but it was a huge success
25:11for the company to push us to get us
25:13going and get us to where we are today
25:16since then i've worked on every pretty
25:19much every cpu
25:20k7 k8 if you've heard of these
25:23they work on the bulldozer line and
25:26along the way i actually got my masters
25:29here
25:29at the university of texas uh very
25:32painfully one
25:33class a semester yes i was still working
25:37but uh but yeah and then you know
25:40i got that in 2003 uh now we're
25:44um i was a lead architect on zen
25:47and now i actually drive the the whole
25:49road map
25:50uh going forward um so i'll let tom
25:54introduce himself all right thanks mike
25:57um so my name is tom bird
25:59i graduated from uc berkeley i did
26:02undergrad
26:03masters and phd there so i was there
26:06boy for over three decades started at 88
26:09and left the 2001.
26:11um after tried my hand at doing some
26:14independent consulting for a few years
26:17i've been at amd now 15 years so
26:20uh when i start amd on the back end of
26:23the
26:24kate family through the bulldozer family
26:27and then of course now more recently on
26:28the zen family um
26:30so uh i can't stress it enough
26:33that as much as you kind of hear oh the
26:36world is where you know suffers is where
26:38the world's going
26:39without hardware that software is
26:41useless and hardware innovation
26:44it's you know we it's become a lot more
26:48challenging which has made it a lot more
26:51interesting
26:52and really pushes you to think outside
26:54the box
26:55um you know going forward you want to
26:56have the nice smooth cmos scaling that
26:59we skated through with through the 90s
27:01and
27:01early 2000s and it's absolutely um
27:07you know for years to come there's going
27:09to be enormous
27:10challenges in hardware design all right
27:13you want to
27:13take questions yeah i'll i have a
27:15question this is something that happened
27:17uh
27:18uh so first um we are doing uh
27:214k videos in this class and these are
27:23snippets of like
27:24you know 10 minutes or so of uh of a
27:28chunk of a lecture because we are
27:29delivering
27:30most of the stuff remotely and we are
27:32pre-recording things so dan and i got
27:34into you know this we are going to do
27:36this over a green screen green screen
27:37over power points
27:39um and all of that so we are doing this
27:41in adobe premiere
27:43um i tried doing that on the
27:45three-year-old laptop
27:46and i mean you know it gave me an
27:48estimate at six hours for the first clip
27:50or something like that
27:51i got a brand new one um apple mac
27:54with an amd radeon gpu you know the
27:58the fastest one that there is and then
28:01suddenly that that
28:02ate the time i mean these clips that are
28:0410 minutes long are done in 15 minutes
28:06or less
28:07spectacular i mean this is you know only
28:09changed my life but then
28:11adobe pushed a new update last week
28:14and they went to 20 minutes
28:18and i mean you can't keep up with this
28:20kind of stuff
28:21software bloat right it's software
28:22upload somehow
28:25you know apparently it's better and
28:26there is many more features but it is
28:28slower
28:29but there was another spectacular thing
28:30that happened last night
28:32i started rendering a video i started uh
28:35you know producing this video
28:37and the power went up after about one
28:40minute
28:41and it finished it took 30 minutes
28:43instead of 20. but it rained my battery
28:45i mean this was
28:5020 of the batteries
28:53how do i mean this is very much um in
28:56the line what tom was working on uh you
28:59know you can do a lot of things but when
29:00you're plugged in but you have to be
29:02very careful
29:03uh how these things are are are doing
29:07when they're left on their own on just
29:08the battery charge
29:10what can you say about i mean i mean uh
29:12students probably don't know what is
29:14happening in dynamic voltage and
29:15frequency scaling
29:16uh yet they might have heard about that
29:18um but it
29:20it obviously is a very big deal nowadays
29:24i mean we had the amd had a huge
29:25initiative uh
29:27my from sam napster i forget the name of
29:30it maybe
29:3125 by 20 that improved performance
29:34per watt by a factor of 25 times
29:38from 2015 to 2020 and i think the final
29:41calculation we came in at about
29:44i don't know 25.6 or something like that
29:47that
29:48you know and absolutely because of that
29:51software bloat
29:52um we're continually pushing
29:56you know the envelope on uh power
29:59efficiency
30:00and you know we'll continue doing that
30:02performance per watt
30:04right because if you know we come out
30:05with a new chip and yet you get 12 hours
30:08of battery life 13 hours battery life
30:10say hey that's great
30:11i bet you in two years of another few
30:13software updates
30:15you'd be down to you know four hours or
30:17less so it's a
30:18continual battle but absolutely
30:22you know i mean you see that in the
30:23mobile space and phones i mean in all
30:25market segments but yes absolutely
30:28it's uh you know i think market alluded
30:30to that that's uh
30:31i spent a lot of time on the zen 3 about
30:34that to make sure that we could deliver
30:35this huge performance uplift you got to
30:37do that
30:38while staying within your power budgets
30:41and there's an enormous amount of
30:42work and desire that has to go into the
30:44cpu to support that and the system
30:47you know level as well it's all levels
30:48of the hierarchy
30:52thanks uh here are some questions so uh
30:55perhaps we can go to mike what were some
30:57of the biggest challenges that had to be
31:00solved in order to develop the zen 3's
31:02cpu
31:03and one of the biggest challenges roblox
31:05that amd needs to overcome
31:06to create the next generation of chips
31:11yeah so i think i mean tom was kind of
31:13getting to it and it
31:15really i'm sure you're learning in your
31:17classes too
31:18i like to talk about you know kind of
31:21the wheel of performance
31:22you know there's ipc there's frequency
31:25uh there's area and there's power and
31:28you know a good architect for
31:29each design is trying to balance all
31:32those and then you have software on top
31:34of it with
31:35security and iso that you have to take
31:37into account as well
31:38so starting with zn3
31:41kind of what another good property we've
31:44learned along the way
31:45is that we do a grounds up design
31:49uh then you know like zen with 52 ipc
31:51then we did what we call a derivative
31:54you know along the way as architects a
31:56lot of stuff changes
31:58you know we get things wrong we see
32:00things
32:01we want to fix but you know because of
32:03the schedule they make us
32:05you know put our pencils down and we're
32:07like you know how engineers are they're
32:09always like man
32:10i just wish i could have fixed that and
32:12so we do a derivative
32:13and then we can get good amounts of uh
32:16you know performance uplift 15
32:18ipc on his m2 but for zen 3
32:21is one of those we we've kind of learned
32:24we needed to
32:25do a whole new grounds up core again if
32:27we are really going to stay on that line
32:29that mark showed you
32:30a performance and still keep the perf
32:33per watt that tom was talking about
32:35we have to re-look at the whole
32:36architecture you know
32:39come up with a grounds up core and make
32:41sure you know we
32:42effectively use uh every transistor
32:46because
32:46uh as the scaling has slowed down as
32:49time was hinting and costs are going up
32:51we have to be even more prudent and even
32:54better architects to make sure
32:57every transistor gets used uh wisely
33:00so all right thanks
33:03um here is a a question uh that was from
33:06our but
33:07you guys know this well um how do you
33:09see the future of cisc versus
33:10specialized risk processors
33:12in the world of iot uh
33:15will this be a symbiotic relationship
33:17perhaps not in the world of iot you can
33:18think about it
33:20more broadly yeah
33:24i can go first and tom can go as well
33:28i mean you know interestingly even
33:29though i've only worked on
33:31you know x86 at amd for 27 years
33:34uh you know to me isa is
33:38you know not not that big a deal
33:42as far as you know
33:45so you know we don't our machine
33:48doesn't directly execute x86
33:51instructions right
33:53we break them up into whatever
33:54operations we need we actually fuse x
33:5786 instructions together to get more
33:59work done
34:00per cycle than what was in the actual
34:02original instructions so
34:03we make new cisco instructions in the
34:05architecture we'd a couple cisco into
34:07risk
34:08so i think uh i think it is going to be
34:12symbiotic i mean i think we have um
34:15you know and it's really more about kind
34:17of
34:18the underlying micro architecture than
34:20the actual iso
34:22uh on top so i i think if you're a good
34:24architect you can design a good
34:27uh cpu in in any iso but we tell them
34:30here
34:31in in class that the x86 processors are
34:35actually
34:35risk data parts and there is other stuff
34:37that makes it appear
34:38like cisc is that true i mean you can
34:41think of it that way
34:42there is like i said we a lot of times
34:44you think of as breaking it down into
34:46simpler operations
34:48um so but you know again that's the
34:51point i mean
34:53as you know it's not really the iso
34:56it's the architecture that delivers the
34:58performance
35:00and so i think you'll see
35:03and the design point you're trying to
35:04hit right if
35:06i think that's one of the other
35:07challenges we have is trying to use this
35:09one micro architecture to go from those
35:12super computers
35:13down into that laptop and give you not
35:15make your battery run out
35:17in 30 minutes you know so it really is
35:21the same basic core
35:23across those uh dimensions you know we
35:25do
35:26cut down some things and we obviously
35:28don't put in as many cores but
35:30you know the basic root uh what we call
35:33the
35:33engine is the same in all those devices
35:42yeah i'm just going to add in that you
35:45can go
35:46look at architectural yeah
35:49good bar
35:53yeah it finished though in 30 minutes
35:56spectacularly
35:58other than the gpu what are some other
36:00hardware accelerators are being
36:01developed
36:08you wanna go for that one tom
36:12i don't know how to what extent i could
36:14answer that but let's just say
36:16we you know uh
36:19look at the ecosystem around us and
36:22i don't think that there's any things
36:25that we've announced has their
36:27mid mic that but you know
36:30uh you know you know probably
36:34at amd of course but any high-tech
36:35company there's going to be a host of
36:38of exploration r d
36:41that goes on in you know various
36:44side programs a lot of times those are
36:46you know quite interesting and
36:47challenging things to go off and look at
36:49so it's i could say it's on our radar
36:52but i probably can't go to too many more
36:53details
36:55okay um a couple of more uh questions we
36:59probably need to
37:00let you go so of what got both of you in
37:02the into the world of computer
37:04architecture
37:07yeah that's interesting i mean um
37:10my mom uh was a teacher and um
37:14we actually actually had an apple two
37:17was a they had like
37:18you know a discount program uh and so
37:22we got an apple ii computer when i was
37:24you know
37:26in grade school and just
37:29i just love that thing just playing with
37:31it but you know of course then more
37:32software but i it was just so
37:34intriguing to me that i wanted to learn
37:37how all the hardware worked
37:38inside of it and so um
37:43yeah that's a i you know i wanted to be
37:45a computer architect
37:47as a kid it's kind of weird until
37:50um and yeah i guess that's it
37:56i guess for me um i mean i there's
37:59probably only like one person that might
38:00even recognize this but i remember
38:02the zx80 microcomputer that's that's
38:06really dating me i know
38:07one of the first ones you could buy for
38:09home use and then commodore 64 after
38:11that
38:13i found like at berkeley i think
38:16like even this class i would have taken
38:18it at a different class but
38:20the one that i found most interesting
38:21was probably the closest class is cs152
38:23going off and building at that time your
38:26own spark
38:28cpu core and then going off
38:31and spending grad school at berkeley
38:35did the research project building an
38:36entire rm8 core
38:39building the infrastructure to go off
38:41and do the dynamic voltage scaling
38:43and that was a phenomenal experience you
38:47know covering all the
38:48the various disciplines um so that i
38:51could take that experience with me
38:52whether it's cad whether it's physical
38:54design
38:54whether it's rtl architecture and the
38:57more
38:58breadth that you get while you're in
39:00school the
39:02higher value you will be in hardware
39:04design down the road
39:05because as if you've already heard mark
39:07you know hound on that it's
39:08there's so much cross team collaboration
39:11that
39:12you know these huge complicated ships
39:14are a team of
39:15uh hundreds of people and the more that
39:18you can talk to
39:19people outside your domain and your
39:21specific expertise the more effective
39:23you can be to move the whole train
39:25forward so you know on a day-to-day
39:27basis me and
39:28mike do pretty different things but
39:31there's
39:32enough overlap in terms of our overall
39:34long-term road map
39:36that we've got semi-regular discussions
39:38and forums and
39:39that makes it you know the very
39:41interesting and exciting
39:42to keep working in this area i met tom
39:46when he was a grad student i just
39:48thought he had a junior faculty one
39:49secret
39:50he was the i think my first
39:53[Music]
39:54dissertation that they signed off as a
39:56second reader uh tom
39:58bob brotherson was tom's uh tom's
40:00advisor
40:02but tom was a one-man michael proster
40:04show uh uh i mean he designed him
40:06you know with a very little help uh
40:09micropascal from scratch and made it
40:11scale its performance with supply
40:13voltage which was the first one of the
40:15time
40:15uh quickly just uh why does
40:18amd use partners like psnc to
40:22manufacture chips instead of making them
40:24in-house i think you went through a
40:26painful experience of doing that
40:31um you want to take that one mike drove
40:33me too no i can yeah
40:35yeah so i mean yeah i mean but
40:38fabricate fabs are really really
40:41expensive and well
40:43we did have a ceo who used to say real
40:45men have fabs
40:46we uh he he doesn't say that anymore
40:52but um yeah i mean when you you you
40:55really
40:56you really either because they keep
40:57costing so much um
40:59either need to continue to
41:02develop other types of chips to keep
41:05filling them
41:07and and sucking in more you know
41:10things to fill that fab that it's a
41:13pretty tough business model whereas
41:15using tsmc and others who can take
41:19multiple customers to leverage the cost
41:22of that fabrication facility
41:25looks is you know a better economic
41:28model
41:29for us all right one final question i'll
41:33let you go
41:34uh what role has grad school played in
41:36your career development and what
41:38some some thoughts about doing grad
41:40school right away after
41:41uh undergrad versus doing it from
41:43industry or after
41:44uh spending a few years in industry
41:49yeah i can go first yeah i mean to be
41:51honest
41:53i was broke and i i always wanted to go
41:57to grad school
41:58and but i really needed to get a real
42:01job and
42:02start paying down my student loan debt
42:04um
42:06so yeah but then it was a wonderful
42:07opportunity amd here they had a program
42:10where they would pay for
42:12you to go back to school and so
42:15i put in the effort to put in the you
42:17know nights and weekends and we have
42:19flexible schedules here to be able to
42:22take just one class spread it out over
42:24six years one class a semester
42:26uh but yeah no i think i believe in
42:30you know continual learning uh the
42:33collaboration you get in that
42:34environment
42:35uh i even i was 2003 when i graduated i
42:39took a machine learning class if you can
42:41believe it and
42:43um you know so i
42:46new ideas uh come from they're all old
42:49ideas that get better
42:50right as well as just totally new ideas
42:52and just trying to uh
42:54i think even if you go to grad school
42:56you need to continue to keep
42:57this industry move so fast you just have
42:59to continue to keep trying to
43:01uh grow your learning and knowledge
43:08you just kept going yeah yeah i would
43:11say i would say keep going
43:13but it's i would advise
43:16doing internships to get some experience
43:18in industry to help you
43:20understand what you'd want to focus on
43:21in grad school but like i couldn't
43:24imagine i actually knew some you know
43:26somebody we work with uh you know went
43:28off for his phd
43:29and he was playing his mid 40s and i
43:31don't know if i could mentally you know
43:33handle that
43:34trying to do that is in my mid 40s so i
43:38think if you're going to go through phd
43:39i would recommend doing that sooner
43:41rather than later
43:43um you know masters perhaps a little
43:45more you know flexible but
43:47definitely i wouldn't i'd highly advise
43:49to go
43:50do the internships and that's to help
43:52you get some exposure to
43:54the industry of that do you guys hire
43:56interns
43:57uh can you send me a collection of links
43:59if you do do you hire people with
44:01bachelor's degrees
44:03we have a link of those we do i'll i
44:06i'll have to send you some links after
44:08so you can post for the class
44:10that would be spectacular we have we
44:13have a few
44:14um and somebody just said yeah yeah we
44:17have 1200 people
44:19100 people i'm sure there's a couple of
44:20them there who we'd love to have it
44:22thank you very much thank you so much
44:25folks outstanding
44:26wonderful wonderful to have you here
44:29let's give a hand folks thank you thank
44:31you
44:31thank you glad we could help out that's
44:34great mark did a great job his
44:35presentation as always
44:37that was outstanding that was very good
44:39that was great all right
44:40thanks for watching there you go
44:43take care take care all right all right
44:46folks
44:47let's uh continue on have some fun uh
44:50i've got the next slide which is a do a
44:53quickie see if we can do this one minute
44:54a slide computing the news
44:57robot cars without drivers are coming to
44:59san francisco i think we introduced this
45:01that was coming to arizona before
45:03and they're now coming to san francisco
45:05that's very exciting
45:06uh they're going to start neighborhood
45:07by neighborhood and try to socialize the
45:10idea that you know cars are going to be
45:12coming around without a driver don't
45:13freak out
45:14the problem is what if they socialize in
45:16this neighborhood but the person who
45:17gets in the car wants to go to that
45:18neighborhood and they haven't socialized
45:20yet
45:20so what are they going to do not drop
45:21them off or something but anyway we'll
45:23see what happens this is a gm product
45:25and we'll see what that works but i'm
45:26very i'm sure this is going to go viral
45:27if you know to be able to first
45:28there's going to be somebody who gets in
45:30the car says and then takes a big
45:32picture of the movie and that goes viral
45:33so we'll see how it works but
45:34fingers crossed for safety and fingers
45:36crossed for the success of that that is
45:38the future and it's here now it's here
45:39in our city very exciting
45:41all right uh i don't know if you can all
45:43see this in
45:44if you zoom into this maybe i should
45:46zoom in zoom in a little bit so i can
45:47see it better as well
45:49we've got uh you all many of you are
45:51taking the midterm
45:52uh i was going to pull up the actual
45:54numbers i had this in another window of
45:56the actual numbers people who were
45:57where they were um we
46:00are in still in midterm season so we're
46:03still we're still uh processing that
46:06many of you have lost power
46:08so we're gonna announce this in the next
46:10slide we've already announced it but
46:11we are uh we have certainly extended the
46:14exam
46:15from sunday and monday to be sunday
46:17monday tuesday
46:19so that's exciting uh so just so you
46:21know that that's what that's where we
46:22are
46:23we're in the middle of a week or the
46:24beginning of week 10 um
46:26the today's lecture is caches part three
46:30and wednesdays is caches part four the
46:33upcoming we also have a lab this week
46:35lab seven we're gonna talk about that
46:36labs are no longer
46:39required for points you attend the
46:41session to get those points
46:42but they uh you don't have to do the lab
46:45so if you can do this if you do this lab
46:46this week great
46:47if you need more time it's no problem
46:48you won't lose points with that just
46:49show up for your check-in session
46:51it's no longer a check-off it's a
46:52check-in we have a whole slide talking
46:54about that
46:57then on friday is os and vm intro and
46:59then next week we start a whole new
47:01topic well it's still related but it's
47:02part of the memory hierarchy
47:03of virtual memory one and two and io is
47:06on friday
47:08so what's still pending i believe i
47:10showed here is uh homework seven is due
47:1211 9 and project 3b people are still
47:15working on so we're going to talk about
47:16that but that's just the schedule
47:18so you kind of know where we are but
47:19there's nothing officially new assigned
47:21here it's just going to roll over
47:21assignments from
47:22before last week's last week's homework
47:247 and lab 3b
47:25and hope project 3b all right if you if
47:28you did not notice if
47:30we we moved the report to that due date
47:34correct okay so let's now pull to the
47:37next
47:38slide here um this is the update
47:41and i'm going to pull the scores up or
47:43at least the numbers here here we go
47:44i've got the numbers here out of i see
47:47about basically 1200 because for the
47:49quest
47:49section 112 1190 took it so we'll just
47:52say 1200 for
47:53for the sake of it so out of 1200 who
47:55took the quest
47:56i've got about 900 who took section 1
48:00600 who took section 2 720 you took
48:03section 3
48:04680 you took section 4 and 58 section 5.
48:07so roughly if we talk about 1200
48:09about half of you took sections two
48:12four and five a little bit more check
48:15section three
48:16and 900 out of 1200 so maybe three
48:18quarters ish
48:19took uh section one so people are still
48:21taking it but
48:22we do know that there are outages and we
48:24do you know wish
48:25safety and uh for everybody around or
48:28safety for all of us certainly that we
48:29hope
48:29no fires spark up because of the high
48:31winds this is a high wind event as they
48:33say it's the
48:33it's the strongest wind event all year
48:35so pg e and a preponderance of caution
48:38uh turn power off for many people so
48:40that the you know things don't
48:42brush into their lines and cause a fire
48:44as this happened before
48:45so here's the idea my house is without
48:47power i'm i'm presently in corey hall
48:50uh because uh i have no internet no cell
48:53coverage and no power right and bo this
48:55is new information that they're actually
48:57the cell towers
48:58i thought the cell towers were battery
48:59powered or some other power so that even
49:00turned power up you wouldn't lose a sell
49:02signal
49:02that kind of worries me for what power
49:05goes out during a big earthquake or
49:06something
49:06to have no cell phone i just think of
49:08cell phones or something that
49:09automatically always happens
49:10but that's that's that's a significant
49:12thing for safety
49:17[Music]
49:19okay okay um so anyway the examination
49:22is tuesday for everybody
49:24so this is it for everybody we'd love
49:25you to take it earlier so if you still
49:27have power take it now don't take it
49:28tuesday
49:29and there's partly a reason for that and
49:30i'll just read this to you if you didn't
49:32take it by then
49:33um you could just by the way if you
49:34didn't take it by the end of tuesday you
49:36say oh something happened something
49:37happened something happened
49:38we'll just put you in the pool of
49:39midterm retake students and that's going
49:41to be
49:4111 7 11 8 that's going to be after the
49:44election so we're not making you do
49:45anything between now and election
49:46because
49:46of so many things are going on we're
49:47going to send out a form for you and
49:48figure that out
49:50um we are going to have we had to now
49:52scramble and get ta proctoring for
49:53people who had the exam on tuesday
49:55because we didn't plan for
49:56exam a tuesday exam day um and
49:59uh we're still working on that schedule
50:01and trying to have people put pull extra
50:02extra hours but you know we're really
50:04stretching our staff thin we were going
50:05to have them
50:06all signed up to do this and now you
50:08still have somebody manning the store on
50:10monday
50:11but now they have to man the store on
50:12tuesday also so like they're going to be
50:14half
50:14it's going to be their workload is less
50:16on monday and and now they're stretching
50:18into two two days we'd rather just be
50:20you work a certain surface you answer
50:22questions as fast as you can now the
50:23questions are going to come much slower
50:24because
50:25fewer people take you out of monday and
50:26now they have to also work double time
50:28so it's kind of a shame that we have to
50:29push
50:29our staff to do this but we're working
50:31with you and we're working with them and
50:32they're they're also dealing with the
50:33same thing they're also students
50:34themselves
50:36so please take it today if you can if
50:38you don't have any power issues take it
50:40today because we don't have a lot of
50:41people tomorrow just letting you know
50:43that
50:43and piazza is going to remain closed for
50:45tuesday okay so that's awesome so if you
50:47finished your exam
50:48you want to work on the thing and ask
50:49questions you could ask a private
50:50question but we're focusing and
50:52prioritizing all the questions for the
50:54exam just so you don't know
50:55for one more day just one more day we
50:57can't slip anymore
50:58project 3b we have extended it to
51:0211 1 so this is monday 11 1 i think 11
51:062 is the tuesday or is 11 3 maybe it's
51:07sunday let me make sure i get the date
51:09right
51:09uh sunday 11 1 okay we're get that's two
51:12extra days which means we only actually
51:14moved
51:15we only actually stole one day for you
51:16in terms of piazza being closed and all
51:18this but we actually give you two extra
51:19days so that
51:20accompanies you know that is able to
51:22react to maybe you need two days of
51:24extra so we're actually giving two days
51:25because people may have actually lost
51:26power for two days
51:27piazza is going to be closed tuesday so
51:29you can't ask questions
51:30but we're going to oh this is very
51:31important we're going to open up a
51:33project 3b
51:34text channel on discord but
51:37tas are prioritizing midterm monitoring
51:40over project support on tuesday because
51:41that's really the most important thing
51:42that's a much higher stakes you know
51:44it's like a huge chunk of your total
51:46cost points to the exam how much
51:48a smaller chunk is project 3b so we need
51:50to focus on that part of it
51:52you are not allowed to discuss the
51:54midterm on discord
51:56if you do we will consider that cheating
51:57maybe somebody's out here looking at
51:59discord to get answers we don't want any
52:00of that happen so if you discuss it on
52:02discord we
52:03we are going to pull your name we're
52:04sending off a student conduct you cannot
52:06do this you cannot discuss the exam
52:08anywhere period
52:08on any channel period until tuesday
52:12at 11 59 pm pacific time okay that's the
52:15rule so make sure you don't do that
52:16don't get in trouble for that make sure
52:17you don't think oh i didn't know i can
52:18tell you no you can't do that
52:20okay that's a bad thing all right
52:23here is some more miscellaneous policies
52:24just to clarify that we're running a
52:25little bit over but we're going to
52:26clarify these misleading policies and
52:28we'll be able to take any questions
52:29after this
52:30the quest retake is hopefully happening
52:32this week that's a much lower priority
52:34it's
52:34like five percent of your whole grade so
52:36we kind of haven't prioritized that
52:37we're really prioritizing making sure we
52:38beta test the midterm
52:39and launch that's worth much more so
52:41that's what took over
52:42all of our kind of circle of inner
52:44circle of tas who normally would do
52:47the autograders for the quest and also
52:49beta testing the midterm they were all
52:51focusing on the midterm to make sure we
52:52delivered i thought a very good product
52:53i hope the midterm
52:55we've had some initial feedback which is
52:56people were very pleased with the
52:57midterm
52:58what the numbers are showing by the way
53:00the mean and mean numbers
53:02is they're showing i'll just show the
53:04first question
53:05you know you have an hour the average is
53:07about 23 minutes remember i'm not
53:08telling you this is not immigration
53:10the the c part that's close caliber is
53:12two hours you not
53:27five is available minutes even though
53:28you have an hour so people are mean is
53:30well below the number no you know when
53:31the mean starts to creep up to the limit
53:33we're like oh no people are having to
53:35cut it up before the end
53:36actually we're doing very well with that
53:38so i think the overall for
53:39all those times ends up being about two
53:41hours and 20 or something so that's
53:42something i told you two to three hours
53:44and the mean is showing it to me that so
53:45i'm very very pleased that there the
53:47exam is doing well
53:49we're hopefully get the midterm grading
53:51by next week there's you know there's a
53:52lot to do we have to do it right and we
53:53have to make sure we cover all the cases
53:55so we don't have a lot of those
53:56iterations of
53:57please do a regrade of it we want to try
53:59to get it right the first time so we'll
54:00spend a little bit more time doing that
54:01that probably is going to happen by next
54:02week so they're very
54:03focusing on the quest for take this week
54:05and we have to rewrite the midterm for
54:07the midterm retake also so we've got
54:08we're doing a lot folks it's
54:10much more than we used to do we should
54:11just write an exam and be done and we
54:12all you know
54:13there's a couple of great scopey small
54:15things no there's like four exams for
54:16every exam now so it's a lot harder than
54:18it used to be
54:19um but hopefully that's working well for
54:21for folks who are remote and able to do
54:23this
54:24lab check we want to clarify what a lab
54:26check-in is called before they were
54:28called lab
54:28check-offs where you got points for
54:30being there for answering technical
54:32questions
54:32okay this is just clarify this is no new
54:35information this is just telling you
54:37what we really
54:38meant if you didn't know that so the lab
54:40completion is no longer mandatory
54:42so that you don't have to finish the lab
54:44if you want to it's out there but which
54:46is a resource for you to use and it's
54:47going to be tested so
54:48we recommend you do that but we're not
54:50going to push you on it because we're
54:51not going to put all these things so if
54:52you need to do an rr week
54:54that's fine not optimal because you
54:55probably need them for homeworks and
54:56projects but
54:57you can do that and you don't lose any
54:58points for that the check-ins again
55:01previously those check-offs are still
55:02mandatory you still
55:04need to come to check-ins every week and
55:06again that can
55:07this is driven by you you drive what you
55:09want to talk about you have to come
55:10have a time where you have like an
55:12appointment and you have to make that
55:13time if you don't you'll get points of
55:15that week there's no
55:16do that the week after the week after no
55:17you have to come that week if you miss
55:19it you don't lose you lose points for
55:20that one
55:21and all it is is a chance for you to
55:23have a face-to-face time with a staff
55:24member
55:25and what that means is if you need to
55:27just talk or vent
55:29or cry or get help or get a technical
55:31question anything you want we're here
55:33for you we're here to support you and
55:34our staff knows about resources to
55:35direct you to
55:36for either technical or social emotional
55:38support if you say you know what i'm
55:40feeling okay
55:41ask me the lab and i did the lab ask me
55:43lab questions then we're ready with
55:45those we're already ready with those lab
55:46questions and the check off question to
55:47do
55:48there's no points for getting them right
55:49you just help and that's kind of a
55:50chance to have a little one-on-one with
55:52with the staff member okay we got those
55:55and those are great so we have them
55:56ready and as much as possible we're
55:58gonna have the labs have auto grading
56:00components to them that's again just
56:01kind of a free thing for you to practice
56:03your stuff
56:03not for any points but well you know if
56:05you can order a grade rather than use up
56:06a person's time do that one so
56:08we'll hopefully have as much as that and
56:10we've moved homework seven that was
56:12there
56:12so now they're due we've slid it back
56:14but we gotta have it you know at some
56:16point it's a stop because we have an
56:17exam at the end and it back and so it
56:18back it backs up obviously like
56:20it backs up like a like a sink drain
56:22does so homework seven is now doing the
56:23same day with homework eight
56:25which means that's eleven nine so you
56:27know two weeks in the future two and a
56:28half weeks in the future we've got
56:29some homeworks due and they're out there
56:31and by the way don't forget you can work
56:33in teams on your homeworks
56:35so get your teams together and support
56:37each other and work together on
56:39homeworks and we love that that's just
56:40wonderful okay good stuff
56:42um i believe you still have to submit
56:44them independently
56:45right you work together and whatever you
56:47create you can create to be identical
56:48you and your partner have identical
56:49projects they have
56:50mechanical homework assignments but you
56:52need to submit them in grade scope
56:53together there's no way in grade school
56:55for you to indicate who your partner was
56:56that's what we're saying so a great
56:57scope is treating you as a one-on-one
56:58person but you can be an identical
57:00solution
57:00just make sure you have your partner's
57:02name on your solution
57:04i'm done poor anything to add um
57:08yeah get going with these homeworks as
57:10soon as you have time um because they
57:12both of them
57:12do have a bit of work um you know
57:15they're like two hour homeworks
57:17yeah finish 3b first obviously you know
57:19finish the midterm first
57:20then take a decade of breath and to
57:22breathe and rest
57:24then jump in on that project 3b finish
57:26that up you have till sunday so that's
57:28good we give you some more time
57:29hopefully that's okay
57:30then take your breath and then jump back
57:32in your homework so hopefully that's
57:33okay and then
57:34farther farther field is waiting for you
57:36which is going to be fun and hopefully
57:37you enjoyed this one
57:38i'm going to give lectures after borah
57:39talks who does this vm and io lecture
57:41my lectures are going to be thread level
57:43parallelism how to make how to make use
57:45of those multiple cores finally going to
57:46learn how to do this class exciting
57:48we'll teach you that and then you'll
57:49have a project for to jump on that so it
57:50should be really fun people really have
57:52enjoyed
57:52the parallelism that should be good
57:54stuff okay
57:56phew about appointments um
57:59i think they're closed now but we gotta
58:03open them
58:04uh so we'll remind the that's right yeah
58:06i think that's right on the appointments
58:07okay
58:08good stuff gotta open them so people can
58:10sign up okay
58:11and we've got still about 58 people here
58:14anybody else
58:15anybody else want to share some stuff
58:16it's it's an exciting weekend we wish
58:18most of our
58:19we wish everybody safety you know that's
58:20the most important thing with all of the
58:22if if they say to evacuate you should be
58:24evacuating folks a lot of people in the
58:26in the berkeley hills i'm not sure many
58:27students live in berkeley hills but
58:28please listen to them when they say
58:30evacuate
58:30or even even if they're volunteer not
58:32mandatory please think about that for
58:34safety reasons because that could be
58:35scary the berkeley hills have had
58:37major major fires in the past and if you
58:39live up in the hills i don't know how
58:40many people are renting houses up in the
58:41hills but though i used to live up there
58:43every time i lived across the street
58:45from where the oakland fire was across
58:46the street from the oakland fire
58:48uh and it was a terrible loss of life so
58:51please don't
58:52don't do that and i understand that
58:53irvine is on fire right now and i'm so
58:55sorry
58:56that folks are suffering that but i
58:58think that that's that's a challenge
58:59we had fires up in napa recently in
59:01sonoma um and please do reach out if
59:03you're having troubles with this we
59:04certainly have
59:05extensions and by the way just to remind
59:06you we have an incomplete for anybody
59:08who needs it and i will give you an
59:09update i just want to give you
59:10some numbers update uh i see that eight
59:13students have taken advantage of it and
59:14that's eight students who are now
59:15breathing a little bit easier
59:16now realizing that's off their shoulders
59:18if you need it we're here for you so
59:20grab that in getting complete it's a
59:22pre-approved incomplete
59:24i don't i know if no other class is
59:25giving this we are giving it to support
59:27you if you need it
59:28you guys are you guys are uh you guys
59:30have that support okay
59:31and don't feel bad about that yeah don't
59:33feel guilty it's not like oh i do it i
59:34feel
59:35like we're looking down on you that's no
59:36problem if you have it just pull the rip
59:38cord and we got it there for you no
59:39problem folks we're there to support you
59:41that's fine no big deal no big deal
59:44good stuff okay all right
59:48i think we're good i'm i'm seeing it
59:50i'll get powered at some point i i need
59:51to
59:52record the next four lectures
59:55porn's going to be in his core office
59:56with his books hi i'm in my office i'm
59:58recording a lecture
1:00:00yeah virtual memory enjoy it
1:00:04virtual office
1:00:08and by the way thank you so much for
1:00:10borah for great inviting those guests i
1:00:11want to give borah credit for this he's
1:00:13been mostly the one
1:00:14handling and making their relationships
1:00:15with all the guests so i want to thank
1:00:16bohr for doing a great job with our
1:00:18reaching out to so many by the way this
1:00:20has never happened before and this is a
1:00:21great gift to the class so i want to
1:00:23thank you what you've done for the
1:00:24for the students we try to get these
1:00:26guys to to to to stay and pay attention
1:00:29to us as well right i mean you guys are
1:00:30working hard i do want to make sure that
1:00:32you're connected to jobs uh
1:00:34and definitely trips and stuff like that
1:00:36and andy supposedly doing
1:00:38really well we'll find out tomorrow how
1:00:40well they're doing is the earnings
1:00:42report
1:00:43yeah i think that's what the paperwork
1:00:45is working on now i'm sure
1:00:47the only last minute people i i guess
1:00:49that's
1:00:51right as well but look boy i've been
1:00:53around berkeley since 1990 that's 30
1:00:55years i
1:00:56have never seen an undergraduate course
1:00:57that had this powerhouse
1:00:59array of speakers in the history 30
1:01:01years i've never seen it happen i know
1:01:02but you too
1:01:03never i i mean i that that's the idea i
1:01:07mean this semester
1:01:08nominally sucks it's online right at the
1:01:10very least
1:01:11grab somebody we can bring
1:01:15people who there is no way that you know
1:01:17that they might be
1:01:18become like victoria coleman or or
1:01:21martha master right
1:01:24yeah yeah good stuff good stuff and
1:01:27darpa come on forget don't forget darpa
1:01:29folks
1:01:30it's proper project upper lead here all
1:01:33right folks
1:01:33good to see you all stay safe everybody
1:01:35do your midterm today if you can
1:01:37don't push us tomorrow because we don't
1:01:38have a lot of staff to support that so
1:01:39do a bit trip today good skill to all of
1:01:40you
1:01:41knock it out of the park hopefully you
1:01:42guys are going to enjoy that take care
1:01:43folks thanks again bye
1:01:44all right bye thanks franklin good stuff

About this transcript

This page contains the full transcript of [CS61C FA20] Weekly Lecture 10.LIVE - Cache by CS 61C Departmental, generated from the public captions YouTube serves with the video. The transcript has 11,280 words across 1,867 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.