12:33. We’re ready! Is everybody excited? How many of you are excited? Alright! I’m gonna make you more excited! Because I am going to ask you questions. If you give the right answer, you get a toy. Alright, are you excited now? YEAH! I thought you might be. Okay, good! Yeah! Alright, are you excited now? is everybody out in TV land ready? I hope you are. Hopefully nobody’s having audio problems. Let’s get started! So, by the way, I’m Henry. Hi! Who are all of you? Are you nice people? I hope you turn out to be nice. That’s good Alright. This is an experiment Everything will fail. Are we excited? I mean, are we happy that it’s all gonna go wrong? Badly, badly wrong. Good, okay. By the way, if you’re out in TV land and you are on anything where you can talk, make sure you mute yourselves because otherwise we’re gonna have all kinds of trouble Okay. And, by the way, make sure you mute yourself. If you have questions, you can send them, other than the folks here in the room, if you have questions you can send them to supercomputinginplainenglish – all one word – at gmail.com. Okay, and you can get all the instructions for this from the webpage: www.oscer.ou.edu/education. So, if anything goes wrong, we are still good. If you want to watch on youtube, just go to youtube and search for two words: supercomputing space inplainenglish – all one word. And we thank Skyler Donahue at OneNet for pulling this together for us. We are also on Twitch. Did anybody here know that Twitch was a thing? I had never heard of it before. Apparently it is a service where you can watch people play video games. I didn’t know that was a thing, but apparently it is a BIG thing. So, twitch.tv/sipe2018 and you can watch all of these, by the way, any of these methods should work from your laptop, your desktop, your phone or your tablet. And, then we have wowza, which is out there broadcasting for us. So here’s the URLs for that. There are two different URLs. The first one is much, much better, so use that one. The phone bridge is no longer toll-free, but it is on. It only handles like a hundred connections, so only use it if everything else fails for you. We do have the phone on, but we have the sound turned down to zero you know they use it if everything else fails for you. We do have the phone on, but we’ve got the sound turned down to zero so you can’t ask us questions on the phone Okay, mute yourself. Questions by email Again, supercomputinginplainenglish- all one word – at gmail.com If you try to use any of the chats for questions nobody’s monitoring those. We have two people monitoring the email address and nobody monitoring anything else. So, don’t try and use the chats. If you’re in the room…again, your two choices are: either you fill out and turn in the talent release form, or you sit behind the camera and you don’t say anything. Okay, I put up a tentative schedule that is subject to change Hopefully, it won’t change very much, but there will be at least one week when we don’t have a session because I felt it would not be fair to people here on site if we didn’t have a session the week that the university was closed for spring break. So, we won’t be doing that. I know some of you are grad students, so you don’t get to take spring break anyway. Some of you are faculty or staff, so you totally don’t get spring break But, there’s that option. I want to thank everybody. Alot of people are helping with this and they are working very hard Pretty much, the whole team of the OU Supercomputing Center for Education and Research is helping us today, so we’re very grateful to them. Also, the network team has been helping us and our CIO and over at OneNet, which is our state research, education, and government network, Skylar Donahue has been a hero and has made all of this possible. He’s the one who told me that Twitch was a thing and that people would be excited to have access through that. And then, over at Oklahoma State, my counterpart there, Dana Brunson, is lending us one of the Zoom licenses. So, we’re doing REALLY good here. And, again everything will fail It’ll be so exciting! Are you excited? How many of you are excited…about the failure? Yes! Two people are excited about how much it will fail! Many exciting events coming up this year. Some of them in spring; some of them in fall You won’t want to miss them. And, those of you out in TV land, you’re gonna want to show up and enjoy it with us, so make sure that you have an opportunity to come to all of these events. Some of them are free and other ones of them cost money. So, alright, so literally, the only closer to the microphone I could get would be right here, but I’ll try and shout real loud when I’m over there. How’s that? Yes, a question. Oh, jeez…Okay…let me see if I can fix that The best I can do for that…let me see give me a second, I’m gonna try some

magic. Nope, nope. They took that away from me. Back to meeting. There we go! I had a thing…I’m gonna cancel…new share…now stop share…share again…and let’s see, if I share my desktop whether that changes anything. It was working beautifully before. What do they see now? See, I told you this was an experiment! We have to give them a second. Nothing? Anything? Okay. I’m gonna try one other thing and then they can tell us. If they say it was working perfect before, then let me know But, do-te-do-te-do…and we are recording, too. Ops…Hang on I’m gonna go back to share, but I’m gonna try and share the slides. I swear, it worked beautifully before. Let’s see what happens here Alright. And, we’re just gonna have to do it large screen. And I’ll just try to get rid of this as much as I possibly can. Unpin. There, we’ll make that work Okay. That stepped all over my great joke! – alright, so. Did you turn the lights down in here? In the back, that metal thing with little lights on it and then all the way on the other side its companion. Yeah, turn it down to like the lowest setting or something. There, now we can see the slides. Same thing over on the other side. See, it’s another miracle of modern technology! Alright, so, I’m gonna distill this whole top down to its essence. Are you ready? These are people Thank you. And, here are some things. Okay, that’s it. Thank you. Okay Wouldn’t that be nice! I actually went to that talk one time. A guy got up and he showed a page of formulas and he said these are some numbers and then he put up a picture and he said this is a result. And then he sat down And he got a standing ovation That’s beautiful Alright now, the way I give this talk is a little weird but I promise you I would make it worth your while So, the way I give this talk is I’m gonna ask you questions and the answers to those questions are on the slide behind me. So, you can make yourself seem like a genius without doing any actual thinking, by just saying what it says on the slide. So, what is supercomputing? So, you get a prize! It’s a t-shirt! Yes, there! Actually, it’s a polo shirt. You won the best prize and we only just got started. Alright, good! Supercomputing is the biggest, fastest computing, right this minute. Therefore, what is a supercomputer? One of the biggest, fastest…oh, look at this! We’ve got a travel mug. I am going to start throwing these to the people in the back of the room. So that’s going to work out really well the room so that’s gonna work out really Right, so supercomputer..one of the biggest, fastest computers, right this minute Why do I say right this minute? It’s always changing in what direction? Faster and…well, as long as by smaller, you mean bigger. So, this is where I get to tell another joke So, if cars had improved over the last 50 years the way computers have improved over the last 50 years, your car would fit in a matchbox, it would go the speed of light and it would crash every 10 minutes. Okay, so by smaller you mean the size of the components. But, as the components get smaller, what gets bigger? Well, speed…yes, but also capacity. Yes, so when I say big by size, I don’t necessarily mean how much of the room it takes up, I mean how much capacity. Good. So, supercomputers are always getting bigger and always getting faster. How fast are they getting bigger and faster? I have a slide for that. Now, here’s a good rule of thumb, if it’s a hundred times as big and fast as what you have on your lap or your desk, then you can probably call it a supercomputer. Okay. That’s just round figures, but now in the world of supercomputing, we don’t call supercomputing…Yes, question? Which camera? Looking at me..yeah, cause it can’t. Right. I

can’t get it to do that. They should just see the slides. That’s correct. Echo in the webcast? On which media? Are other people on YouTube getting an echo on YouTube? Really? Okay, can somebody pull up…can you pull up YouTube on yours and see? You just search for “supercomputing” as one word and “inplainEnglish” as the other word Okay, they may have something old then. Tell them to refresh. Okay, now in the world of supercomputing we don’t call supercomputing supercomputing because if we call supercomputing supercomputing, then you might think you know what we’re talking about and then we can’t charge you a lot of money for our time. How many of you do that professionally? Or plan to? Everybody in this room is lying to me. Okay, so we have other terms like high performance computing or high-end computing or cyber infrastructure is a wonderful term, and, of course, we’re techno geeks so therefore, we’re terribly fond of TLA’s anybody know what a TLA is? It’s a three-letter acronym; we love our three-letter acronyms! So ,we don’t say high performance computing, we say HPC so you’re gonna see that in the slides That’s just to save room. Okay, now I said supercomputers are getting faster and faster and bigger and bigger all the time Okay. So, this is an actual picture of that. So, we can see, I’ll point it out, over here, we can see this sort of pinkish straight line here, that’s Moore’s law. How many of you have heard of Moore’s law? What does Moore’s law tell us? Yes, computing speed and capacity go up with…how fast? Anybody know? Every two years they do what? It doubles every two years! So every two years doubles two years. So, every two years the speed of the computer doubles Every two years the capacity of the computer doubles. Right? Now, by the way, notice, it’s logarithmic in the vertical here, because if I made it linear in the vertical, this goes smooth it wouldn’t be a very interesting graph, but here again, this straight, sort of pinkish purple line, that’s Moore’s law. That computer is doubling in speed every two years. Now, this dark, jaggy line above it, that’s the fastest supercomputer in the world measured about every six months. It’s a wonderful website Top500.org. I referenced it down here and it shows the 500 fastest supercomputers in the world and they publish a new one every June and November. Yes Stay pointed towards the mic. Okay, so I’m gonna try and get louder and softer as we go. It’s gonna be brilliant! Okay. So, fastest supercomputer in the world. And, you notice that supercomputers are getting faster and faster. Faster than computers are getting faster and faster. So why is that? We’ll explore that toward the end of the session, time permitting. Okay, what is supercomputing about? Speed and size. Okay, here is a lovely toy. Bring out the toys from that bag. Okay Yeah, size and speed! So when I say supercomputing is about size, what am I talking about? Good. Flesh that out slide what does it say on the slide behind come on end up with shy so what do I need thy size memory good so how many of you have a laptop or desktop PC that has one gigabyte or more of RAM one gigabyte or more good to gigabytes or more of RAM four or more eight or more 16 or more 32 or more okay what do you what do each of you have a VRAM er of disk by the way by the way you can buy a terabyte of RAM today so I was gonna accept that as an answer okay what and what do you have on RAM how

much 80 Wow okay I have 12 so I hate you you’re 16 I hate you anyway I only hate you merged no I hate him a lot okay so now how about hard drive how many of you have 250 gigabytes or more of hard drive or SSD okay 500 gig or more described terabyte ahora two terabytes okay so the biggest is a how much how many terabytes of disk you have you have two terabytes okay so I hate you still hate you less than him but I still hate you okay so on this one I’ve actually got two hard drives in here but together they add up to less than a terabyte so I don’t have that much okay so suppose I had a problem that was bigger than 80 gigabytes of RAM and bigger than 2 terabytes of disk or bigger than 2 terabytes of disk could I run it on his PC or any PC in the room so then what would I need I need a supercomputer good now when I say supercomputing is about speed what am I talking about hints slide okay so let’s Klaus floating-point operations per second good so what about them in slide I want to be able to process more in less time why is that good by the way I agree but why would I want to process more in less time why is it good at the same time why do I want my results quicker by the way I totally agree but why why is that good I completely agree with you but why because okay say it again time is money exactly correct yes because time is money right we’d like to think that it’s not all about the money right we want to think that it’s the good warmhearted stuff what is it all about it’s all about the money okay so and by the way so what are the advantages of getting the result quicker why is why do we want to get the result quicker because you can get more stuff down the same amount of time absolutely why is it good to get more stuff done in the end that’s what it comes down to we want to be more productive whatever your value system is whether it’s earning a profit whether it’s publishing papers whether it’s providing a service to someone whatever your value system is doing more of that is better right so if you can do more in the same amount of time that’s better right I can publish your papers this year I can earn more profit for my company this year whatever it may be okay so suppose I had a problem that I could do on my laptop in a month is that good or bad doing this problem on my laptop in a month how many of you thought that’s good how many nobody voted good okay how many do you vote that’s bad okay everybody votes is bad all right why is it bad that’s a lot of time okay so if that’s a lot of time then what’s the negative implication of it taking a lot of time in real life practical application you’re gonna run something big number crunching Java on your laptop for months you can’t do much else on your laptop because it’s going to consume all of the CPU probably a lot of the RAM you’re not going to get anything else you can’t even do your email right your laptop is running super hot and you can’t use it for anything else how many of you can go without doing email from us believe you can do that I’m sorry without getting fired okay so not practical to not have your laptop for a month what’s another downside Thank You yes how many of you have had your laptop stay up for full months without crashing okay somebody here’s lying okay goodness knows it’s never happened to me so I don’t know if you guys know some magical secrets but I’ve never had that happen right so what happens if you run for 29 days and on day 30 your laptop crashes you’ve got to start all over again so now it doesn’t take one month right takes two months except what’s gonna happen on day 28 in the second month your laptop’s gonna crash again will you ever get done no you’re never gonna get done it’s that good or bad what’s gonna happen to you yeah you’re fired or if you’re the CEO of the company the company will go bankrupt because you never get anything done right okay but suppose this probably would take a month

to do on your laptop suppose we could run it on a supercomputer in an hour would that be good or bad how many of you vote that’s good lots of votes tell me vote that’s bad how many of you vote it depends Oh why do you say depends well let’s suppose this problem I can run on a supercomputer in an hour where I’m making up a magical theoretical I’m positing that this particular problem is the sort of problem that on the supercomputer we can run it out so I’m making up this case it’s it’s it’s purely hypothetical but suppose it’s a case of something to run on your laptop in a month or on the supercomputer in an hour is that good or bad to run down on the supercomputer in an hour okay so what if I mean I’m going to teach you an important rule so anytime I ask you for a value judgment is it good or bad the answer is always it depends so suppose you could run out a supercomputer in an hour but I have to charge you a million dollars a minute Oh suddenly you’re not interested okay all right so yeah all of the conditions change don’t they right so the answer is gonna always be it depends so what is the answer he’s a good or bad to run in an hour on supercomputer it depends by the way I will not charge you a million dollars a minute to run on our supercomputer if you are an academic or rather a not-for-profit research and or education organization then you and you’re in Oklahoma you can use our supercomputer for free all right so what do we use supercomputing for what’s a category hint blue the bold-faced underline italics of things we use supercomputing for what simulation okay what does it mean to simulate reproduce good what else give me another word first you late-model good I love that that’s a nice science word now give me a kindergarten work we’re gonna get that down that in the third one something we used to do when we were in kindergarten we would say let’s play that we are Disney characters say again pretend yes simulate means pretend we’re going to have the computer the supercomputer pretend to be a tornado or an oil reservoir or a molecule or a star system whatever it is that we’re studying we’re gonna have it pretend so for example oh you of course this big weather school so one of the things they do they do lots of weather simulation the way you forecast the weather you take a snapshot of the weather right now so satellites and radars and ground stations and weather balloons and even commercial aircraft are constantly sampling weather data as they fly around you’ll take all of that you mush it together to get a picture of the weather right now we call the initial conditions you plug that into the software that simulates weather physics processes and outcomes a simulation by applying those physics processes of how weather behaves and what the simulation shows 24 hours from now that’s our prediction of what the weather will be 24 hours from now and then 24 hours goes by it turns out we’re totally wrong but that’s ok because we already have your money and then tomorrow we do it again right how many of you are in the weather business okay one person is that roughly the business sure ok yes in back and not on camera so you can get away with nada all right good ok what’s another category hint blue italic underline bold-faced data-mining what does that mean you’re absolutely it is you get other people in the face ok so yes finding needles of information in a haystack of data so a big one of course nowadays is genomics how many of you have heard the term next-generation sequencing is anybody in the bioscience business okay we have one bio scientist okay so next generation sequencing how much data does that produce a little or a lot a lot it used to be a little but that was like 25 30 years ago I guess early 90s so 25 years ago it used to produce only a little bit of data now it produces an enormous amount of data I’ve got a picture of that later on so you need a way to choose because you can’t just look at the data by eyeball right anybody know how many ACGT there are in the human genome roughly by a science person you should know that yeah you didn’t know you were coming to an exam if you remember roughly the number that looks like an O okay three billion base pairs three billion of the ACGT in human genome can you just look at

that with your eyeballs and make sense out of it no you’re gonna need help from the computer to chew through that stuff looking for interesting patterns that then turn into interesting proteins that then turn into industry interesting structures like you know your nose or your brain and then you’re especially looking in many cases from interesting structures that are not how they should be so to give you a very concrete example my aunt and my grandmother both died of Alzheimer’s when they were in their mid 80s and I’m feeling okay so far but I just want to let you know and bio scientists in particular you’ve got about 30 years to get this worked out so no pressure but it’s got to get done and guess what’s gonna be super helpful for you supercomputers absolutely okay the the gene sequencing stuff that is job security for people like me okay and then what’s the third category bold-faced underline italic blue visualization was that mean turning data into pictures so suppose I have a stack with a trillion numbers right I print out a trillion numbers and I hand you this giant pile of paper with a trillion numbers on are you going to be able make head or tail of that will that mean anything to you shake your head loader with your mouth no not at all but suppose I give you a picture of that trillion numbers or even better a movie of the trillion numbers or even better a video game of the trillion numbers where you can walk around inside the trillion numbers now will that make sense to you now will you be able to get some insight out of your trillion numbers because the purpose of computing is not numbers the purpose of computing is insight okay now this is the single most important slide in the entire talk this is everything important about supercomputing with nothing left that how many of you believe me how many of you believe that I can distill everything about supercomputing down to two bullet points how many believe it anybody to a few of you how many of you do not believe it reserving judgment okay a lot more reserving judgment than actually believe me this is I promise you everything important about supercomputing with nothing left out are you excited so what are the only two issues that we are concerned about in the world of supercomputing inside the storage hierarchy and parallelism what does it mean parallelism yes doing multiple things at the same time you used a $10 word simultaneously but I still bet back at the same time so how many of you either you do this or you have family members or friends who do this you are reading a book while watching TV while talking on the phone with your friend while listening to music while playing a video game how many of you do that or know someone who does that okay now when you’re doing all of those things are you literally doing all of those things in the same moment shake your head louder with your mouth no you’re not what are you doing instead you’re switching from one to another in the computing world we call this time slicing so that’s the technical term for that is concurrency but that’s not parallelism ilysm is where you’re literally doing all of those things in the exact same not excited okay so all boast of that how many of you have seen Pirates of the Caribbean and by this I mean the good one the first okay all right how many of you then went and saw the other four and you sort of felt stupid and dirty when you came out of the theater because he paid money for that hey how many of you admit you’re gonna go see the next one when comes out which would inevitably will because they each make a pile of money right okay so with the first one there’s that scene where they’re marooned on the desert island and they have been drinking the rum there’s big bonfire going Captain Jack Sparrow says this he says what a ship is it’s not a keel and a hull and a debt and sails that’s what a ship needs but what a ship is what the Black Pearl really is is freedom then he curls his mustachios and passes out but so this is a really important point now how many of you have heard the term cluster in the context of a supercomputer okay so what is a cluster it’s a bunch of CPUs and in order to get that we typically have a bunch of PCs okay so it’s a bunch of pcs what else do they need hints slide they

need to be connected with a network what else they need some software to make it possible to do stuff right but that doesn’t make it a supercomputer you can have all of those components and you’re still not a supercomputer yet to be a supercomputer what you actually need is you need all of those pieces to believe so hard that they’re one big computer that it actually comes true and what I mean by that is you don’t become a supercomputer until all of those components are at least a big subset of them are working together on a single problem that no one of those pieces could do on itself a problem that’s either too big or would take too long or both so here’s a picture of that this was our very first cluster supercomputer here at oh you back in 2002 so this is now what 16 years ago before my long white beard made it all the way down to the floor you know and the reason I like showing this picture is you really can see this in a very clear way in this photo so if you look at the part on the right and in particular if you look at the part on the right in the picture on the right what you’ll see that stack of silver-backed boxes those are pcs now they’re not shaped like a regular PC but they got the same stuff inside them same kind of CPU same kind of RAM same kind of disk same kind of network they do have an extra network that’s even better than your usual network but chugging along very nicely and then on the left of the right hand picture that’s the network now what you can’t see because it’s a still photo is lots and lots of lights blinking there’s nothing better than blinky lights to tell you you’ve got a really high-quality supercomputer so we had beautiful blinky lights on that one tragically the networks we have nowadays not enough blinky lights okay how many of you are experts on computing and how many of you are normal people okay good all right so in that case let’s talk about what’s what’s inside a computer so and by the way here’s my actual laptop these are the actual specs for my actual laptop I did add a second hard drive an SSD but other than that this is pretty much the same specs as when I bought it okay what are the components that we find inside a computer and by the way I should tell you anytime anybody starts categorizing in an engineering context they’re totally lying to you so I am officially lying to you everything on this slide is a lie however it is a useful lie I am oversimplifying to make this more straight phone okay so what are the components of any computer what do we find inside CPU yes and probably have some points I can give you in Spain for reading these Oh a frisbee who would like a frisbee yell off what’s on the slide and you get a frisbee nobody wants the frisbee I gotta get rid of it I promised my wife I would get the stuff on the garage you want the frisbee okay hey I did pretty good all right now you have to read out loud what it says on the slide though no go ahead yep yep mm-hmm and exactly correct good cpu primary storage secondary storage input and output all right let’s talk about the CPU what is the CPU it’s the brain now is it really a brain no so I put brain in scare quotes but it’s the thing that does what we think of as you and the CPU has several components inside of it and we’ll just talk about a few of them so what’s the the big one that that’s kind of the gateway to everything else and I will dangle a toy oh I’m not even sure how I got these but I have a lot of them so if anybody would like a necklace anybody have kids who want something shiny no seriously I have a wooden spoons from Louisiana State University I’m not gonna throw this but who would like a with the spoon in exchange for telling me about the first component yes yes give me some examples of what the control unit might do that which we know as yes so all those wonderful things so basically the control units job is to decide what everybody else is supposed to be doing okay how many of you have had that boss really nobody okay one person willing

how many of you are that boss okay but many of you will become that person okay what’s next after the control unit we have the Oh put it this beautiful bag oh oh sure now you wanna tell me about the arithmetic logic unit yes it does calculations like what any examples yes so beyond the ones I listed on the slides what other kinds of calculations might a an ALU do in addition to addition and multiplication subtraction sure so yes subtraction is closely related to addition so is multiplication for that matter what else division exactly correct but division is closely related to multiplication but on the other hand how you do division is very different from how you do multiplication how many of you did long division in grade school that’s basically what CPUs dinner which for the record is why it takes a lot longer on a computer to do a bunch of divides and a bunch of multiplies multiplies SuperDuper fast divides painfully slow guess what’s even slower than that square root square how many of you have ever done the long division like way of doing square root of anybody learn that yeah I’ve seen that one okay so that takes even longer um we’ll talk about that as we get deeper into the semester but one of the big ones that takes forever is raising this non integer to that non integer power that takes forever I’ll I it takes like a tiny fraction of a tiny fraction of a second but still okay and then what’s the last component of CPU registers let what are registers right so there are little tiny storage locations inside the CPU where data lives when it’s actually being operated on so if I want to add two numbers together the add end has to be in a register the organ has to be in a register and when I press the magic Add button the sum is going to end up in a register okay so that’s essentially what registers are now let’s talk about primary storage by the way there are two kinds of people in the world people who split the world into two kinds of people and people who ducked so there are two kinds of storage in the world what are they hit look at the title of the slide primary end very good okay so under the category primary storage what are the two types of primary storage RAM and catch okay so tell me about RAM or main memory by the way what does RAM stand for random access memory what’s the opposite of random access memory yeah we’re trying to figure out what’s the opposite of random so the answer anybody else so the answer is sequential access so what’s an example of a sequential access storage technology Hey yes how many of you remember audio cassettes good how many of you regularly use audio cassettes anymore okay literally not one hand goes up okay so in the olden days when there were audio cassettes and you used to make someone if they were your friend or if you wanted to go out with them or something you’d make them a mixtape right how do you remember mix tapes I mean remember mix CDs what are the kids make now playlists how much love and effort goes into making a playlist about five minutes how much time had to go into making a mixtape it’s at least an hour usually several hours he actually had to record it you couldn’t you couldn’t interrupt that you actually had to go through the whole thing at speed right so suppose someone makes you this beautiful mixtape full of love and your favorite song is at the beginning of side a and your second favorite song is at the beginning of side B what do you have to do you’ve got a merge from one end of the tape to the other right you can’t just skip so you either fast forward and then flip it over or you flip it every Y right but you’ve got to merge for one under the other that’s the opposite of random access random access means you can jump from anywhere to anywhere else magically in zero time now of course it’s not literally zero time but it’s close enough let’s think of it as zero time later we’ll find out is worse than zero time but okay so one bday to live in RAM it’s like yes when it’s being but used by program that’s doing what currently running exactly correct so if I’m running a perm so right now I’m running PowerPoint where are my slides my data in RAM exactly correct good okay what’s

cash a small area of much faster memory that’s exactly correct okay so um when do day to live in cash if it’s faster when do we want them in the fast stuff yeah by the time we need them right when they’re about to use so data show up in cash when they’re about to be used or if they’ve just been usually probably a mini kicked out all right I’m sorry go ahead question Oh such a great question I love your question can you wait maybe three or four slides and we’ll totally talk about that because that is the perfect question to ask I love that question okay let’s talk real quick about secondary storage so give me some examples of secondary storage in sly CD go on our drive sure floppy oh how many of you remember floppies how many of you still use floppies Oh so zero I mean oh wait you do use it for decoration that does not count but I will however give you a prize this is a bouncy ball oh all right very nice okay so you got a bouncy ball all right so when do day to live on secondary storage so when you plan to use them sometime in the future hey when you’re gonna get to them so let’s talk about the exciting part of that but before I get to that I oh very boring I have one slide we’re not gonna talk about it all right let’s talk about the storage hierarchy and this is gonna roll back to your question right you’re gonna love this part so they give you Algie suppose I want you to move your best friend from Maine to Mexico on land as fast as you possibly can here’s a million dollars so what are you gonna spend your million dollars on what are you gonna buy to move your friend from Maine to Mexico on land as fast as possible a Jaguar right I put a formula one I’m not sure actually whether you can take the formula one out on the open road but never mind so a super-fast expensive car right okay so far so good okay now suppose I tell you I want you to move your hundred best friends from Maine to Mexico on land as fast as possible but not her person as fast as possible for the hundred of them now what are you gonna spend your million dollars on a bus a train I don’t know if you get a train familiy knowledge maybe get a bus for sure four million dollars absolutely I put here a fleet of little cars right but the same idea I’m gonna want to get something cheap and junky but a lot of it right because that’s gonna move the bunch of them as fast as possible okay same principle applies to computing so in computing we want things that are fast for a few we can’t afford enough of that for men so the rule is if something is fast then it’s expensive if it’s expensive can you afford much of it no you cannot but if something is slow then it’s cheap if it’s cheap can you afford much of it absolutely so registers are blindingly fast I’ll show you some real numbers a little bit registers are blindingly fast they’re terribly expensive although I can’t show you numbers for that and therefore you have hardly any registers so current CPUs are running at several kilobytes how do you remember kilobytes how many of you ever owned a computer where the word kilobyte even came up be honest okay I did back in 1970 mumble when my dad got me a trs-80 I was the first week on the block and have there computer we don’t even call them personal computers back then we call the microcomputers right that’s a word that’s fallen out of use but my trs-80 had four kilobytes of RAM so very very excited by the way huge excitement when my dad paid 200 bucks which is a lot of money late seventies 200 bucks to get me the upgrade to 16 kilobytes of RAM to give you the feel of it that’s about enough RAM to hold 16 emails worth of text which is to say it’s like no RAM at all right you literally cannot buy that small amount of RAM okay so cash is almost as fast it’s not as fast as registers still quite fast therefore pretty expensive and I can talk about

what cash not therefore quite small so typical CPU today has a few to several megabytes of cash okay main memory is quite cheap therefore sorry it’s quite slow therefore quite cheap therefore pretty big so today several – even tens of gigabytes is normal on a PC penny how much money you want to spend so again this one here my laptop twelve petabytes hard drive slower therefore cheaper therefore bigger right again hundreds of gigabytes Ruby terabytes removable media you can have stacks and stacks of them and then of course the Internet is gigantic but terribly slow how many of you have had that experience the internet is slow oh if your hand isn’t up you’re kidding yourself okay yeah let’s talk about this Ram is slow business is RAM faster slow how many of you vote Ram is fast how many of you vote Ram is slow how many of you vote compared to what oh yes so if I say something is faster slow is that an absolute or is it a relative it’s relative so it’s not that ran as fast it’s not that Ram is slow its ran as faster or slower is slower than CPU by a little or by lot by a lot so on my laptop and these are the real numbers I look the stuff up I even ran some benchmark performance tests to find out on my laptop if I could keep my CPU fed it could chew through over 650 gigabytes per second my RAM can only do 15 gigabytes per second so which one of them gets to decide how fast I get my work done the RAM the slow one always wins ready it’s like trying to shoot a fire hose to a drinking straw you’re just not going to get that wet on the other end so that’s the term for that is that’s a bottleneck right so cache doesn’t solve this problem but it does improve it right in engineering can we ever solve a problem in an engineering context you ever really solved the problem or do you just make it be not that bad you got to make the problems not that bad so cash helps make the speed issue not that bad so if you can get the data you need into cash by the time you need them then you can get your work done much faster again that’s such a big if it’s job security for people like me okay all right so here’s my laptop and now here is an actual table of actual numbers of real life okay so I mentioned about the over 650 gigabytes per second that my CPU could chew through can’t talk about the price of the registers because you can’t go out and buy more registers for your CPU you can’t even necessarily go buy a CPU that’s the same as the other one except it has more registers that’s not really offer but it turns out if you ask the people who designed CPUs oh yes the registers are SuperDuper expensive that’s why you don’t have much the other reason you don’t have much as it turns out is having lots and lots of registers doesn’t make the computer that much faster than only having a few is that weird let me view it that’s weird how many of you go that’s not weird it’s an actual paper where they did a bunch of tests this and it turns out there’s an upper limit to how much money it’s worth spending to get more registers and after that the incremental improvement in performance is like a few percent doesn’t come okay so much much slower than the registers but still pretty fast much bigger so by the way everything here is megabytes except that first one so three megabytes compared to 10 kilobytes so a factor of 300 bigger and I actually found two CPUs that as far as I can tell the specs are identical except the amount of cash in them and I gotta tell you I like to look at a lot of CPU specs to find these to a different price today or at least last week by about $20 of per megabyte so that’s roughly the cost of a megabyte is 20 bucks ok so now I’m going to answer your question why do we even have RAM why don’t we just have cash so let’s say I wanted to have 10 gigabytes of this primary storage on my laptop at $20 per megabyte how much is my laptop can cost so let’s see so 10 gigabytes so that’s 10 thousand megabytes at $20 per megabyte so 200,000 dollars for that laptop how many of you are willing to pay $200,000 for laptop every four years

okay how many of you are nowhere near that dumb okay not that dumb very good okay so if you’re not that dumb you want a better solution so you’re willing to trade that the storage will be slower than cash which is and that’s a lot slower see the numbers over here you’re willing to trade that away in exchange for you get a much cheaper laptop and your laptop even though it’s much slower than it would be with all cash he’s good enough I’d much rather pay two hundred or two thousand dollars for laptop again every four years and two hundred thousand yes oh yes so if you are if you are in the financial industry you are willing to pay top dollar but even then you can’t afford two hundred thousand per because you still need a lot of them right you don’t just need one PC to do the work you need all right then Ram you can see is slower and it’s much bigger so what do we get we get a factor of four thousand more ram and cash in my laptop right so one penny per megabyte instead of $20 per megabyte so that’s a pretty big difference right but then if you look at my hard drive of this it’s three percent sorry it’s three tenths of one percent of a penny per megabyte of harddrive that’s a pretty good grade by the way that’s just going and looking on on various websites there’s a wonderful website called price watch comm that will show you the prices everywhere else so you can see what things I just looked at what’s the cheapest one there so you can see all of these things are actually happening now why is it why does the storage hierarchy actually work why is it that fast things are expensive but slow things are cheap why anyone want to venture a guess oh you’re way ahead of me yes okay see you solved the problem usually we get guesses like well because the fast ones are very expensive to manufacture or whatever and oftentimes those reasons are even true but also irrelevant what you just said is the correct answer so I’m gonna put this in terms of what mathematicians call proof by contradiction so suppose that I have a slow expensive storage technology I’m going to magically make up a name for this I’m going to call it arbitrarily I will say that it’s called floppy okay so suppose I offer you a floppy at about 50 cents per megabyte how many of you will buy 50 cents per megabyte as opposed to when we come up with one third of 1% of a penny per megabyte for hard drive right how many of you will pay 50 cents per megabyte okay also I should mention that a floppy runs at about 30 kilobits sorry kilobytes per second versus a hard drive will do a few tens of megabytes per second so that sounds like a good deal no it’s a terrible deal and by the way this when I realized this and I’m so embarrassed about how recently it was that I figured this out when I realized this it actually restored my faith in humanity how many of you believe that this fact restored my faith in you how many of you are very skeptical reserving judgment about whether it was okay so this is true I’m not making this up so before I figure this out I actually I am a natural cynic and a pessimist and my up until that point my belief about not just my fellow human beings but me as well was we were all a bunch of morons with our eyes closed feeling our way in the dark stumbling over the furniture and occasionally accidentally doing something useful this is what I genuinely believed about all of us okay and then I figured this out and this told me something very important about kimby because what happened to Fluffy’s how do you own a floppy drive today how many of you use floppies regularly how many of you have a box of them somewhere but you haven’t used them in like ten years or more how many of you don’t even own any floppies anymore or you’re not sure okay so why did we give up on floppies because they are a terrible idea anymore they used to be a great idea 30 years ago but now they’re a terrible idea because they are tiny and slow but expensive so we don’t buy them so anything that doesn’t fit the pattern of fast implies expensive anything that doesn’t fit the pattern of slow rise cheap nobody buys so it’s not that you can’t have a technology that doesn’t fit

that it’s that nobody will use it because it’s terrible right but anything that’s a if tomorrow a new technology came up that was faster than the registers we have today all of the stuff that we buy today would become obsolete but then immediately other new technologies would pop up and guess what the fast ones would be more expensive the slow ones would be more cheap right because fast and slow are really faster and slower that help so that’s why we don’t buy all right let’s talk about parallelism okay and I need to grab something from my bag while we’re talking about parallelism so I’m gonna yell really loud don’t be frightened okay so what was the parallelism what did that word mean not just work what’s happening inside the machine so suppose that I go to Bass Pro Shop and I buy the most expensive fishing pole and the most expensive fishing line and the most expensive fish hooks and I go out to the lake and you and 99 of your best friends the hundred of you get broom handles and baling twine and bet paper clips and the hundred of you go absolutely who’s gonna come home with more fish me with my amazing gear or the hundred of you with your cheap job the hundred of you with your cheap job every single time you’re gonna win right because more instances of it happening at the same time is gonna get the work done faster now how many of you like jigsaw puzzles I just saw you volunteer you didn’t know you were volunteering did you okay come up and sit right here okay now the folks in TV land are not to be able to see this but I’ll show you the slides that captured the idea all right I have this beautiful diesel ready now let’s suppose for the sake of argument that you can do this puzzle in an hour by the way thousand puzzle pieces they just twelve enough so you could be fine thanks so no no don’t wait just go right ahead don’t be shy just get right to work a thousand puzzle pieces and we’re gonna say oh you need to look up May right so it’s a beautiful picture with a pair of kitty cats cats okay so beautiful picture memorize that because it’s gonna okay so can we agree for the sake of argument that he used this thousand piece jigsaw puzzle all by himself in one hour and we agree on that for the sake of our cut every degree thousand puzzle pieces one hour everybody’s agreed that you can do that in one out right you’re gonna have to work a lot faster than that now you despond to you didn’t you say yes of course it did scooch on over come sit here and the two of you are gonna work together alright so the two of you are working together now if one of them can do the puzzle all by themselves in one hour how long is it going to take the two of them working together give me sort of the naive brute-force answer no you’re jumping ahead here I won’t see the take the two of them half an hour why same amount of work twice as many workers get it done in half the time that seems reasonable right except from time to time will it happen that the two them reach into the pile of puzzle pieces at the exact same moment and then they have to do after you know after you or maybe they’ll fight does that happen from time to time is that gonna happen will that take time yes so there’s an overhead cost associated with contention for the shared resource so far so good okay now let’s say that you’re gonna do this kitty cat and you’re gonna do this Kitty right and let’s say that they get to the point where they both finish their kitty cats and now they have to bring their halves of the puzzle together yes so are they going to have

to communicate to bring their piece of their halves of the puzzle together yes will that take time absolutely will so there’s another overhead we have the overhead associated with contention for the fair resource and there’s another overhead associated with communication at the shared interface okay so now if I say how long will it take the two of them do we still believe it’s half an hour less more or the same as half an hour more because in addition to doing the work they also have to have the overhead for contention indications all right how about the two of you can you come join us yes come on don’t be shy you can just sort of stand here up in front this is sort of facing the puzzle pieces here all right so now there are four of them working diligently on the puzzle together will there be less more or the same contention for the fair resource more will there be less more or the same communication with shared interfaces more so if one of them could do it in an hour two of them can do it in let’s say thirty five minutes how long is it gonna take four of them 20 minutes 25 minutes somewhere in there right okay now everybody coming on kidding but if I did bring you all up here would that make you faster what would it make you slow or why because there’s too many people so the overhead for contention would go right more fights more often and the overhead for communication go up more people to talk to right so far so good are we excited okay so now I want to change this up a little bit by the way so this is what we call in the computing world by the way in the computing world what would be the workers it would be the CPUs right and what’s the resource that they’d be sharing together storage typically Ram but sometimes it’s disk so first good okay so in the computing world typically what we’re gonna see is that this shared memory parallelism is gonna top out somewhere around 32 processors give or take okay so much beyond that you’re not gonna get more value out of adding more processors because they’re going to spend all their time fighting or talking now it isn’t always 32 it can be less than that 16 8 4 2 1 sometimes the fastest parallel approach is slower than the non parallel to serial approach good now I’m gonna change the conditions of the test so I’m going to send the two of you back to your seats I know you’re having a lot of fun take exactly correct 0 5 if we re on that segment ok and I’m gonna give you the exact I’m gonna ask you to go sit in that corner way over way over there ok away in the corner there it’s not because you misbehave it’s just part of the thought experiment ok so he’s got his half of the puzzle pieces she’s got her half of the puzzle pieces so far so good will there be less more or the same contention for the shared resource less how much how much contention for the shared resource will there be why do you say zero and I agree by the way we’re not there yet they’re not they don’t have a shared resource anymore they each have their own private resource right so contention actually goes to zero some pretty good now what about communication let’s think about this case of the literal case of the jigsaw puzzle right you’ve got your part of the jigsaw puzzle here she’s got her part of the jigsaw puzzle over there when they each finish their half of the jigsaw puzzle what are they going to have to do in the literal piece of the jigsaw puzzle forget computing the literal piece of jigsaw puzzle what are they gonna have to do to bring their halves together yeah you’re gonna have to move this table over there or they’re going to bring the tables together they physically have to carry the tables because you can’t pick up the puzzle right it’s gonna fall apart you have to physically pick so compared to the cost of an individual communication when you’re sitting right next to each other is the cost of an individual communication now less more of the same more a little more or a lot more a lot more on the other hand are they tempted

to communicate more than absolutely necessary are you gonna yell at her a lot well yeah that’s just a personality thing right okay so it realized you’re not going to communicate because it’s expensive to communicate right so you’re not going to be doing that what question please in the context of parallelism is this like the law of diminishing returns and economics in the case of shared memory parallelism that is absolutely true that as you keep throwing more and more processors at the problem with shared memory almost always there are exceptions but it’s quite common that adding more processors at some point and usually that point is not that many processors it doesn’t add more speed and in fact can start slowing you down that is absolutely true so a law of diminishing returns from economics is a very good analogy to what’s happening in the CPU context for shared memory parallelism we’re gonna talk in a minute about this distributed parallelism where it’s a bit different okay so now we’re distributed how many of you by the way had this happen when you were in grade school at the beginning of the school year you sat next to your best friend and then one day the teacher said you sit over there how many of you have that happen okay why did the teacher make you go sit over there because you weren’t spending too much time talking and not enough time getting your work done and also of course you were disturbing everybody else but let’s stick with the analogy to the computing right you were spending too much time talking so by separating you we reduce your ability to talk because the cost of communication is very high therefore we reduce your temptation to talk more than absolutely necessary so even though the cost of an individual communication has gone way up the aggregate cost of communicating might come down will that affect how many processors can work on the problem together can we get more if there’s no contention and if the aggregate cost of communication has come down if the problem can be split up properly that’s a very good point good so we can add up to some point now by the way we also want to make sure that we get a balanced load load balancing means getting the right amount of work to each of the processors so suppose I give him twice as much work as Earth is that good or bad how many of you bet that’s bad let me field that’s good how many of you vote it depends who well what does it depend on okay so you might work faster so if you work faster than her should you be given more work now I want to make sure we subtract out the morale judgment part of this this is only about getting the work done not about whether that’s the right thing from a moral perspective remember these are computers so the moral part doesn’t apply that much okay so it really depends on the relative speed so if they each work at the same speed and I give him twice but twice as much work as her is that good or bad if they work at the exact same speed how do you feel that that’s good don’t you feel that’s bad let me vote in two patents now we get rid of all the dependence okay so I agree it’s bad why is it bad yes why do I care about that and by the way I do but why time is money so if she finishes up before he does and she’s just sitting idle could we have taken some of his work and thrown it and thrown it to her and then she could work more which means he would finish earlier and by the way she was going to finish it at whatever time anyway right so it doesn’t matter if she keeps working it matters when he gets to stop so it turns out that the fastest that the two of them why do you stop the fastest is the two of them can finish the problem is when they finish at the exact same moment because if one of them finishes before the other we could have given the other some of that ones work in which case that would have ended earlier okay so getting them to end at the exact same moment that’s the minimum amount of trying to do the work now on the other hand suppose he works twice as fast as her then should he get

twice as much work to do yeah because then they’ll finish at the exact same moment and that’s our goal because that’s the fast the real goal of course is to do the work as quickly as possible the way to do that is to have them finish at the same time so load balancing does not mean giving them each the same amount of work load balancing does not mean giving them each a fair amount of work load balancing means you than the right amount of work so that they end at the same time okay turns out load balancing can be easy or it can be hard I will now prove this see okay so on the left is let’s say weather forecasting so you take let’s say the continental US you split it up into chunks and equal size you give a chunk to each of the processors and they do the number crunching and from time to time they talk to each other the one on the right was my dissertation which happily I did not have to load balanced somebody else did that but that was an arbitrary collection of an arbitrary number of arbitrary size arbitrary shaped blocks that was changing arbitrarily over time so it turns out that the best load balancing scheme for that thing what I just said which is hard to say by the way the best load balancing scheme is start by just randomly throwing the blocks to the processors just randomly putting them wherever and then observe and see which processors have too much work based on when they finish and which processors don’t have enough and then just redistribute throw some from the overly busy ones to the less busy ones and you’re gonna have to do that anyway because remember it’s changing arbitrarily over time so when you do it that way you get not by any means perfect load balancing but about as good as it’s likely to get ok now I promised we talked about Moore’s law that be how are we doing on time oh ok so we’re okay we’re gonna go a little over but it’s not a crisis and if you need to drop off your cool to drop off ok so more and more he later went on to co-found and was I think the first CEO of Intel back in 1965 he was at a one of the early chip companies they were hardly any they just started a few years before these were the first what we called back then integrated circuits and he noticed that the number of transistors you could squeeze on a chip was growing very rapidly initially it was growing it was doubling about every year but it pretty quickly sort of steadied out to about doubling every two years the number of transistors you could squeeze on a chip so call it transistor density it turns out that the speed and the Cassity of a chip is roughly proportional to the transistor density so essentially what the implication was is that computers were getting bigger and faster by doubling every two years and again that is the strain sort of pinkish lower line there which again this is logarithmic in the vertical in a linear linear graph goes swoosh but so the straight line that’s Moore’s Law doubling every couple of years and then the upper jagged line the dark line that’s the fastest supercomputer of the world from top 500 world and you can see that a loud nose in a much more stair-step kind of way where something will be the fastest in the world for a year or two sometimes you can see that supercomputers are getting faster and faster faster than computers are getting faster and faster so how’s that possible how could supercomputers be improving faster than computers what did we get better at we got better at parallelism exactly correct and I have some actual numbers okay so in 1993 in June of 93 the very first top 500 list published was about a thousand CPU cores and it could do about 60 billion calculations per second the most recent one which came out in November of 2017 the next one will be June the most recent one was ten point six five million CPU cores and yes they have run a single application on all ten plus million CPU cores all at the same time they have actually done that okay now a lot of them but they have done it and so that one is capable of doing 93 million billion calculations per se ok and what we got better at a lot better we got a lot better at parallelism okay and Moore’s law is an uncannily accurate predictions one of those accurate predictions in human history so Intel came out with the great great great great great great great-great-great great-great-great-great great grandmama of the current ships that we have from

Intel a and B there’s other manufacturers who make what we call x86 they came out with the four thousand four back in 1971 and it had two point three thousand twenty three hundred transistors so in 2010 this is already now almost eight years ago they came out with a chip that had 2.3 billion transistors on same size chip a million times more processing power and capacity so transistor density has doubled about every 23 months over that what was that 40 year period over 40 year period that held steady it’s actually starting to slow down a bit now because we’re starting to get to the physical limits of what the current technology can do the physics is starting to get weird right but unbelievably accurate prediction so here’s kind of a picture this doesn’t matter the year doesn’t matter the speed by the way again logarithmic in the vertical because otherwise it just looks like swoosh but here’s a picture of that CPU doubling every couple of years now here’s the good news network bandwidth is getting faster and faster faster than CPU speed is getting faster and faster isn’t that exciting so the network’s gonna solve all our problems right how many of you go at the net which will solve all the problems only if you vote not so much okay good smart people in the back they’re good um now the bad news is RAM is not improving as fast to seat you so you don’t let every 24 months RAM is 32 36 may be worse than that man okay so the gap between what the CPU can do and what the RAM can do is getting bigger and bigger all the time but then here’s the worst one so I said network was getting faster and faster right but there’s two measures of speed in networking there’s bandwidth bits per second and that’s important but an even more important measure in some cases is what’s called latency the time it takes for the first bit to show up latency is improving very very slowly so I think we’re now at the point where latency is getting cut in half about every five years or so maybe even more than that so that’s terrible right so the gap between what the CPU can do and what the network can do for small messages is getting worse and worse and now here’s the really bad news a software software is hardly improving at all and you know why because it turns out writing fast software is very hard to do very labor intensive and it’s literally the exact opposite of what you were taught in any programming course you ever took in programming courses we teach you have little pieces of data and do little actions on them if you want to get high speed you need to have big chunks of data and do big actions on them and then who okay by a scientist gene sequencing right yeah a DS equals a gene sequencing is improving by a factor of 10 every 16 months there’s no way the computing can keep up so the big problem that the gene sequencing folks are facing is that they can’t get the computing they need right that is the fact that you know by the time you install the machine there’s a new one come along let’s maybe it’s obsolete but leaving that aside enormous amount of data so here’s a picture to show you to illustrate Moore’s law so in 1997 the very first machine that was capable of doing a teraflop okay trillion calculations per second named al B cost well nine figures I don’t know the exact dollar figure but a few hundred million I guess and fill up an entire room it was owned by one of the Department of Energy in 2002 we installed our first Buster supercomputer here at O you I think you could do a trillion calculations a second just over and it cost let’s say very high six figures right so the price had come down by a factor of a thousand in five years that’s not too bad by 2012 10 years later you could buy a card this big that you could pop into your workstation we do that really calculations aside and late last year both AMD and Intel came out with chips individuals who chips that will do a trillion calculations a second take a moment and Wow on that right so the course of 20 years we went from a room and hundreds of millions of dollars to a single chip the costs I don’t know I may be low four figures I guess you need mid for leaders that’s just amazing okay I’m trained by the town but let me just say this real quick supercomputing is hard

supercomputing is a big pain remember we said would it be good to do a problem on your laptop takes a month or better to do it on a supercomputer in an hour well here’s the hidden secret about that it’s something I’m going to charge a million dollars per minute to use the supercomputer it said it’s going to take you six months or more to rewrite your software to run in an hour on the supercomputer is it worth it it depends thank you excellent answer what does it depend on you’re gonna need to solve it later right yeah ah say again you’re gonna need to solve the problem many many times if you’re gonna be solving that kind of problem many many times that is it worth the six months or more that it takes to rewrite the software then it’s worth it if you’re only going to do it once it’s not worth it on the other hand is there such a thing as run one software and realize how have you have ever taken a programming course okay so after you’ve finished the programming project in the course did you ever look at that software again no that was what we call run once code you wrote it you tested it you got it to run you ran it once with the test data and then you threw it away in real life is there such a thing as runs code no because as soon as you run it you show someone your boss your colleagues whoever and then they say oh that is fantastic I want the following 30 features tomorrow morning right and it follows you around forever you never escape the software you right okay so it’s a lot of trouble is it worth it well so it’s good that supercomputers get you to solve the same problem faster the thing that you used to be able to do in a month you can now do an hour that’s a win but that’s the trees that’s not the forest the bigger win is if you’re willing to invest the month now you can solve a much bigger problem because if you can solve the month long problem now in an hour then in a month you can solve a problem I’m going to do this 24 hours a day times thirty days in a month a problem 720 times bigger in that month and doing the bigger problems is the big win it turns out that the bigger the problem you can do the more money you make or the more papers you publish or whatever your value system is doing the bigger problems is gonna get you more of that so it turns out that’s the win the other win many of your students by the way okay so I’m gonna use undergrads because the numbers are a little simpler so let’s say for an undergrad already but is anybody here an undergrad okay yeah you undergrads good okay undergrads to one significant figure how old is a typical undergrad that’s to one significant figure good to one significant figure what’s the life expectancy the u.s. oh I wish 80ish okay so how much time do you have left 60 years by the way for the record it goes like that okay they don’t believe me yet but they will yeah so you got 60 years left computing speed doubles every two years how many doublings will you experience in your lifetime the rest of your lifetime you got 60 years doubles every other year 30 doublings what’s 2 to the 30th you’ve memorized your powers of 2 to the 30th right no okay anybody know two to the tenth no we’re not Google anybody know two to the tenth is 1024 which two one significant figure is well just two significant figure so just one thousand okay so if two to the ten is a thousand was two to the twenty what’s two to the ten times to the ten a thousand times a thousand thousand ten thousand is a million okay so then let’s two to the thirtieth the thousand times a thousand tons of thousand a billion so how much faster so thinking about the computer you have today and the computer that’s going to be on your lap today you’ve got well today before you die the day you died something useful but the day before you die the computer on your lap how much faster will it eat compared to the one you have today say it again a billion times faster can we possibly imagine what we’re gonna be able to do with a billion times more computing speed and capacity than we have today all we know is it’s gonna be amazing right supercomputing doesn’t get you to the factor four billion but here’s a rule that’s very important whatever happens in supercomputing today will be on your desk or your lap in ten to fifteen years and by the way in 30 years it’ll be a cell phone right so my cell phone is the fastest supercomputer of thirty years ago some parts of it so supercomputing what it gives you is it gives you a peek at the future it gives

you a look at ten to fifteen years from now because whatever we’ve got in supercomputing now that’s gonna be your laptop fifteen years is it useful to know the future is it easy to know the future in real life how many of you are good at predicting future I love you are terrible at predicting future good your normal human beings right predicting the future is hard predicting the future of technology I don’t say it’s easy but it’s easier because again and again and again we’ve seen that whatever is happening in supercomputing today that’ll be your laptop 15 years from now okay so it gives you a competitive advantage because you know the future of technology because the problem with the future is the future comes too fast and in the wrong order now it’s