Cray Research - A Story of the Supercomputer

supercomputers that seem are designed had a tremendous impact on our everyday lives the safety design of cars crash testing exploration for petroleum certainly the design of military aircraft and then later the design of civilian aircraft we're all areas in which supercomputers had a tremendous role you welcome to the triple falls Museum of Industry and technology CFM IT is dedicated to the history of industries in Chippewa Falls and also to provide a showcase for current industries in this video you are going to learn about and hear from some of the people who played key roles in the history of super computers in Chippewa Falls my name is Dave frosh I've been associated with one form or another of Cray since 1972 I during most of that time I was corporate secretary and chief technology Council held a similar position later with Cray Inc successor company as you will hear cray and Chippewa Falls took different corporate forms from 1972 until 1996 Cray was cray Research Inc a separate company from 1996 until 2000 Cray operated as a division of Silicon Graphics in night in 2000 Terra supercomputing of Seattle bought the Cray assets and became Cray Inc cry Inc continues to to exist in Chippewa Falls a significant part of the Krait story in the Cray history involves Seymour Cray himself however there were many other people who played key roles some of those people as you will learn had employment positions with all forms of Cray throughout its long history in Chippewa Falls hello my name is Jim Mann Dallaire I'm a docent here at this Chippewa Falls Museum of Industry and technology and we're going to introduce the story of Seymour Cray in Chippewa Falls today we are at the exhibit that we have here at the Museum Seymour grew up on the west hill of Chippewa Falls and here he is on Dover Street Seymour had a sister Carol and the two of them spent their summers on the upper end of Lake with soda when seamer was 10 he built a device out of an erector set components that converted punched paper tape into Morse code signals and that's the Morse code signal that we have on the wall there here in the bow and Lomb Science Award and taught physics after the teacher left for World War two-seamer attended the uw-madison under an ROTC scholarship program but was drafted in the US Army in 1943 serving as a Code Breaker in the Philippines at the end of the war after graduation he was repairing televisions to make ends meet he looked up his college professor for a job lead and was told to check out the work going on at the unused World War two glider Factory he did an engineering research associates hired him while at engineering research associates he was designing computers to break codes for the government in 1958 when ara then Remington Rand decided to stop producing scientific computers Seymour left the company and joined with several others in forming control Data Corporation so a little bit of a background on myself I was in the Navy from 1951 to 1955 and was fortunate enough to attend electronics school when I was discharged from the Navy I was looking around for work and I saw this ad and the lad in the paper about a company called er a minute that I was hiring engineers and technicians to work on computers and I answered the ad and I was fortunate enough to get a job as a electronic technician working for engineering Research Associates at that time they were making computers and I frankly didn't know anything about computers while it controlled data seamer did not like the constant interruptions and interferences from salesmen therefore he convinced control data to to bill set up a new laboratory in Chippewa Falls along the chipra River in 1962 Seymour was drawn back to Chippewa Falls to build a new laboratory on the banks of the chipra River as well as his new home he also but came back home to raise filled children here where he it was where his parents still lived the museum has Seymour's desk that he actually used their control data what you see on the desk today is the boolean algebra that Seymour generated to describe the computer we also have a slide rule which seamer is very proficient at and you also see the clock that was on the upper corner of the desk and this is a mechanical clock that buzzed whenever it was running no the powers-that-be in mini and yeah in Minneapolis had decided that you know Seymour needed a secretary he needed somebody to answer the phone I really wasn't that interested in meeting him or anything I mean I wasn't that impressed I didn't know any better yeah but so I came for the interview and it went alright because he he really you know he didn't want a secretary under foot it was he was just obliged to take someone and you know apparently he thought I was alright he offered me a I can't remember what the salary was but it wasn't as much as I was making and I remember him saying whatever the figure was I think I think he said how about you know this figure and I said how about and this was a higher figure so he said all right so I went to work for him but I remember him saying to me would you call a meeting in the lunch room so I told everyone and we went to the lunch room at the time and he stood up and spoke what he had to say very minimal amount and he just turned around and hand walk out I mean he was painfully shy I would say that Seymour's first computer at control Data Corporation in the Twin Cities was the 1604 which was a of 48-bit computer that had 32k of five magnetic core storage CMOS next computer at control data in the twin cities was the 168 which was basically a scale down from the architecture of the 1604 it had used 12 bit words and had only 4k of magnetic core storage basically this is the whole thing in in the desktop had an optical tape reader which was a good thing at the time at a tape punch and the display was Nixie tubes vacuum tubes that could display numbers and everything all the control is available right here a friend of mine checked out one of these machines voted all of the diagnostic programs into the core storage crated it up delivered it to keep the nav roll took it out of the crate plugged it into the wall ran all the diagnostics and the install was done so it was there was a very convenient portable computer this is your your first laptop computer this is a 6,400 computer which is based on the 6600 which came first the 6500 is yet another variation developed from scaling the 6660 600 wisely the original supercomputer as you can see that the modules are the same the memory is in the same location the logic modules are pretty much the same architecture same hardware and the same wiring on the back which was all done with color-coded 26-gauge twisted bare wire and after this 6600 was completed work started on a 7600 and again the goal there was to make a system that was at least four times faster than the 6600 we completed the development of the 76 conard we're standing here in the the interior of a control data 7,600 computer the reason that was built in this sheep the C shape is to try to shorten the wires because the one of the limitations on how fast the computer could run was the length of the wires from one module to the next and after delivery of the first systems and work started rather quickly on a follow-on system which was to be called e 8600 it was a pretty steep step trying to make four times the speed performance by eighty six hundred over that of a seventy six Center partly because the technology just wasn't there to allow us to do it quite easily so we had it's quite a bit of difficulty trying to build the 8600 this is the control data 8600 but the 8600 was the last computer that Seymour designed for control data but on the on the test bench we were executing 8600 instructions on the day we found out we've been cancelled at the same time control data was experiencing financial problems and one of the efforts I believe to reduce cost was too perhaps cut back on the development cost of the 8600 that didn't sit very well with similar because we had it actually a very small budget and it was some time in that time period where Seymour felt that he would leave control data and start his own company with a number of friends of his Frank Mulaney George Hansen Knoll stone pet Dirk and perhaps some others there started a company called grey research that was in 1972 and development then rather quickly started on building a new machine that would have been significantly faster than what was available in a marketplace and of course that was the 7600 as well as some top competitors from IBM and a couple one or two other companies and the work that was started then was basically that of the cray-1 so this is a cray one in fact it's cereal one the very first machine that was built by Cray research it's beautiful its iconic and it turned the entire numerical computing world upside down and there's one in the Smithsonian so how did that happen we also had advancements in the we also had advancements in wearing and we were able to make wearing a lot smaller and were able to shrink the Machine down good thing that always comes to my mind about Chippewa Falls and Cray particularly in the early days are the people mostly women that put the christ together they wired him up this is in the old days before micro processors and all this microscopic technology that we actually had to make machines by hand and these ladies would knit the Machine together in a sense they were doing the wiring it was very dense very intricate there were thousands and thousands of individual connections that they had to make by hand and they would solder the connection and it took them probably three months to put the machine together fact les Davis explained to me he's the guy that built these machines and I'm sure you're going to be talking to him any case he explained to me at least in the early days that that the production process was about six months for a cray cray one like the one I'm sitting on and he said but it was really three months of assembly work and then three months of fixing the errors that had been made in the assembly work and that was very hard very difficult to do to diagnose the errors and fix them well these ladies got so good that they could make this machine in three months absolutely error-free no errors at all so as soon as the machine got plugged in at work you know Sonya Gardner I started a crane November 1st 1976 however I worked at Cray from about 1974 in the wiring department I was a temporary worker until 76 I'm Yvonne Branstad started working for cray-2 through tennis two of my tennis one of my tennis partners was Jer Seymour's wife Geri and her business partner sandy sandy McCausland it was like a little cottage industry in that we all worked at home my name is Shirley Gormley and I started at great November 1st of 1976 and we started in the wiring area and then I went on into the cable area shortly after that and made all the cables that powered the systems and my name is Jim Mann Dallaire I started with Seymour in 1974 January and I was hired as the early checkout engineers on the cray-1 and then I stayed with the company through the STI split and then left STI in 1999 bring on the first chassis donut the little building down here and they brought that thing in and they laid it down in here was I know five million of these little blue and white wires hanging all down they said now you've got this crochet hook and you've got this book that tells you you've pulled from point A to point B you put a solder sleeve on it and you solder it together we're looking going oh my lord how are we ever going to do this but as I can remember my girlfriend and I we started after we got used to what we were doing when we got up to doing I forget if we got up to like 500 little wires in a day that we pulled out and of course then you had to make sure you didn't make a mistake because if you did you had to take them all apart and find the right one so after a while everybody could really go pretty fast you could get your speed up you did the little short wires first and then they gradually got longer and then pretty soon the mat would be I don't know six eight inches deep and then you would have to take you would have to take your fingers and dig down in this wire mat to be able with and look with a flashlight and be able to count the little wires on these bars so that you could get your crochet hook and pull out the right one so after a while it got to be quite a challenge but as we can all tell we all did pretty good well there are probably two if you want to call it that teams at the time one was maybe even three one where the mechanical engineers headed by Dean Roush the other was a software group headed by Bob Allen and then the third would be the electrical engineers that I was a part of and Harry Runkle and others were some of the engineers that were involved in doing the electrical testing and design of the machine so basically what we did was we kept basically the same architecture and we went from a 5-4 gate to a 16 gate array so we were able to put more punch into into into each chip which means we could put more into each module the other thing we did is we took what used to be a single module and we put two of them together and put connectors between them so we could have more conversation happening at a closer in a closer range other than that from a mechanical standpoint which was my expertise the cold bars which used to be cast with pipes in them became now a solid piece of aluminum and we could gain some efficiency that way we did the same thing with the power supply cool place the power supplies in the in the XMP were identical to the power supplies in the cray-1 the only thing we did is we encase them in a block of aluminum ore and get more power out of them instead of having one CPU and four columns like we did in the cray-1 we now can in those four columns put two CPUs so we could have double the processors and plus gain the speed from having everything closer and we also had a gain in memory the plates the coal plates were a little bit different in that we put little little um copper pads on the boards underneath each chip to help cool it to the coal plate it spread the heat out if you will we also had to have a program where we when we put the chips down we had to make sure that there weren't a whole lot of hot chips in one place so they had we had a software program that we just put into place and we had that going so that we could put a lot more power under that coal plate and get things done at the same time we were doing the XMP we also added the solid-state disk we didn't have that before we just had the i/o which is back over here and basically we wanted a fast faster solid-state solid-state death we had some faster memory quickly that we could - and back every wanting to go faster and furious Seymour who was always determined to take technology and shrink it down and the crate too is a perfect example of that Seymour decided that he was going to have to use a different cooling system though and one of the unique things about the crate too was is that he came up with a liquid system using what's called liquid floor inert and floor not really can think of it as liquid Teflon it will not react with anything but yet it will be able to cool the machine by passing the fluid across the modules and Seymour designed the modules that take advantage of that by flowing the fluid over the tops of the chips in between along sideways across the module and on the front end would be connections for power and on the back end would be connections for logic we had to find a place to put the computer put the fluid when we were working on the machine and so we came up with a reservoir that you see in the background here this was the first model of the reservoir and this machine would and then we also had to be able to pass the fluid through the machine to cool the modules down and that was done with these month with these cabinets that you see here and what they contained was heat exchangers that would take cold water from the customer and cool down the floor inert and return it to the machine what we have here is the ymp this was a follow-on to the XMP things things went pretty quickly into a larger scale integration here if you look if you look you can see the module down here we have large chips that are now 2500 gate arrays so we've significantly increased what we can do in a chip and I think there were eight CPUs in here and then the memory top and bottom beside it if you look at both side you'll see that the remnants of an ILP and the XMP same one we had on on the XMP I mean the i/o and the SSD sorry so those those really weren't changed at that time by the way the when we did do the SSD that was done at the same time XP that was actually a separate engineering project not part of the XMP that Jim Sun debt actually I ran that program when that was introduced this instead of being cold bars we actually ran floor nerd through a coal plate which was a brazed assembly so that was basically two pieces of metal with fin stock in between braze together and we ran the fluid to no beyond that we had I don't know if you can see see this but there's four boards stacked up here on two cold plates so again we're talking between the cold plates so that's why we're able to put so much in into into one one module and actually have a whole CPU when we put the CPUs and they're actually quite heavy we actually had a thing to hold them on to put them in so that we didn't bend connectors the connector technology on this if you see the connectors are along the side was something new we had to come up with because usually we just plugged in in the back in the XMP and cray-1 and create two here the connectors were on the side obviously you have to slide in and then somehow move something ters and we had a little pointed ramp if you will that we would push in there and push all these tabs over one at a time all the way through and we to unplug it you just turn the rod around the other way and push them back so it was prepped a unique little design yep now we've got through the ymp and that was really a fascinating machine for its time and Cray wasn't about to quit so what went on after that you know you had brought up the C 90 and Cray 3 and we've been talking about sort of the old days at Cray when we were single focused on just the biggest computer in the world and there came a time where we reached a turning point and it might be of some interest to talk about that because one of the things about Cray research is the name is Cray and the name comes from a person Seymour Cray and he is you know fundamental obviously to the company but fundamental to the business itself he was involved in the early early days of computers just after the Second World War he worked with people like John von Neumann he's in many ways as important to the computer business as Thomas Edison is to the electricity business so that was always a dominant factor in the company and yet those of us in the company other than Seymour and Seymour - wanted it to be more than that more than just one man in his work we wanted and had a prize that could carry on and has carried on actually since those days so I knew at least all along that we had to somehow be able to make computers without Seymour Cray or beyond Seymour Cray and until we did that it really wasn't a done deal it wasn't complete and we did do it shortly after the Cray one was out last led a team with a fella named Steve Chen on it that that came up with something called the XMP and then that became the most powerful computer in the world but at the same time Seymour was working to and then he came out with something called the crate - and that became the most powerful computer in the world and then there was a group I think it was Steve Nelson that led a part of it at least that came out with the ymp and then that was the top computer in the world so that was really fundamental very important for the company to be able to be beyond Seymour if I can put it that way but in the middle late 80s we ran into a problem there were two projects going as there had been all along one was called the sea 90 the other was called the Cray three as you can imagine the sea by me was being developed by other teams inside the company and the crates 3 was being developed by Seymour and as I recall the original schedule that the Cry 3 was to be completed before the sea 90 and then the sea 90 would come out say two years later and frankly Seymour ran into some difficulties with a Cray three and it it got delayed and the sea 90 kept apace and pretty soon we had a situation where the machines were going to be overlapping they would come out at the same time and at that point the business as a whole kind of slowed down and we didn't we didn't have the resources or the need to have both machines at the same time it would confuse us it would confuse our customers and I was left with a dilemma what to do one of the projects had to be cancelled and then put all of our resources in the other project we didn't have the resources really to do both and to do them well so I thought about it a long time trying to figure out what are we going to do and if I chose the Cray three Seymour's project the signal would be clear okay Cray is cray I mean it really is cray and unless you're on the Cray team Seymour's team you're vulnerable because look they cancelled the scene iv when would the Cray three that would not be good so let's look at it the other way well we could do the c90 well how in the world can I do that I'm going to cancel the project of Seymour Cray I mean his name is on the door I can't do that so it was a real dilemma and what we did is decided in a way to do both but not in the same organization so I went to see more and I said I know you're making progress I know it's going to be a great project we can't do both of them so here's what I'd like to do tell me how much money you need to finish the play three and he said well I think about a hundred million dollars I said fine so here's what I'd like to do I'm going to set aside one hundred million dollars and give that to your project we're going to organize your project as a separate entity and we'll spin it off from the company and you can take the hundred million dollars and that the technology are working on and make a company call it what you will and he wanted to call it Cray so I said fine and that's became Cray computers and he moved to Colorado and then the c90 stated Cray research and I stated Cray research that seemed more appropriate and we came out with the c90 but it was a very traumatic experience to have to split the company into like that very very fundamental then things changed when obviously after the cray-1 crate - the natural progression is cray three I think this is just my opinion was what I think happened and in its I'm sure a lot of other people will disagree with me but part of what happened to see where I wanted to make a big step forward again and we had the XMP program going on at Cray research and it had actually outperformed the Krait - great - had its niche where it could work well but the XMP did some other things Seymour took a what I call a big step building the Cray three and I think it was the downfall of the Cray three with technology every time you go forward what's important as you step out and take advantage of the technology that exists at the time but you don't step out too far that it's not practical you can't make it work and in my judgment see more stretched out too far the first thing he did with the Cray 3 is he chose a new technology gallium arsenide gallium arsenide had the potential of really speeding up all quickly circuits process information and it wasn't a technology that was available out in the marketplace we actually started our own IC Development fab at Cray so we could build a circuits and we successfully built circuits for it Seymour used both gallium arsenite which was a technology that was essentially unproven yet and he coupled that with putting them on printed circuit boards that use a lot of gold a lot of very fine ink reading intricate technology to bind things together to bond things together and in in the end what happened is he had a package that was pushing too many things too far the my understanding of the early Cray 3 is it they got one built they shipped it to a customer but it had too many problems problems in the sense of being intermittent because there were too many connections that were done in a fashion where you just couldn't build it reliably enough to do what you wanted to do I think had Seymour changed his plan early on it's that we're going to do it in gallium arsenite we're going to go to a little larger array we're going to simplify the designs and we make the packaging easier a Cray 3 or another product out of Seymour probably would have been successful so it's 1987 and the ymp and it's MEC maximum configuration of eight processors was selling well and the cray 2 is moving into production and in the background not known to the general public there's a project that's been going strong to build a follow-on computer to the ymp to follow on that computer to a more powerful version but something happened the original project create that computer was cancelled for a variety of reasons and along with that canceled project the company lost a number of very talented people so very quickly we had to play catch-up and that was the birth we know exactly the day and the week when we started the c90 project and worked very fast and hard to be ready when the marketplace was ready for that next machine to replace their ymp this computer would turn out to be a 16 processor supercomputer the ymp had eight processors maximum he had some other nifty features in it we had double double-wide vector registers so we could pop out more results per second we had larger memory in fact the word large describes the c90 to be honest we were pushing the envelope of what you could physically do the c90 turned out to be colossal in size holding all these processors with all the cooling equipment even though it was large and it was expensive it was a very successful product that many of them were sold 90 was farmed as skip pointed out when mr. krei went to Colorado to continue his efforts on the Cray three as an independent company and he was the group that was the state behind were the primary the heart of his design team and it was an interesting sidelight that he would come back every month at life after I was hired and he would interview these people which was he was given permission as a part of the package and to try to induce these people to move to Colorado Springs and there I think the magic of Chippewa Falls comes into play in my mind is that is that these people could see they could build a machine there was a tugging pole of Chippewa Falls on one end and the inventor of the supercomputer on the other end and the idea that the relationships were so tight that there was a very heart-rendering set of decisions it had to be made with this design team but the heart of the previous design team stayed in Chippewa Falls and went to work and effectively and very well designed the the t90 machine which took about four years from the beginning to the time it went into manufacturing the difference is on the t90 machine from all the others was kind of a combination of of the best I'd like to think mean part of it of what went on be with the other two we we did use cooling that was similar to the Cray to Oni we accelerated the fluid across it because the chips were were more dense in logic and therefore higher power and we also use the attributes of the printed circuit board facility that was going of trying to achieve the objectives of the Cray three and we backed them off a little bit and we use the attributes of the c90 and and using three dimensional packaging of the board so that you could get between any point on the board and we put everything on one board and put the circuits on two sides of one board there the other thing we did which scared some of the menu during people the wonderful ladies who did all the wire Metz is we tried to do a wireless machine and we did that with some pretty exotic connectors that were struggle for a while but very creative people one being Nick Krajewski who was really the backbone of the of that area worked worked it through so we could reliably connect our circuit boards to our memories and our boards to each other our computer boards through our memory boards and we will achieve that these people were put to work on assembling the boards and because that was another trick to put 150 chips or so on two sides of a board around yeah two sides of a single board and then bring this thing to assembly so micro miniaturization and the attributes of the assembly people really came to play there you we were often running with the c90 but there was another set of stimuli that were coming at us and that was the microprocessor market these small single chip computers and we had them those days in our personal computers and it became clear to people our customers particularly some of our government customers that these little micro processors like the processors that are in iPhones and in iPads and in Android cellphones it became pretty clear that they had a lot of room to run well as luck would have it we were approached by a company that was Digital Equipment Corporation out of Massachusetts and said hey we've been working on something you guys might be interested in it's a 64-bit processor and I like that we have some that are working we've got some working ones and we thought we should talk and so we did and that was the beginning of the Cray t3d project and we were proud to present this product to the marketplace you SGI acquired cray in 1996 and at that time Cray had two products the t90 which was the product that curry had developed that was the traditional based on the traditional vector processor and also had the t3e system which was the second MPP system for high-performance computing that Cray had developed that system the t3e system specifically became the standard for MPP high-performance computer systems and it was the standard for approximately 10 years I would say in in 1998 the Cray business inside of SGI was developing a new system that basically blended the t3e technology and the traditional vector processor technology so this was a cost shared development between the government and SGI so then in 2000 along came Tara computer and Haier computer purchased the Cray business from SGI there were three things that were of real significance and importance here one of them was the Cray name honey so they purchased the cup the name from SGI and they changed the name of the company to Cray this was very meaningful because Craig was a name that was well known throughout the high performance or the super computer industry and also a part of that technology transfer was the the contract with the government and so we had funding for doing development so we in some ways we were a weird startup where we didn't have a lot of product at the time we had a lot of people we had a lot of resource at Cray Inc the very first system that create introduced create ink I should say introduced and was a very successful system the the third thing that was really critical in the acquisition was the people and it's always the people that make up the company at the end of the day so today if you look at that that microprocessor that the MPP systems are based on it continues to ride the integrated circuit technology curve and if I were to look at a chip today using the most leading-edge technology it can have up to 16 individual processors in a chip and this and these are at a performance of approximately 50 times the performance of the cray-1 from back in 1976 when that was introduced so a phenomenal improvement in just individual processor performance and now this is on a chip versus in a system like we are looking at right here and so for Cray that differentiation is in the interconnect technology the memory system technology and the software I mean that if you took a single application and you were able to run it on this system that it can scale to the point that it can be running across greater than a hundred thousand processors on a single application and so if you go to a hundred thousand processors you want to have the performance of a hundred thousand times it versus running on a single processor so with these with these capabilities and with this differentiation Cray has been able to create Inc now has been able to maintaining a leadership position in high-performance computing as I look back to the late 80s and specifically in 1988 creig announced the system that could perform on a given application a real scientific engineering type application could perform at one times ten to the ninth calculations per second and since that time the performance improvement on these systems has gone up a thousand times approximately every 10 years and when that performance or that performance increased has been realized by Cray Inc prior to anyone else being able to demonstrate that capability so that started with in 1988 that was the ymp system in 1998 that was the t3e system and now today inside of Cray Inc in 2008 that was our MPP offering that demonstrated that before anyone else in the early days of crit was something called the cray style that described ideal culture yes yes yes it was really like that it was you know the way I think about that is it was selflessness and it was about not taking yourself too serious taking what you're you do serious and I believe that's what drove us I still worked there today it's uh you know going on 38 plus 40 years and it's still a great company it's still friendly people you know we only have about 200 people in Schiphol Falls but I think we've got about 800 worldwide so the rest of them are overseas and we still get to travel and you know it's it's still a great place to work and I think it's it's good for triple falls people didn't want to leave triple falls once they lived here it's a great place to live and Chippewa Falls with where they want to be and even to this day there's a lot of companies out there that were started by some of these people and there's more people employed in the Chippewa area than create ever employed because of that yeah and there's stupid and there's there's people out there with PhDs that are teaching school right now just because they don't want to leave the area in this figure when weren't here anymore but the people really made this happen and there's there's some great people still in the companies that are left here we hope you've enjoyed learning about the Seymour Craig collection at CF MIT and Chippewa Falls and we certainly hope you have a chance to come and visit you

Channel: ThisMakesMeThink

Views: 45,192

Keywords: Supercomputer (Invention), seymour cray, cray, nsa, amd, intel, chippewa falls

Id: wn03wn3k47Y

Length: 48min 10sec (2890 seconds)

Published: Tue Jan 07 2014