in many ways the town of Livermore California was the ideal getaway destination it was quiet dry sunny and just an hour away from the home drum of 50 San Francisco during the war it was home to a military Outpost the climber and remote location were perfect for training new pilots locals would often see gliders soaring in the deep blue sky most expected the Bas to close after the Allied Victory and indeed it was abandoned shortly after but it soon became clear that the end of the war would not signal the end of military development far from it a new research lab was set up in the small town a branch of the University of California radiation lab its aim to create the most destructive weapons mankind would ever develop you see for the United States government ensuring the country was the world leader in nuclear research was Paramount the development of the Manhattan Project at the sister Los Alamos National Laboratory in New Mexico had ended the war in the Pacific in an instant the progress on the next generation of nuclear weapons had slowed and researchers felt that a competing lab would help speed up advancements leading the lab was Sydney fernbark a master physicist who had studied under the wing of Robert Oppenheimer fernbark would put an emphasis on using computers to solve the complex equations that underpinned Nuclear Physics while they were still in the infancy there wasn't much Choice testing nuclear weapons was hugely destructive extremely expensive and politically dangerous in less than 10 years detonating nuclear weapons above ground would be prohibited mechanical computers using Wheels cogs and relays have been utilized the development of the first nuclear weapons a few years prior at the time the technical limitations meant the equations had to be solved in one dimension however the hydrogen bomb was a far more complex weapon and one dimension simply wouldn't be enough the addition of the second dimension increased the number of mathematical operations required exponentially and as the difficulty of problems continued to increase only one thing mattered speed the first electronic computer of the Livermore lab was the univac 1 it took 30 days to assemble major components were made out of fishing wire and it would break down all the time but the machine could crunch numbers thousands of times faster than anything else had existed and that was all that mattered As Weapons development continued faster computers were acquired at break neck Pace the near unlimited budget that the atomic energy commission had given both the Livermore and Los Alamos labs created the opportunity of a lifetime for computer manufacturers the labs needed the fastest machine possible and as soon as it arrived another that was even faster the question was who is going to build them as the war ended and the need for large numbers of military recruits disappeared many of them had to make a decision about their future most wanted to return to their pre-war life working on farms or in factories for some however the last four years had changed their outlook on life and wanted something completely different at seesaw the US Navy's code breaking division servicemen faced this very choice they were some of the first people in the world to use a mechanical computer each day they would feed the machine's encrypted radio messages sent to German ubos to decipher when the operation came to a close some began to wonder about the potential use cases of computers outside seaw one individual of note was Bill Norris a lieutenant commander who had joined the division four years prior Norris came from a farming background but was technically minded and had worked his way up to an engineering position his experience with the codebreaking apparatus led him to think about the wide range of applications for a computer and wanted to set up a company to explore this idea further he was not alone on this Commander Howard angstrom who before the war had worked as a professor of mathematics at Yale was eager to pursue this plan surprisingly HPS at the Navy were fully supportive this was for two reasons they knew that most of the engineers would leave the Navy and it would be easier to subcontract the work to the companies than start fresh with a new internal team also due to the War ending their budget had been cut significantly and they were unsure whether they would be able to fund the development of a new codebreaking machine themselves thus Norris angstrom and a few other Navy recruits had a plan to leave the Navy and set up their own computer company the first thing they needed was funding to Norris and angstrom the commercial use cases of the computers were plentiful American Airlines could automate their flight reservations Chase could instantly store trades of shares on the stock market and General Motors could immediately see the stock levels of all their dealerships but all of their previous work on computers was classified so they couldn't give any details about how the machines worked or how reliable they were as a result these stuffy bureaucratic boardrooms of corporate America didn't share the young Engineers visions despite the Navy support it was simply too radical of an idea there was one contact who was intrigued by the concept an investment banker called John Parker he was the owner of a large glider factory in St Paul Minnesota which during the war had made thousands of gliders as part of the war effort Parker had gained large wealth in the process but like many contractors across the United States the orders had dried up and the business was quickly bleeding Ash as a former Navy officer himself he was more understanding about the restrictions Norris and engstrom faced when discussing their ideas and having exhausted all other funding options the Navy hups repeatedly emphasized just how important it was that this new company was set up Parker was skeptical but he was smart he wasn't really sure about the technology but he knew that this company would have lucrative contracts with the Navy and and with the Glide of business on the verge of collapse he had little choice so he agreed and together with Norris angstrom and a few others set up a new company engineering research Associates to work on this new technology the contract ra had with the Navy allowed Engineers total freedom to design and build equipment however they wanted while the company would take on a variety of projects including data recorders and instrumentation the the main source of funding was from the government who would pay however much it cost to develop a computer this flexibility created a hot bed of innovation which manifested with the group's first computer The Goldberg computers of this time used vacuum tubes for much of their logic and memory and The Goldberg was no exception their use allowed the computers to process hundreds of thousands of calculations per second but their reliability was poor if one blew while the computer's running which wasn't uncommon then often the machine would have to start from the beginning to prevent this Engineers wanted a way to store and retrieve information without using a vacuum tube to do so they've placed a metal drum coated in Magnetic Tape inside an enclosure with Magnetic greed rad sticking out by rotating the drum data could be read from and written to the Magnetic Tape in The Goldberg machine it was incredibly slow taking nearly 10 minutes for a full rotation but it could store more than 40,000 numbers the company hit a snag with their next computer demon while the drum memory technology evolved the project was rendered useless just before completion due to the Soviets changing their messaging format due to the high cost of producing purpose built machines Engineers realized that the only way forward was to build a general purpose computer with the program stored on the magnetic drums this way if something changed the machine could be reused and the program simply Rewritten so for Ra's next machine atas they did just that the magnetic drums used in these machines were truly revolutionary so revolutionary in fact the concept would become the basis for the hard disk drive that inclusion brought a host of advantages for one the machine could Access Data very quickly without having to read yards of tape the drums are also nonvolatile meaning when the machine was turned off the data stored on the drum remained and it would be there when the machine was next turned on most importantly of all the user could write any program they wanted to run it and make near endless modifications almost instantly ra knew that the magnetic drum technology would give their machines a huge advantage over their competition in said about modifying the outlas into a commercial machine the finished product would prove to be popular and its design would last well beyond 10 years the Navy contracts and the success of the commercial products grew the company rapidly from 40 to 400 staff in just 5 years however despite this the company was in a dire Financial State and was bleeding Capital it was 1951 and the market for computers simply wasn't well established in addition the company was coming under heavy increasing legal pressure due to the questionable almost monopolistic Arrangements Norris and angstrom had negotiated with the Navy the company would be acquired by typewriter manufacturer Remington Rand in 1952 who wanted to increase their presence in the computer Market many in erra were skeptical of the purchase but Remington Ran's prior acquisition of univac who made the world's first commercial computer showed they were serious ious about the machines however the differing approaches of the two teams almost immediately cause tension unlike ER who had spent years pitching their machines to businesses the univac machines were built by academics who had little concept of the world outside the lab they would Hemorrhage cash often spending more than five times the allocated budget developing a machine which would usually suffer from poor reliability their first machine the univac 1 would only run for 10 minutes before component failure in comparison the ra Atlas could run for 30 hours without issue the original plan was for ra to focus on the Navy machines and univac to focus on the commercial side but when Remington Rand was purchased by sper Corporation another Electronics manufacturer the two divisions were combined under the new sper Rand management the new unit would retain the univac name as it had better brand recognition than erra but with the merger the funding for projects rapidly dried up and tensions between the two teams who were now forced to work together was reaching boiling point no one was more frustrated than Bill Norris who had founded ra 11 years prior he'd seen his company and the engineers become a shadow of their former selves and he'd had enough Willis Drake a senior erra engineer shared Norris's frustration he' witnessed years of incompetent management and felt that Remington ran would never be able to succeed in the increasingly crowded computer Market they both knew that they would get no support from the Navy and would have to raise funds in order to start a new company to find this Norris Drake and top ra engineer Frank melany met up with ex ER worker and Harvard Business School graduate Arnold Ryden to discuss funding options Ryden suggested a unique approach simply sell shares in the street convincing people to invest based only on a business plan this seemed risky but Ryden explained that a loophole in Minnesota law allowed him to do this while they were still working at remon Rand by doing so they could stay at their old company for as long as it took to gain the required funding Drake liked the plan and left the now monolithic univac division to join Ryden in selling stock in the new company it didn't take long before they had enough Ra's aircraft hanger had never been considered a glamorous place to work but compared to 501 Park Avenue Minneapolis the headquarters of the newly formed Control Data Corporation it was a palace Engineers arriving at their new jobs walked into a dark noisy and pollution filled Warehouse used by local newspapers to store printing paper most workers would have walked straight out but the lack of overbearing management created an atmosphere that was filled with enthusiasm the Navy was well aware of this and to reduce the brain drain at remon Rand asked Control Data not to take employees who were working on key Navy projects the most prominent of which was Seymour cray cray was a University of Minnesota grad who had joined ra straight out of college despite his youth he had the confidence and abilities of a veteran engineer within 2 years at the company he was already making important decisions about computer components breezing through simpler tasks in his first 6 months Craig was assigned to work on the control system for an upcoming machine the ra 1103 the function of this component was to convert the lines of code being fed into the machine into computer instructions it required sending signals to various parts of the machine in a specific order which would often take weeks to work out such a mechanism would seem trivial these days but but in 1951 there were no manuals and little prior research a fresh college grad would have bked at just a thought of Designing such a component but not cray he knew he could do it and his fellow Engineers did too by the time Control Data had been Incorporated cray was one of univ back's top Engineers the Navy had kept in there to finish designing a data processing machine for warships but it was only a matter of time before he to walked out the door the new company was initially unsure about where their revenue would come from spending its first year selling drum memory technology that the engineers pioneered ER large computers which is what the company wanted to develop cost millions of dollars millions of dollars the Control Data did not have but when cray joined as head of engineering he simply told them all I know is how to build computers so I'll do that to try and entice investors cray set about developing a small prototype computer which could be used to demonstrate cdc's Technologies the machine the CDC little character was significant due to its use of transistors instead of vacuum tubes they were far smaller used far less heat and were more reliable when the machine was finished it was an impressive piece of equipment being a useful sales tool and a vital test bed for new technology however Control Data was bleeding money it needed to develop a flagship machine that was a Quantum Leap above the current fastest and on a Sho string budget the design would be based on the unimac 1103 that many Engineers including cray had worked on at Remington Rand many small improvements would be made but the biggest difference was that their machine would replace the vacuum tubes used in the 1103 with transistors having used them on the little character cray knew that any state-of-the-art machine would need to use them their advantages were simply too great the problem was that they were expensive costing $4 each and the Machine would need 25,000 of them face with no other choice he tracked over to a local hardware supplier to purchase the cheapest transistors he could costing just 37 cents each cray was so impressed by the price he bought the entire stock their performance was terrible with widely inconsistent electrical properties that surprised even cray himself but they worked to adjust for the lackluster capabilities of the components each transistor was paired up with another which brought the properties of a pair in line with a good quality part the components would be mounted onto hundreds of small circuit cards 5 by 8 cm in dimension which would plug into a mass of connectors allowing the cards to communicate development was was a slog it seemed like the machine would never get completed the project was consumed by a double whammy of Financial and time constraints employees had to take deep pay cuts to ensure the company didn't go bankrupt and as fall turned into winter there was still no working example but slowly the printed circuit cards would fill up the Prototype casing and by mid 1959 the machine was working at a size of 192 kiloby its memory systems would struggle to hold even a small jpeg today but when the newly christened Control Data 1604 was tested it was simply incredibly fast being the first fully transistorized computer it was easily 10 times faster than any vacuum tube machine surprising even the engineers themselves the capabilities of the machine started to attract investment including a $600,000 injection from an insurance company when the machine launched in 1960 demand was steady but in the Years following this would rapidly increase initial customers were branches of the US government the Navy various research Labs but it was industry that really filled up the order books universities defense Aerospace oil and gas for Control Data each of these sales meant one thing cash profit soared increasing five-fold in just 3 years out of nowhere a small computer company operating out of a newspaper Warehouse had built the fastest computer in the world and its designer seore cray was regarded by many as a Pioneer anticipating the success of the 164 the company had been working on a cutdown machine which would incorporate similar technology but at a lower cost the 16604 was a success for the biggest businesses and research Institutes but for general purpose Computing it was simply too expensive for most Enterprises and management saw the potential for a cost reduced system the machine the CDC 924 would use a 24-bit processor compared to the 48 bit of the 16004 and the memory would be cut from 192 KOB all the way down to 24 however as a result of these compromises The Machine's 1961 launch price was just $180,000 about a fifth of the price of the 1604 the product was well engineered but the 1604 success failed to trickle down primarily the world of business computers was dominated by IBM and cdc's lean architecture made programming business applications unnecessarily difficult less than 15 were ever shipped compatible than 50 604s as a result the head of Control Data Bill Norris decided that their computers should be more business friendly new machines would need to include more instructions and full backwards compatibility with the 1604 even in the early 60s growth was the key to startups and this direction would allow the company to springboard off the success of their first product the idea was to have a lineup of machines to be dubb the 3000 Series that could service both commercial and scientific Enterprises the first machine the 3600 would be a flagship or a placement for the 1604 it would be seven times faster and have eight times the memory all at the same price a lowend machine the 3200 would replace the 924 it would also be 7even times faster and have four times the memory then there would be the 3400 a mid-range machine with the same memory size as the 1604 and the same clock speed as the 3600 but at half the price Cay would leave the initial development of the machines but grew frustrated at the new business focus Direction the company was heading in the ballooning number of instructions and backwards compatibility were a compromise something CRA saw as poison IBM's newest machine the 7030 stretch had been the subject of constant modification from management and have flopped upon release it was a textbook example of how Corporate bureaucracy could ruin a project and cray worried the same thing could happen at Control Data there was no doubt the new machines could be successful but he had little interest in their development he was focused on creating the fastest machines in the world and nothing else crra and the engineers started to design a new system with a performance Target of three times that of IBM's latest the first stages of development were focused on replacing the 37 Cent transistors in the 16004 with new state-of-the-art parts but when this was done the performance was nowhere near as expected months of tweaks and modifications went by without any success facing hurdle after hurdle meanwhile the business focus 3600 wasn't tensions Rose as management started to question the validity of such a risky from the ground up project it was burning through the newly acquired cash with very little to show and visits from Executives became more and more frequent eventually cray had enough he stormed into Norris's office and gave an ultimatum something had to change or he would leave the company for Norris he too was nervous about the new machine the 6600 project as it was now known was going nowhere unlike the rest of the executives however he understood cra's anti-corporate working style and the effects the interference from above was having besides cray was simply too important and engineer he let the man have what he wanted a remote research station away from management Engineers were allowed to join cray under their own free will and with the prospect of working with complete creative freedom many did working in a new office built in cra's hometown of chipware Falls excitement filled the air and progress on the new machine started to accelerate to address the issues facing Minneapolis the engine began testing a bleeding edge transistor from Fairchild made from Silicon rather than geranium the proposed clock speed of 10 MHz 50 times faster than the 16004 has simply been impossible with the old transistors these new parts however were in a different League they offered Lightning Fast switching and using them would mean the 6600 clock speed goal would finally be attainable but with such a fast clock there was now a new bottleneck the wires in this era of computing CPUs have grown increasingly large in order to handle the complex are of instructions that required the size introduce signaling delays the idea that one part of the processor can't progress because it's waiting for a signal from another part to get around this the central processor was simplified to only perform mathematical calculations with other functions such as memory management and IO being handled by other components distributed throughout the system this meant the processor could be small and that meant that the wiring could be much closer together in fact this is often considered the first implementation of a reduced instruction set computer a technology that would become ubiquitous 40 years later using this approach the logic circuits were made 32 times denser than in the 16004 so dense that the heat would almost certainly fry the circuit boards to address the issue crat returned to engineer Dean Rous who had been working on the machine's cooling Rous was an expert in refrigerants working in freezer and air conditioner design before joining Control Data up until this point all computers had been air cooled but seeing the eye watering heat dissipation of the new system a different solution was needed this new approach pumped dupon free and coolant in between each of the circuit boards and then sent the warm liquid to a refrigeration chamber where would be cooled the approach worked making the machine the world's first liquid cool computer the final Innovation was perhaps the most significant in Prime machines the main processor acted as a single unit executing instructions one after another cray noticed that this approach resulted in many parts of the system left idle for example if the instruction was multiply then the circuit's handling addition would be unused the 6600 100 would be different its Central processor would contain 10 functional units each handling different instructions as a result the machine would be able to multiply divide and add floating Point numbers in parallel today this technique is known as super scaler processing and has been common place in microprocessors since the release of the Intel Pentium in 1993 but this was 30 years prior and something no other machine could do the result of these Innovations was a machine unlike anything that came before when initial testing was performed in July 1963 the moonshot targets CR set at the start of the development really were met it was three times faster than the current fastest IBM 7030 five times faster than the CDC 3600 and 50 times faster than the 1604 at the Press launch journalists were astounded that a team of 34 People based in a small Wisconsin Town had created the fastest computer in the world as a result of the incredible performance the next few years would be golden for Control Data it would sell over 16600 $2 million each generating half a billion dollars a year in sales un adjusted for inflation the company had grown from 11 employees to 9,000 in just 8 years becoming the third biggest computer manufacturer in the world it positioned itself as the most agile computer company out there offering a suite of services that no one else could a physical machine too costly you could simply buy timeon one that Control Data owned once a write software but don't know where to start the company would send over programmers to help you need accessories no problem they made them all in house this was a result of CEO bill Norris's desire to stabilize the somewhat volatile financial status that had plagued the company in its first 5 years sales of services and components would help balance out any LS in demand for a specific model or line of computers the company would begin a flurry of Acquisitions to achieve this it completed eight in 1963 and more than 10 in 1964 from the outside Control Data was a perfect example of the American spirit and Innovation that defined the 1960s but from the inside all was not well many of the Acquisitions were not profitable and those that were still built parts of vacuum tube machines desperately needing funding to upgrade their equipment the vast cash reserves that had been built up from 1604 and 3000 Series sales were rapidly decreasing and many in the company began to wonder if this was the right direction cray however detached from management as usual was largely unaware of these struggles he was focused on the 6600 successor the 7600 the success and performance of the 6600 prompted many to wonder whether a brand new machine was necessary 2 years had passed since the product's release and it was still the fastest computer in the world but the industry was Cutthroat especially for Control Data who despite the Acquisitions was still far smaller than IBM with this in mind cray was ambitious the 7600 would be 10 times faster than the 6600 to achieve this a breakthrough design would be required one of the Essential Elements of the 6600 design was its parallel functional units allowing the machine to multiply a pair of numbers while adding together another pair this gave great performance but it had its limitations the most significant ific of which was that an instruction had to completely finish before an instruction of the same type could execute the scheduler could mitigate this to an extent by interweaving different instructions but it only had limited memory if a program contained a series of additions in a loop there would be no parallelism at all each instruction would have to wait for the previous to finish executing Cay realized that the functional units themselves were almost completely linear in nature the instruction would be fetched from memory decoded executed and then the result will be written back to memory and the registers of the processor each one of these stages was independent using different circuitry and different parts of the system this was something that could be exploited by processing instructions as a series of steps instructions could be processed in parallel within a single functional unit this technique known as instruction pipelining had already been tried on other machines including those from IBM but the combination of pipelining and the 6600 super scaler design was not only unheard of it was groundbreaking in theory this could boost performance by around 10 times the limitations in the circuitry meant real world performance would only be about three times faster the rest of the performance would be gained by packing the circuit boards even closer together and cranking at the clock speed at launch the 7600 was unquestioned the fastest computer in the world while the latest IBM could perform 2 million operations per second the 7600 could perform seven its architecture was groundbreaking there was particular praise given of its innovated c-shaped packaging that allowed for the blistering clock speed the machine sold well in the defense sector where governments could absorb the 5 million doll base price however sales elsewhere were not as strong as expected for for one software written for the 6600 was not compatible with the 7600 the machine was also unreliable often breaking down several times a day and many of the internal components were not easily repairable due to the tight packaging for all the 7600 performance it was the wrong time to launch such a product the country's slowing economic growth weakened appetite for Flagship machines and the lack of cutdown versions Limited profitability many customers wanted to buy a cheaper machine and upgrade to a more powerful compatible system when required this was simply not what Control Data offered for all of IBM's bureaucracy their Market researchers knew what they were doing this new low-end market was where the growth would lie customers would purchase a basic machine and upgrade it when the company had grown large enough IBM's system 360 family of computers did just this it was a range of products from $50,000 to 6 million they could run all the same code while it was true that the 7600 sold far more units than the fastest system 360 model the low-end machines sold in such high numbers that it didn't matter IBM had a computer for almost every customer Control Data had a computer for almost none the development of the 7600 successor the 8600 started shortly after the 7600 launch typical for cray his Ambitions were Mighty the new machine should be 10 times as fast as its predecessor unfortunately increasing the clock speed to help achieve this would be much harder than it had been prior the development of discrete components had plateaued and the increase would have to come from packaging alone the 8600 design itself was also ambitious packing four 7600 processor units within a single system the process is would all run the same instructions but would split the workload between the four of them reading and writing from a shared memory pool while it would require programmers to rework their code the potential benefits were immense when the engineers benchmarked the first prototype the performance was astounding in front of them was a machine that went programmed correctly really was 10 times faster than the current fastest computer in the world but there was a problem a problem so big even cray didn't know how to solve it the 8600 processors were made up of modules essentially 8 8 in x 6 in circuit boards stacked on top of one another to reach the required 125 MHz clock speed the components Within These circuit boards were packed together so tightly that the heat generated was simply unmanageable each module was only the size of a college textbook yet used more power than three refrigerators the cooling solution developed by Dean Rous mounted large copper heat sinks between each board drawing heat away into the coolant even with a refrigeration capacity twice that of the 7600 the machine still ran way too hot the combination of typical failing components and the excessive heat resulted in a machine with truly catastrophic reliability cray of course was used to setback on this project the 6600 had spent years in limbo and the 7600 didn't work reliably until at least a dozen had been produced 2 years went by as the team of Engineers attempted to address the issues but it was to no avail the machine simply would not work by 1971 Control Data who had been fighting illegal battle with IBM over monopolization were hemorrhaging cash it was a battle they would eventually win but but to stay afloat in the meantime the company would need to cut spending management sent a memo to the chipware falls research team requiring them to reduce expenses by 10% for Cay this was a slap in the face his team had generated over a billion dollars in sales yet they were pulling funding for a struggling project to keep the project going Cay sacrificed his own salary dropping his pay to minimum wage it dawned on cray however that 8600 would never reach the market in the current conditions sitting down with company chairman Bill Norris Craig spelled out the situation the 8600 did not and would never work the team would need to start fresh and they would need funding to be increased after all this exact situation happened with the 6600 and that machine generated half a billion dollars in Revenue but to craze disappointment Norris said he couldn't do it while cray was building the 8600 another legendary CDC engineer Jim thoron was building an equally ambitious machine the difference was that Thornton star 100 was working the company was already taking orders Norris explained that the company could not afford to finance two Flagship supercomputer projects but if the project was put on hold for a year the company may be able to increase funding once development resumed Craig replied that he needed time to think it through but by the time he walked out the room he' already made up his mind Craig met with colleag Les Davis shortly after Davis was a brilliant engineer one of Craig's closest Partners the two worked well together he was softspoken and likable possessing the people skills that cig didn't when Engineers had a problem they would come to Davis not cray in the meeting Cay spelled out his plans he was going to leave Control Data to start his own company he'd spoken to long-term colleagues including Frank meany and had raised $2.5 million of funding of course with a large computer costing $50 million to develop it wasn't going to go particularly far the new Venture would likely have to do contract work for much larger computer manufacturers it was Ambiguously christened cray research for this reason this wasn't ideal they would prefer to work on their own projects but it would be far more enjoyable than what they were currently doing Davis agreed the cutbacks on the 8600 project had seriously lowered morale and control data management simply didn't see computers as a growth area limiting The Leverage that Engineers like him could have over the next week the two began to crystallize details who they'd bring with them where they'd work what the salaries would be it was important for cray not to poach too many staff for Control Data he' split on good terms with the company the chairman Norris completely understood crazy decisions even choosing to invest $250,000 in the new business the team would be small cig Davis Rous and a few other Engineers working in a small facility a few streets away from the Control Data lab by mid 1972 the new Venture was fully operational although funding was limited the ultimate goal for the new company was to build a revolutionary supercomputer previous cray machines had used discrete components physical transistors and resistors solded to a printed circuit board however during the development of the 8600 it had been clear that this approach was no longer feasible the new machine would use integrator circuits a component that incorporated transistors in digital logic onto a single silicon chip while integrated circuits had been available during 8600 development their performance was not sufficient enough to be used in a supercomputer however in the 5 years that passed the quality of the chips had improved dramatically making them viable for this new project incorporating integrated circuits into the new machine would bring several advantages firstly it would allow for tighter packaging improving performance additionally the cooling could be simplified the integrated circuits cray wanted to use for all the same height so a single cooling plate would have direct contact with every chip to be competitive the machine would need to employ a technique called Vector processing in essence Vector processing allows a computer to perform calculations on a group of numbers in the same time it would take with just one consider a program that adds together two numbers four times a traditional supercomputer would process one additional after another with Vector processing a machine could calculate all four additions simultaneously in addition fetching the numbers from memory and storing the result of the additions could be done just once traditionally this would have to be done four times in practice it was never that simple the star 100 project to Control Data faced enormous difficulty implementing this technique efficiently its Vector implementation worked by reading and writing data directly to memory by doing so the machine could process vectors of unlimited length however since the memory was slow the processor required a long pipeline to ensure it didn't bottleneck the system unfortunately this meant switching between Scala and Vector instructions something done frequently by most scientific programs carried a heavy penalty this combined with a slow clock speed resulted in the machine being no faster than the 7600 in the real world Craig's new machine would be different from the beginning the computer was designed to have excellent Scala performance a combination of advanced wiring packaging techniques meant the machine could run at a clock speed of 80 MHz more than three times faster than the star 100 however it was in Vector processing that the computer would really shine cray noticed that in scientific programs the code would mainly consist of a series of mathematical operations on the same data on the Star machine the result would have to be written to memory and reread after every operation killing performance with a new system a different approach would be taken the processor itself would contain eight Vector registers each 64 numbers wide values would be loaded from memory calculated on the processor and then written back to memory the advantage was that the values wouldn't have to be written to memory until all the calculations on the data were finished the machine would also feature a vector pipeline allowing programmers to chain together Vector instructions to extract even more performance the result was that by 1974 the team felt that they had created something special testing on the newly christened cray 1 showed it could easily compute 80 million operations per second making it roughly four times faster than the 7600 however like many of C's projects the software St was unfinished and hardware unreliability needed to be addressed this required funding the company was spending $42,000 a month and by this time was $2 million in debt but cray knew a man who could help by every definition John rollwagen was exceptionally talented graduating from MIT with a degree in electrical engineering he' worked as an intern at Control Data before completing an NBA at Harvard the man was bright easy to like and charismatic the ideal salesman returning to the company after graduation he had been the driving force behind the sales of the 6600 the most successful machine cray had ever worked on his stint there was short he only stayed for 2 years before moving on to work for other companies but cray knew that if anyone could find funding and customers it would be him roag and joined cray research in early 1975 as vice president of marketing at the time cray had recently secured a $5 million loan from A bank assuring raren that he had plenty of time to find further funding however in the first board meeting Cay announced the loan had fallen through by August the company only had enough cash to last 45 days eager to impress ragen quickly managed to secure a $1 million bank loan enough to last an entire year and and generated a further $600,000 by selling bonds cray however didn't want to borrow any more money he wanted to raise funds by taking the company public ragen was stunned at the time many people saw the supercomputer industry as one in Decline Flagship machines were too expensive and too risky to produce and mass Market machines were where the growth was where computers were becoming smaller and smaller but to his surprise investors were delighted to invest in Cray's new company by this point cray had built up a reputation of a supercomputer magici and with a new machine almost ready the IPO managed to raise $10 million in a matter of days the infusion of funds allowed the company to settle its debts expand and finish the production of the cray one now it was time to sell it to drum up enthusiasm the first computer was leased to Los Alamos National Lab abs for free for 6 months as the supercomputer Community were fairly tight-knit by doing so ragen hoped hype for the product was spread through word of mouth and spread it did calls started coming in monthly from companies asking when they could get a machine production jumped from two to four then eight machines a year by 1979 orders were coming in at a rate of 1 a month astounding compared to cra's original goal of one every months ragen who by now was president of the company heavily promoted cra's genius for marketing purposes a technique that was extremely successful the company would go on to sell more than 100 units generating over a billion dollars in sales during the late stages of the cray1 development cray began to dream up the next machine in the company's lineup the crate 2 would feature a four processor design akin to the CDC 8600 which was now possible to implement reliably using integrated circuits the design featured four Vector processors entirely for data processing and a foreground processor designed to feed the vector processes and manage the system development would begin in 1976 but a combination of Shifting Focus to a new memory enhanced cray one model and personal issues resulted in more than a year passing without any progress to catch up on Lost Time cray proposed ditching the four Vector processor design instead suggesting a 64 processor scaler concept but after a year this idea was scrapped due to compiler difficulties and the design reverted back to the original even this proved a challenge in the cray one heat was removed by placing large cooling plates along each circuit boards running the coolant across the integrated circuits before piping it into a chiller where the heat would be dissipated The increased density of the new machine meant that this design had to be modified the CR 2 would feature a significant amount of components and component wiring requiring holes to be cut in the cooling plates to allow these wires to pass through the issue was that by the time enough enough holes had been cut there wasn't much of a coing plate left by now it was 1981 5 years after the development had started and the Machine was going nowhere the flurry of administ FR ation that Cay faced day to-day certainly didn't help the company had hundreds of employees on the payroll and his position of chairman left little time to work on the engineering Craig would shortly leave this position becoming an independent contractor for the company to focus more on the project with his time now freed up it wasn't long before cray cracked the cooling problem the original idea of building the machine out of large circuit boards would be scrapped instead eight smaller boards stacked one on top of each other would be used in their place communication would be handled by pins running between each board allowing each stack to perform just as well as the first design the boards would be too difficult to cool using air but cray had no such plans instead the electronics would be submersed in a new coolant developed by 3M which allowed the components to have no issues running submerged progress was finally starting to accelerate but the staff including cray were well aware that the project was in its sixth year of development and still years away from release the machine would also have competition not from another company from within you see Les Davis close friend of Seymour and the co-founder of C research had witnessed the damage the 8600 project had done to Control Data 15 years prior thus when the cray 2 project started stalling in 1979 Davis led a separate effort to develop a replacement to the crate one the computer would be designed using the company's new hotot engineer Steve Chen a Taiwanese designer who studied parallel Computing at the University of Illinois upon hearing about this effort many of the k 2 Team dismissed the project they saw it as corporate uninspiring and joking referred to the new machine as a cray one and a half the machine would be an evolution of the cray1 design but with significant upgrades for one the design would use new integrated circuits with improved density the number of logic gates per chip increased from 2 to 16 inside the casing would lie two Central processors sharing a 16 MB memory pool finally memory bandwidth was significantly improved each processor would have four parallel connections to memory compared to just one on the cray 1 even though Chen's team had started just 3 years ago by late 1981 the machine du the XMP was fully operational and at 2 and a half times faster than the cray 1 it gave the cray 2 a real run for its money the cray 2 did eventually launch in 1985 but it was barely faster than the four processor cray XMP that had launched a year prior the machine had its Niche the advantage of the cray 2 design was that it could support a huge amount of memory a launch the machine was fitted with an enormous 2 GB 32 times that of the flagship XMP 48 the CR 2 was a supercomputer in the traditional sense built without compromise incompatible with the previous generation and designed for a niche audience but the truth was the times were changing customers wanted compatibility as well as performance many organizations could now afford the $11 million starting price but the cost of rewriting all of their software pushed them to machines such as the XMP where few changes to their codes were needed the cray 2 would only sell 27 units less than a quarter of the XMP the success of the XMP created a changing power Dynamic many in CR management felt that the machine had saved the company the 10year gap between the cray 1 and cray 2 would have been difficult to survive without it Steve Chen was handsomely rewarded becoming vice president of engineering and was selected to design the replacement for the XMP the ymp this machine would be a further evolution of the original cray1 concept it would feature up to eight processes and Advance vsi integrated circuits which improveed clock speed and simplify the cooling system like the XMP it was also compatible with code written for Prime machines however Chen wanted more he thought that Les Davis whom Chen reported to was not giving him enough freedom on the project he wanted to push the ymp Beyond an evolutionary design using a dual Pipeline and complex new circuitry but Davis now a veteran of the industry had seen ambitious projects spiral out of control too many times and the ymp was too crucial a product to experiment on despite this he couldn't leave Chen empty hand Ed he was a brilliant engineer well-liked among the team's staff and it would be unfair to Simply ignore his demands the two came to an agreement Chan would work on a new far-reaching development the cray MP and Davis would oversee the ymp to completion meanwhile cray had started work on his latest machine the cray 3 the architecture would borrow much of its design from the ymp but would now feature up to 16 to processes the most significant difference between the cray 3 and any other machine was the use of gallium R semiconductors compared to traditional silicon Gallum ar9 transistors ran cooler and were much much faster the issue was that no one had ever mass produced any requiring the company to develop them all in house by 1987 the company was bleeding cash it was funding the cray 3 project Chen MP Venture and the beginnings of a ymp successor called the C90 the last few years had been incredibly profitable for the company but like Control Data before it it simply grown too fast the ymp was still a year away from release and sales of the XMP were slowing Cuts had to be made the cray 3 project was span off into its own company cray computer Corporation with cray himself being ANP dependent contractor Chen hearing that his funding was being cut was Furious and also decided to follow his own path along with other Engineers he would soon leave cray research to found supercomputer Systems Incorporated or SSI to build a machine even more daring than the MP only the C90 would remain in house the reality was that the supercomputer Market had not only stopped growing it was shrinking the fall of the bir wall and the collapse of the Soviet Union led many governments to reevaluate their defense spending after all why spend billions of dollars to defend against an enemy who no longer exists the end of the Cold War drastically reduced supercomputer sales which were already struggling as a result of a recession that had hit many Western countries the cray 3 would never be commercially sold only one unit was ever produced loan down for free to a government agency the company would go bankrupt shortly after and craze design would never be used Chen's new company SSI followed a similar Fate by 1993 his company was also bankrupt only the C90 would see the light of day selling moderately well the 1990s brought in a new era of massively parallel systems replacing a small group of custom processors with thousands of desktop class microprocessors for many years this concept was dismissed by the industry as the performance had always lagged behind Ground Up designs however the massive cost of engineering custom systems combined with significant advancements in compiler Technologies made the traditional approach harder and harder to justify it was the microprocessor vendors Intel AMD and IBM who would reap the benefits of these new approaches but the impact of cra's designs along with the engineers around him I still felt today Technologies pioneered by cray such as pipelining super scaler architectures load store units and simd processing form the basis of all modern microprocessor designs some estimate that today's micr processors would be decades behind if not for Cray's work the cray 3 and its successor the cray 4 would never have any success but it didn't matter CRA Legacy would live on forever [Music]
2023-12-16