The tough way towards Cray-2 – When dinosaurs start becoming extinct

Already in 1976 Seymour Cray starts research and development towards the Cray-2 supercomputer. The machine is supposed to be ten times faster than its predecessor Cray-1, which didn't work out for several reasons. Of the followup model Cray-3 even only one unit got sold, but never paid, because it never worked as promised.

This talk is the fourth talk of a sequel about life and work of Seymour Cray. Main focus of the talk are challenges during development of the Cray-2 supercomputers, one of the latest of its kind.

Transcript

1. 1 Wolfgang Stief VCFe, Munich, April 2018 Picture: Wolfgang Stief, CC0
2. # whoami • Electrician, Degree in Electrical Engineering • Freelancer, self employed (sys4 AG)
 Storage, E-Mail, Projectmanagement, Board of Directors • Computermuseum München, Cray-Cyber.org
 “Everything below 30 A is leakage current.” • https://about.me/stiefkind
 stief@cray-cyber.org
 ◦ stiefkind 2 Picture: Wolfgang Stief, CC0
3. 00_README.txt • Work in Progress • Cray and Surroundings
 Companies
 People
 Technologies
 Branches • series of lectures (VCFE/VCFB) • http://www.speakerdeck.com/stiefkind/
 https://media.ccc.de/search?q=Wolfgang+Stief
 video recordings are German language only 3
4. # ls -l • Cray Research Inc.
 ca. 1976–1986 • Cray-2, X-MP, Y-MP • Supercomputer Systems Inc.
 Cray Computer Corporation
 Cray-3 4
5. What happened so far (pt. 1, 1949-1959) • CSAW (US Navy, WWII) ➛ ERA (1946) ➛ Remington-Rand (1951) ➛ Sperry-Rand (1951) ➛ Control Data (1957) • 1951: Seymour Cray (∗1925 †1996) starts at ERA
 B. Sc. Electrical Engineering (1949), M. Sc. Applied Mathematics (1951) • Control Data Corporation (Fall 1957)
 plan: U$ 600.000 seed money, shares privately hold
 reached: U$ 1.2 Mio. • CDC Little Character (trial run for transistorized computer) • CDC 1604 (1959)
 48bit, 0.2 MHz ➛ fastest computer of the world 5
6. What happened so far (pt. 2, 1959-1972) • CDC6600: 100 ns ≙ 10 MHz, 1-3 MFLOPS (1959-1965)
 > Si transistors, liquid cooled (Freon) • CDC7600: 25 ns ≙ 40 MHz (1965-1969)
 > pipelining;
 > U$ 8 Mio. for only ca. 4x Performance ➛ for many it’s too expensive • Control Data Corporation: expansion worldwide, aquisitions
 > shares from U$ 1 (1957) ➛ U$ 300 (1964)
 > lawsuit CDC ./. IBM because marketing practice, monopoly, antitrust
 > business orientation more and more towards service • CDC8600: 8 ns ≙ 125 MHz, Quad-CPU (discrete)
 > reliability (!) ➛ way too many solder joints, heat 6
7. What happened so far (pt. 3, 1972-1976) • foundation Cray Research Inc.
 > Q1/1972, after development of CDC8600 is stuck, CDC becomes investor • Design Cray-1: single CPU, 12.5ns clock cycle (=80Mhz)
 > vector unit • mid march 1976 IPO
 > 600.000 Shares, ca. 10 Mio U$
 > paying debts, finish Cray-1 • Plan Cray-1: 2 systems/year
 > reached: ca. 80 systems in ca. 6 years
 > price per unit 5-8 Mio U$, depending on configuration • John Rollwagen becomes president
 > max. 10 years
 > stays until 1993, afterwards Department of Commerce under Clinton 7
8. • from ca. 1976 first thoughts and developments
 > market release Cray-1 in 1976, followed by revised versions • typ. Seymour Cray: let others do the cleanup
 > “janitor” Les Davis: CDC6600, CDC7600, Cray-1 • plan: 4 ns clock cycle (≙ 250 MHz) • plan: 4-CPU • Cray distracted by model improvements of Cray-1
 > Cray-1A, Cray-1S, Cray-1M Cray-2 – how it began 8
9. • ca. 1 year break because of revised versions Cray-1 • new plans:
 > 100% scalar, no more vector units
 > 64 scalar processors • massive parallel is new to Seymour Cray
 > industry starts massive parallel ca. 5 years later
 Thinking Machines: 1993; Transputer/INMOS from 1978, introduced 1983
 > so far: always reliable, well known and well understood technology • Steve Nelson: writes compiler for 64 scalar CPUs
 > order by Seymour Cray
 > doesn’t know anything about, has to self educate first Cray-2 – how it began 9
10. • ca. fall 1978: visit @ Lawrence Livermore Lab
 > Cray & Nelson show/explain 64 CPU idea and compiler
 > intensive discussion • response by Lawrence Livermore not as expected
 > Cray throws away 64 CPU idea without any emotions
 > back to original design: 4 CPU, 4 ns clock cycle 㱺 two years of development evaporated! Cray-2 – how it began 10 “Steve, we’re not going to do this. The customers 
 are not ready for a sixty-four processor system.”
11. • shorter wires because of speed demands
 > already successful at CDC6600 ➛ CDC7600
 > Cray-1: max. 120 cm
 > Cray-2: max. 40 cm • “sandwich type” processor module
 > 8 circuit boards, stacked
 > 7 cooling plates in between • communication between circuit boards
 > holes in cooling plates for wiring
 > cooling capacity decreases, no sufficient heat transport • stuck, see also CDC860 Challenge Cray-2: waste heat 11 Quelle unbekannt
12. • 1977 ➛ 1978: +60% people
 > 321 ➛ 524 employees • 1978 ➛ 1979: profit +149% (U$ 42.7 mio.)
 in 1980 expected +50%
 in 1981 expected +50%
 > Cray Research amongst the 10 biggest computer companies • sales 100% from Cray-1
 > 1978 ➛ 4 systems sold
 > 1979 ➛ 8 systems sold
 > 1980 ➛ 10 systems sold
 > end 1980: backlog of 5 systems • and: employees don’t recognise president in meetings Challenge Cray Research: growth 12
13. • throw away all work until now • decrease team size (drastically) • start over • new: create internal competition The Cray Way 13
14. • research lab Cray Labs in Boulder, CO • many engineers from semiconductor industry
 > Fairchild et al. • assignment: VLSI research/development
 > better integrated chips for Cray-2 • parallel: development Cray-2 in Chippewa Falls The Cray Way – Cray Laboratories 14
15. • Data General Mainframe in bed room
 > several racks, 6 ft high
 > many “Super Brains” distributed all over the place
 (PCs? Workstations? maybe Intertec Superbrain?)
 > spares in garage + Hallie Lab, on call engineer
 > test equipment for circuit boards • unreliable power grid
 > power supply at garage • mainframe produces heat
 > capable air con installed Seymour Cray, the eccentric 15 cottage!
 But this way Cray avoids interruptions.
16. • agreement Cray/Rollwagen (1981)
 > retirement Seymour Cray as chairman
 > handover chairman to John Rollwagen
 > Seymour Cray becomes independent consultant • payment: 6% of development costs spent per year
 > U$ 1 mio. development budget/year ➛ 60.000 U$ annual salary • Seymour Cray finally is back at 100% engineer. Finally again being engineer only 16
17. • totally new approach: immersion cooling
 > electronics bathing in coolant 
 > experiments at CDC 10 years earlier, Cray knew about • coolant Fluorinert™
 > inert, isolating, inflammable liquid
 > Minnesota Mining and Manufacturing (heute: 3M)
 > still sold today So what about waste heat of Cray-2? 17 cCu = 382 J kgK cFluorinert ≈ 1200 J kgK cH2O = 4182 J kgK
18. • press conference
 > management: development looks promising and nearly done
 > jump in stock price
 > day before press conference: +U$ 3/share ➛ short-term trading pause
 > wild speculations @ analysts • details on Cray-2 
 + deal Cray/Rollwagen • after press conference: 
 stock price decreases
 > U$ 36/share ≙ as before
 > stock price decreases further Public announcement Cray-2 18 https://www.dkrz.de/systeme/historie
19. • VLSI development behind plan
 > with immersion cooling not needed anymore • Lab gets terminated in April 1982
 > 1981..1982
 > by management “enacted startup” won’t work back then either • future for Cray Research lays in Cray-2
 > that’s what John Rollwagen still beliefs in Cray Labs ➛ blacklisted 19 “The hope of the future is in Boulder and that’s why we’re making such a big investment here.” Seymour Cray
20. • introduction Cray X-MP (April 1982)
 > 5× faster then Cray-1
 (10ns clock cycle)
 > dual CPU design, 10 gates/modul 
 (Cray-1: 2 gates/modul)
 > backward compatible to Cray-1 • X-MP came into being 
 completely without S. Cray
 > Steve Chen, since 1979 @ CRI, 
 becomes chief architect quickly
 > skunk works project • press: building supercomputers also works without S. Cray Meanwhile in a Cray Research parallel universe… 20 Quelle unbekannt
21. • Les Davis
 > foresees, that Cray-1 comes to an end
 > foresees, that Cray-2 will still need some more time • new team, ideas from 3 years development Cray-2
 > Chen acts as the brain behind
 > most in the team are < 40 years old
 > within only a few weeks development of Cray-2 gets outpaced • backward compatibility becomes important in market
 > Seymour Cray always built something new, no backward compatibility
 > customers aren’t willing to spend resources into development of
 operating systems, compilers, tools • Les Davis immediately starts development of Y-MP after X-MP Why Cray X-MP? 21
22. • end 1981 already “tuned” Cray-1 in lab
 > multiport memory for dual CPU
 > 2 CPUs in ½ Cray-1 chassis
 > CPUs are faster/better because of higher gate density
 (X-MP: 2.3 mio logic gates = 8× Cray-1) • started with 200 MFLOPS • backwards compatible to Cray-1 (design goal) • unit price U$ 11 mio … U$ 14 mio (1981) • Company Digital Production
 > leasing for U$ 275.000 / month
 > rendering for TRON Cray X-MP Facts 22
23. • Cray-2 still unfinished
 > development of machine finalised
 > challenge: manufacturing process
 > Les Davis is not available for “cleanup” (X-MP, Y-MP) • 1985: market introduction Cray-2
 > nine years development!
 > completion by Jerry Brost & team • overall only 27 Cray-2 sold
 > unit price U$ 12 mio … U$ 17 mio
 > Cray-1: 80+ units sold
 > memory borrowed from X-MP
 > Cray-2 supposed to be faster then X-MP, but is also 3 years “younger” Fast Forward: 1981 ➛ 1984 23
24. • X-MP is available in a 4 CPU configuratin
 > Cray Research in total sells 126 X-MP machines
 > unit price ca. U$ 16 mio • ca. 3 year “gap” between X-MP and Cray-2
 > Japan catches up: Fujitsu, NEC, Hitachi At the same time… 24
25. Life after Cray-2 25 • Cray-3 becomes ruinous for Cray Research
 > spin-off Cray Computer Corporation • Y-MP is really successful
 > see also current VCFE exhibition of a 
 running Cray Y-MP EL (entry level) (1st floor) • Steve Chen establishes Supercomputer Systems Inc. (SSI) • competitors catching up further: Japan, CDC/ETA • sponsorship massive parallel systems @ DARPA
 > Thinking Machines
 > Transputer (z. B. INMOS) • cliffhanger/spoiler for next year’s talk :)
26. Beyond this talk — reading • Charles J. Murray — The Supermen
 The Story of Seymour Cray and the Technical Wizards behind the Supercomputer • Kay A. Robbins, Steven Robbins — The Cray X-MP/Model 24
 A Case Study in Pipelined Architecture and Vector Processing 26
27. Beyond this talk — reading • Bitsavers PDF Archive (Mirror Uni Stuttgart)
 http://bitsavers.informatik.uni-stuttgart.de/pdf/cray/CRAY-2/ 27
28. • Oral History of Les Davis (Mai 2010)
 http://archive.computerhistory.org/resources/access/text/2012/04/102657915-05-01-acc.pdf • Gordon Bell — A Seymour Cray Perspective (presentation)
 http://research.microsoft.com/en-us/um/people/gbell/craytalk/ • Cray 2 Reboot Project
 Building “a a clock and gate equivalent recreation of a Cray Research Cray-2 supercomputer that will run on an FPGA board.”
 http://www.darkwavetech.com/index.php/cray2blog/ 28 Beyond this talk — reading
29. Beyond this talk — watching • Cray Research — A Story of the Supercomputer (documentation)
 https://www.youtube.com/watch?v=wn03wn3k47Y 29
30. EOF Thanks! Questions? 30 https://about.me/stiefkind stief@cray-cyber.org " stiefkind http://www.speakerdeck.com/stiefkind/ https://media.ccc.de/search?q=Wolfgang+Stief Bild: Wolfgang Stief, CC0
31. License The entire presentation without pictures is licensed under CC0 1.0 (public domain dedication). Licenses to the pictures are specified directly at the pictures. 31