2. Table of Contents
Introduction
What are They Used For
How Do They Differ From a Personal Computer?
Where Are They Now
Main Parts of Supercomputers
Processor Types
Conclusion
References
3. Supercomputers
The category of computers that includes the
fastest and most powerful (most expensive)
ones available at any given time.
Designed to solve complex mathematical
equations and computational problems very
quickly.
4. What are They Used For
Climate prediction & Weather forecasting
5. What are They Used For (cont.)
Computational chemistry
Crash analysis
Cryptography
Nuclear simulation
Structural analysis
6. How Do They Differ From a
Personal Computer
Cost
range from $100,000s to $1,000,000s
Environment
most require environmentally controlled rooms
Peripherals
lack sound cards, graphic boards, keyboards, etc.
accessed via workstation or PC
Programming language
FORTRAN
7. History
Seymour Cray (1925-1996)
Developed CDC 1604 – first fully transistorized
supercomputer (1958)
CDC 6600 (1965), 9 MFlops
Founded Cray Research in 1972 (now Cray Inc.)
CRAY-1 (1976), $8.8 million, 160 MFlops
CRAY-2 (1985)
CRAY-3 (1989)
8. Early Timeline of Supercomputers
Period Supercomputer Peak speed Location
1943-1944 Colossus 5000 characters per second Bletchley Park, England
1945-1950 Manchester Mark I 500 instructions per second University of Manchester, England
20 KIPS (CRT memory) Massachusetts Institute of Technology,
1950-1955 MIT Whirlwind
40 KIPS (Core) Cambridge, MA
40 KIPS
1956-1958 IBM 704
12 kiloflops
40 KIPS
1958-1959 IBM 709
12 kiloflops
1959-1960 IBM 7090 210 kiloflops U.S. Air Force BMEWS (RADC), Rome, NY
1960-1961 LARC 500 kiloflops (2 CPUs) Lawrence Livermore Laboratory, California
1.2 MIPS
1961-1964 IBM 7030 "Stretch" Los Alamos National Laboratory, New Mexico
~600 kiloflops
10 MIPS
1965-1969 CDC 6600 Lawrence Livermore Laboratory, California
3 megaflops
1969-1975 CDC 7600 36 megaflops Lawrence Livermore Laboratory, California
100 megaflops (vector),
1974-1975 CDC Star-100 Lawrence Livermore Laboratory, California
~2 megaflops (scalar)
80 megaflops (vector), Los Alamos National Laboratory, New Mexico
1975-1983 Cray-1
72 megaflops (scalar) (1976)
9. Where Are They Now
www.top500.org
List released twice a year
Scores based on Linpack benchmark
Solve dense system of linear equations
Speed measured in floating point operations
per second (FLOPS)
10. Architectures - SMP
Symmetric Shared-
Memory
Multiprocessing
(SMP)
Share memory
Common OS
Programs are divided
into subtasks (threads)
among all processors
(multithreading)
11. Architectures – MPP
Massively Parallel Processing (MPP)
Individual memory for each processor
Individual OS’s
Messaging interface for communication
200+ processors can work on same application
1. A large retailer wants to know how many camcorders the company sold in
3. Each sub-query is assigned to a specific processor in the system. To
1998, and sends that query to the MPP system. allow this to happen, the database was previously partitioned. For
2. The query goes out to one of the processors which acts as the example, a sales tracking database might be broken down by month, and
coordinator, it breaks up the query for optimum performance. For
example, it could break the query up by month; this “sub-query” each processor holds data for one month’s worth of sales information.
4. The responses to the queries are returned to a processor to be coordinated—for
then goes to all the processors at the same time.
example, each month is added up
5. Final answer is returned to the user.
12. Architectures – Clustering
Grid computing
Many servers connected together
Relies heavily on network speed
Easily upgraded with addition of more servers
13. Processor Types
Vector processing
Expensive
NEC Earth Simulator
Scalar processing
Grid computing
Based on off the shelf parts (ordinary CPUs)
14. BlueGene/L
IBM
MPP (massively parallel processing)
#1 on top500 as of November 2004
32,768 processors (700Mhz)
70.72 Teraflops (trillions of FLOPS)
Runs linux
DNA, climate simulation, financial risk
Cost more than $100 million