2. Outline
• Vector Supercomputers
• Processing Speed-FLOPS
• Generations of Vector Supercomputers
• Application Areas of Vector Supercomputers
• Scientific Attached Processors
• Working methodology
• Applications of Scientific Attached Processor
• Projected Speed and Performence Evaluation
• Advantages and Limitations of Scientific Attached
Processor
• Conclusion
• Bibliography
3. Vector Supercomputers
• commonly called as supercomputers
• came to lights in late '60s
• machines built to handle large scientific and engineering
calculations
• handles one dimensional array of data called vectors
• characterised by-
Computational Speed
Fast and large Memory(primary/secondary)
extensive use of parallel structured software
• has a large application domain
• data elements are arranged in vector/matrix form
• current supercomputers has the processing speed in
petaflops.
4. Processing speed-FLOPS
• Floating Point Operations per Second
• Measure of Computer Performance
• Best used in Scientific Calculation requiring Floating Point
Calculations
• performance measure Table:
Name Unit Value
kiloFLOPS kFLOPS 103
megaFLOPS MFLOPS 106
gigaFLOPS GFLOPS 109
teraFLOPS TFLOPS 1012
petaFLOPS PFLOPS 1015
exaFLOPS EFLOPS 1018
zettaFLOPS ZFLOPS 1021
yottaFLOPS YFLOPS 1024
5. Generations of Vector Supercomputer
• First Generation Supercomputers:
came into light in 1960s
Equipped with multiple functional pipelines
to achieve parallel to vector procesing
Peak speed=40 mFLOPS(STAR-100,TI-ASC)
examples:
STAR-100 (used memory to memory architecture
with 2 pipelines)
TI-ASC(can handle 3D vector computations in
pipeline mode)
Illiac-IV
6. Generations of Vector Supercomputer
• Second Generation Supercomputers:
Built from first generation supercomputers
increased speed
increased Accuracy
Different variations were available for different uses
Example:
Cray-1
Cyber-200
Fujitsu-VP 200
Cyber 205
8. Scientific Attached Processor
• Designed as back-end machines attached to a
host computer
• most of it are pipeline structure
• designed to enhance floating point capabilities of
the host machine
• can be attached to minicomputer/mainframe
• most used systems:
AP-120B
FPS-164
FACOM 230/75
9. Working Methodology
• Processors are attached to Host
• In a host backend computer organization:
Host is a program manager that handles:
Input-Output
Code compilation
Operating System Functions
• backend processor concentrates on Arithmetic
computations with data supplied by host machine.
• Also called as Scientific Array Processor
10. Applications of Scientific Attached Processor
• Stactural Analysis
• Analog circuit simulations
• Computational Chemistry
• Electromegnetic Modeling
• AI applications like Recognition,Classification etc.
11. Projected Speed and Performance Evaluation
• Figure describes the projected speed and
performence evaluation of Scientific Attached
Processors.
• Parameters taken:
Horizontal Axis: Year
Vertical Axis: Million Operations Per Second(MOPS)
all theoretical peak performance of the processor
inside the parenthesis.
• Relative speed comparison between the
processors can also be done in the figure
13. Advantages and Limitations of Scientific Attached Processors
• Advantages:
Enhances the speed of Host Machine
Can be attached to different computers
Attached processor cost much less than Mainframe and
Super Computers
• Limitations:
Most attached array processors must be microcoded for
various operations
The software might be inadequate and costly than the
hardware
High speed interfacing might be required to maintain the flow
of data
14. Conclusion
• Vector Supercomputers has become an intregal
part of various application domain
• Its speed and accuracy has shown the wide
range of possibilities in todays world
• In a world where flow of data has crossed every
limit, evaluation and manipulation with
maintaining the efficiency is one major question
• However, with proper help of Vector Computers
and the processing powers of Scientific Attached
Processors, this can be achieved.
15. Bibliography and Reference
• Bibliography:
Hwang,Kai,Briggs,Faye A.;Computer
Architecture and Parallel Processing
• Web Reference:
http://www.johngustafson.net/pubs/pub19/Kai
Hwang.html
https://en.wikipedia.org/wiki/Supercomputer