3. Introduction
A supercomputer is a computer with a
high level of computational
capacity compared to a general-purpose
computer
4. History
The history of supercomputing goes back to the
1960s, with the Atlas at the University of
Manchester designed by Seymour Cray
The ”Atlas” was a joint venture between ’Ferranti’ and
the ‘Manchester University’ and was operated at
processing speed of one microsecond per
instruction, about one million instructions per
second.
The first “Atlas” was officially commissioned on 7
December 1962 as one of the world's first
supercomputers and is equivalent to four IBM 7094s.
5. Need & Application
Used for solving highly calculation intensive tasks
like in
Weather forecasts
Explosions - detonation of nuclear weapons
Analysis of data
Astronomical observations
Integrate design of engineering products
6. Examples of super computers
Ex: CDC6600
1-10 MFlops
Single CPU
60-bit word length
RISC
Central Processor has 10 parallel functional units
-superscalar.
Implemented 10 peripheral processors
no pipelining
7. EX:
CDC7600 (1969-1975)
10-100 MFlops
Instruction Pipelining
Pipelined execution on functional units in parallel 8
processors
( Cray series)
Worked on new design - vector processors load large
data at once.
8. Technology in super conductors
Performance of a supercomputer is measured
in floating-point operations per second (FLOPS)
instead of million instructions per second (MIPS)
parallel computing is a type of computation in
which many calculations or the execution
of processes are carried out simultaneously.
Using N processors, we expect the execution
time to come down N times ideally hence
speedup by N
9. 1 processor::T(1) = s + p = 1 i.e., p = 1 - s
n processors:: T(n) = s + (p/n)
Scalability (Speed Up) = T(1)/T(n) = 1/(s+(1-s)/n)
Amdahl's law
10. Architecture of super computer
Different architectures to implement parallel
algorithm
1 Shared-memory systems
Uniform Memory Access (UMA)
cache coherent Non-Uniform Memory Access
(ccNUMA)
2 Distributed-memory systems
Massively Parallel Processing
Cluster computing
Each has a standardized programming model
11. Shared memory systems
Two basic categories of shared memory systems
Uniform Memory Access (UMA): Memory is
equally accessible to all processors with the same
performance (Bandwidth & Latency)
Simplest example - dual core processor
cache-coherent Non Uniform Memory Access
(ccNUMA): Memory is physically distributed but
appears as a single address space: Performance
(Bandwidth & Latency) is different for local and
remote memory access.
Examples: AMD Opteron nodes, SGI Altix, Intel
Nehalem nodes – all modern multi-socket nodes
12. Distributed memory systems
Pure distributed-memory parallel
computer
No global cache-coherent shared
address space No Remote Memory
Access (NORMA)
Data exchange between nodes:
Passing messages via network
14. Massively parallel
Independent microprocessors that run in parallel
Each processor has its own memory
All processing elements are connected to form
one large computer
A number of processors work on same task
Examples : Blue Gene
15. clusters
Completely independent computers combined to
a unified system through software and networking
Components connected through fast LANs -
relies heavily on network speed
Easily upgraded with addition of new systems
Eg. 1100 Apple XServe G5 2.3 GHz dual-
processor machines (4 GB RAM, 80 GB SATA
HD) running Mac OS X and using Infinite Band
interconnect
18. conclusion
Supercomputers of today are the
general purpose computers of tomorrow
Many important applications need the
better main memory bandwidth and
computing power that are available only
in supercomputers