What is the worlds fastest computer?

By: Shane Dempsey

Mankind is obsessed with speed. I mean mankind in the generic sense but most women would agree that part of the male condition is an unhealthy preoccupation with making things, anything, go faster. The world of computing is no different with ever more expensive and powerful machines being constructed by teams of gifted engineers around the world to tackle computational tasks involving massive 'number-crunching' and data storage. Examples of such problems include mapping the human genome [1], analysing geological data, theoretical physics including cosmology and that old chestnut; predicting the weather.
It is clear that these tasks are beyond ordinary desktop machines, which gives rise to the term 'Supercomputer'. A supercomputer is a broad term for one of the fastest computers currently available. For many years, the speed of computers was measured by how many millions of instructions per second, or MIPS, they could execute. This measurement would be a suitable barometer of system performance, were it not for large differences in the instruction sets that different families of machines use. For example Reduced Instruction Set Computing (RISC ) machines, like Apple's G4 computers, have an instruction set that is radically different to Intel's Complex Instruction Set Computing (CISC ) processors found in most PCs. Hence there can be large differences between these processors in the number of instructions necessary to carry out any particular task. Supercomputers are primarily number crunchers so their performance tends to be measured in Floating point Operations Per Second ( FLOPS ). Current supercomputers' performance is measured in billion FLOPS (Giga-FLOPS) and trillion FLOPS (Tera-FLOPS).
The secret to this performance normally consists of breaking up processing tasks into sub-tasks that can be executed concurrently across multiple dedicated processors, potentially using a single communications 'bus' or over a very high-speed network backbone. By simultaneously executing tasks on high-speed, bespoke processors or even clusters of standard PC processors and, by cleverly managing how memory and communications resources are 'shared' between these processors; massive number crunching capabilities can be realised. There are a few schools

of thought on how best to realise these number crunching capabilities. In general these depend on whether the computer will be built from a large number of off-the-shelf processing units or from a smaller number of custom, very high-powered processing units. The Intel processor in most home PCs is an example of a processing unit. Parallelism is achieved using software together with hardware. To provide a real-world analogy, a business needs a management function to enable the employees to collaborate towards the business goals in a focussed and structured fashion. The operating systems and programming languages for parallel machines are analogous to and fulfil this role. For this reasons it is possible to use many standard PCs on a network as a supercomputer using cluster-computing software such as that developed by the Beowulf project. [2]

The fastest computer on the planet is the 'Earth Simulator' developed by NEC Corporation. This machine is housed in the Earth Simulation Centre in Kanazzawa, Japan. Its peak performance exceeds 35 Teraflops (trillion floating point operations per second). More information can be obtained from the 'Top 500' website [3], dedicated to providing information about the top 500 supercomputers every year since 1993. Earth Simulator or ES as it's affectionately known contains 5120 processors, 10 Terabytes of memory (about 40,000 times the memory of the average home PC) and occupies the same area as 4 tennis courts. Its performance is staggering as its computational capabilities are greater than the sum of the other 19 fastest computers on the planet.

For further information or reading on the subject of supercomputing you can take a look at the following sites listed.
[1] Human Genome Program
[2] The Beowulf Project
[3] Supercomputing Top 500