A computer that is at the leading edge or forefront of computing speed is called a 'supercomputer.'  Many years ago, the term 'supercomputer' was defined in terms of the minimum number of floating point operations per second (flops) that it could do, so any computer that reached that minimum was considered a 'supercomputer'.  With rapid and continuous advances in computational power over the last few decades though, it became evident that it is useless to assign a specific number to the definition.  Thus, today, computers that lead the pack in terms of calculation speeds are the ones referred to as 'supercomputers'. 

The era of mega-flops (a million flops) computing started sometime in the early 1960's.  By the mid-60's, a company called Control Data Corporation (CDC) started to dominate the supercomputer market because of powerful supercomputers designed principally by a guy named Seymour Cray.  This market leadership was held by CDC for about a decade until Cray left CDC to form his own supercomputer company in the mid-1970's. 

His company, called 'Cray Research', created the famous Cray computers of the 70's and 80's.  It was during this period that ordinary people began hearing about 'supercomputers' and 'megaflops', thanks to Cray's super machines. The Cray 1, which was first released in 1976, could do 250 megaflops. It turned out to be one of the most popular supercomputers ever produced, with more than 80 units sold worldwide.

    

Figure 1.  Photo of the famous Cray 1 (left) and Cray 2 (right)

As of this writing (July 2009), the fastest computer in the world is the IBM Roadrunner, a machine that is capable of 1.105 petaflops (a petaflop is one million billion flops). This supercomputer consists of a cluster of 3240 computers, each of which has 40 processing cores.  IBM expects to unveil "Sequoia" in the year 2011, a supercomputer capable of 20 petaflops!  To get an idea of how supercomputers evolved over time, see the list of supercomputers since 1961.

    

Figure 2.  Photo of the IBM Roadrunner

   

Early supercomputers produced by CDC were just superfast scalar processors. These were subsequently replaced by supercomputers that employ vector processors in the 1970's.  In the 1980's, supercomputers started using multiple vector processors working in parallel.  By the mid-90's, a typical supercomputer would have thousands of ordinary microprocessors (including "off-the-shelf" ones) working in parallel.  Today, supercomputers operate on the same basic structure, employing thousands of microprocessors working in parallel (except that they are now using much more powerful server-type microprocessors).

Supercomputers pose a variety of challenges to design and build.  Handling thermal runaway issues is a common challenge since a supercomputer produces tremendous amounts of heat.  Making thousands of CPU's work in parallel requires minimum latencies between them, so keeping conductors between components as short as possible is a 'must'. This is why Cray computers are cylindrical (see Figure 1).  External storage as well as I/O-related problems are also big issues for a supercomputer with thousands of powerful processors churning out enormous amounts of data in parallel every second.

See Also:   List of Supercomputers;  More Industry Articles