Supercomputing is a term used to describe computational tasks that are performed at extremely high speeds. There are a number of different types of supercomputers, each designed for specific workloads. The most common type of supercomputer is the general-purpose supercomputer, which can be used for a variety of scientific and engineering applications. Other types of supercomputers include those designed for specific workloads such as weather forecasting, climate modeling, or protein folding. Supercomputers are typically used for tasks that require large amounts of computing power and memory, such as simulations or data analysis.
The first supercomputer was built in the early 1960s by Control Data Corporation (CDC). CDC 6600, as it was called, could perform up to 3 million instructions per second (MIPS). This made it about 100 times faster than the fastest mainframe computers at the time. CDC 6600 remained the fastest computer in the world until 1974 when CRAY-1 surpassed it. CRAY-1 could perform up to 160 MIPS and became the model for all subsequentsupercomputers.
Supercomputers continued to get faster throughout the 1970s and 1980s as new technologies were developed. In 1985, Intel released its i860 microprocessor, which was specifically designed for high-performance computing. This processor powered a number of record-setting machines in the late 1980s, including NEC SX-2/8 and Fujitsu Numerical Wind Tunnel (NWT). In 1993, Intel released its Pentium Pro processor, which increased performance by another factor of 10. This processor powered a number of impressive machines in the mid-1990s, including Paragon XP/S 24M from IBM and ASCI Red from Hewlett Packard (HP).
In 1996, HP released its first Itanium processor, which represented a radical departure from traditional CPU design principles. Itanium CPUs are based on explicit parallelism – meaning that they are designed to execute multiple instructions simultaneously – rather than implicit parallelism like other CPUs(which rely on software to extract parallelism). As a result, Itanium processors offer significantly higher performance than even the fastest conventional processors when running certain types of workloads. Itanium processors have powered some of the world’s fastest supercomputers over the past two decades, including SGI Altix 4700 from 2002 and HP Superdome 2 from 2005.
While Moore’s Law – which states that transistor density will double approximately every 18 months – has held true for much of the history of computing , there has been an increasing trend towards using more than one chip(or core)to boost performance . This is known as multi-core processing . Multi-core processors began appearing in consumer PCs around 2006 with products like Intel Core Duoand AMD Opteron . However , these products were not really designed with high performance computing in mind . It wasn’t until 2008 that multi -core processor s started appearingin supercomputer s with IBM ’ Sequoia system being oneofthe first . Multi -core proces sor architecturesare now commonplace in highend serversas wella ssuper compute rs with many systemsnow using hundreds or even thousands off individual cores .
With each new generationof chips comes anew setofchallengesfor software developers . Oneofthese challengesis how besttocodeforthe availableparallelism so that programs can take advantageofmulti -core hardware without running into scaling issues further down throadwhen movingto even larger numbersoffcores Another challengeis powerconsumption Withtransistor countsincreasingat an exponentialrate , die sizesare also becoming larger resultinginthesame devices requiringmoreand morepower justtomaintaintheircurrent levelsof performance Thisis particularlytrue formulti -coreprocessorwhereasingle socket maynow containdozensorcores Notonly doesthishaveimplicationsforthe amountoffacility spacerequiredbydatacentersbutalsoforthair coolingneeds Powerconstraintshave ledtothe developmentofthe Green500 listwhich rankscompute rsystemsoftheworldintermsoftheirownenergyefficiencyratherthanjustoverallperformance