What is a supercomputer?

Louhi: Cray XT4/XT5 supercomputer (first cabinet row)

Murska: HP CP4000 BL ProLiant supercluster

1996: The IBMSP was CSC’s first air-cooled supercomputer.

CSC's supercomputers.

A supercomputer is a computer belonging to the highest performance capacity class available at any given time.

The unit used for measuring the performance of supercomputers is Flop/s (Floating Point Operations per Second). It shows how many floating point calculations a computer can perform in one second. Floating point refers to a computer’s internal way of representing real numbers. In this representation, the significant figures and exponent of a number are saved separately.

Among the programs used for measuring the Flop/s value, Linpack developed by Jack Dongarra has gained the status of almost a standard. Linpack is used as the benchmark for the TOP500 list of the world’s most powerful supercomputers, which is updated and published on the Internet twice a year.

The architecture of supercomputers has undergone some major changes during their more than 30-year history. Cray 1, released in 1975, is considered the first real supercomputer. That computer introduced a new feature – the vector processor, which meant a crucial improvement in the speed of processing tabular data.

CSC’s supercomputers from the newest to the oldest

The 21st century and the PC cluster

The PC cluster became the dominant supercomputer architecture in the 2000s. The supercomputer consists of inexpensive general processors of the same type as found in ordinary PCs. The low price makes it possible to increase the number of processors to thousands. The processors are interconnected by either a fast connection network or standard GigabitEthernet LAN technology.

2008: Cray XT4/XT5 and PRACE-prototype

CSC's Cray supercomputer was upgraded to over 100 teraflops (trillions of floating point operations per second). This made the new Cray XT5 system at CSC one of the fastest supercomputers in Europe.

Later in 2008, a second upgrade was funded under the EU's Partnership for Advanced Computing in Europe (PRACE) initiative with two Cray XT5 cabinets. One of the PRACE project tasks is prototype testing for supporting decisions for the installation of the future HPC Petascale systems in Europe.

2007: Cray XT4 and the HP supercluster
Cray XT4, ordered by CSC in 2006 and installed in 2007, uses Opteron multi-core processors manufactured by Advanced Micro Devices (AMD). They are interconnected with Cray’s proprietary connection network with high bandwidth and low latency. In the final assembly of Cray, the number of cores is over 6,000. The theoretical computing power will then be over 70 Tflop/s, i.e., 70 trillion floating point calculations per second.

HP ProLiant supecluster, installed in Summer 2007, uses same AMD’s Opteron multi-core processors, but represents the PC cluster design. The number of processor cores is 2,048 and the connection network is InfiniBand.

The 1990s and massively parallel supercomputers

Massively parallel supercomputers were introduced onto the market in the 1990s. Their performance was based on a large number of scalar processors and fast data exchange between them. A massively parallel supercomputer required a new type of programming, parallel processing. This means that the task to be performed is divided into component tasks, each of which is given to its own processor to perform. The processors communicate with each other as the computation proceeds. The computer’s total power depends on how well the task can be performed as parallel processes; in other words, whether the component tasks can be performed independently of each other or whether much communication is needed between the processes during the computation.

1996: Cray T3E supercomputer

In 1996, CSC put to use the Cray T3E supercomputer, which had 192 processors and was a typical representative of massively parallel architecture. The number of processors in the final assembly totaled 544 and the top performance 384 GFlop/s, or 384 billion floating point calculations per second.

The first supercomputers used special processors that required considerable cooling power. Liquid cooling was common. The assembly of supercomputers from components intended for office use started in the 1990s.

1994: IBMSP

The IBMSP acquired by CSC in 1994 was based on air-cooled SP2 processors. The processors were upgraded later to make them faster and their number was increased. In the final assembly, the IBMSP had 512 processors and its top performance was 2.2 TFlop/s.

1989 and Finland’s first supercomputer

CSC’s first supercomputer, Cray X-MP acquired in 1989, had a speed of 0.935 TFlop/s, or 935 million floating point calculations per second.

CSC

Osiot

Aliosiot

Tehdyt toimenpiteet

What is a supercomputer?

CSC’s supercomputers from the newest to the oldest

The 21st century and the PC cluster

The 1990s and massively parallel supercomputers

1989 and Finland’s first supercomputer


Sisältö Methods of computational science Applications of scientific computing Supercomputers and grids CSC Servers What is a supercomputer? What is a grid?	Tehdyt toimenpiteet Tämä sivu suomeksi What is a supercomputer? Louhi: Cray XT4/XT5 supercomputer (first cabinet row) Murska: HP CP4000 BL ProLiant supercluster 1996: The IBMSP was CSC’s first air-cooled supercomputer. CSC's supercomputers. A supercomputer is a computer belonging to the highest performance capacity class available at any given time. The unit used for measuring the performance of supercomputers is Flop/s (Floating Point Operations per Second). It shows how many floating point calculations a computer can perform in one second. Floating point refers to a computer’s internal way of representing real numbers. In this representation, the significant figures and exponent of a number are saved separately. Among the programs used for measuring the Flop/s value, Linpack developed by Jack Dongarra has gained the status of almost a standard. Linpack is used as the benchmark for the TOP500 list of the world’s most powerful supercomputers, which is updated and published on the Internet twice a year. The architecture of supercomputers has undergone some major changes during their more than 30-year history. Cray 1, released in 1975, is considered the first real supercomputer. That computer introduced a new feature – the vector processor, which meant a crucial improvement in the speed of processing tabular data. CSC’s supercomputers from the newest to the oldest The 21st century and the PC cluster The PC cluster became the dominant supercomputer architecture in the 2000s. The supercomputer consists of inexpensive general processors of the same type as found in ordinary PCs. The low price makes it possible to increase the number of processors to thousands. The processors are interconnected by either a fast connection network or standard GigabitEthernet LAN technology. 2008: Cray XT4/XT5 and PRACE-prototype CSC's Cray supercomputer was upgraded to over 100 teraflops (trillions of floating point operations per second). This made the new Cray XT5 system at CSC one of the fastest supercomputers in Europe. Later in 2008, a second upgrade was funded under the EU's Partnership for Advanced Computing in Europe (PRACE) initiative with two Cray XT5 cabinets. One of the PRACE project tasks is prototype testing for supporting decisions for the installation of the future HPC Petascale systems in Europe. 2007: Cray XT4 and the HP supercluster Cray XT4, ordered by CSC in 2006 and installed in 2007, uses Opteron multi-core processors manufactured by Advanced Micro Devices (AMD). They are interconnected with Cray’s proprietary connection network with high bandwidth and low latency. In the final assembly of Cray, the number of cores is over 6,000. The theoretical computing power will then be over 70 Tflop/s, i.e., 70 trillion floating point calculations per second. HP ProLiant supecluster, installed in Summer 2007, uses same AMD’s Opteron multi-core processors, but represents the PC cluster design. The number of processor cores is 2,048 and the connection network is InfiniBand. The 1990s and massively parallel supercomputers Massively parallel supercomputers were introduced onto the market in the 1990s. Their performance was based on a large number of scalar processors and fast data exchange between them. A massively parallel supercomputer required a new type of programming, parallel processing. This means that the task to be performed is divided into component tasks, each of which is given to its own processor to perform. The processors communicate with each other as the computation proceeds. The computer’s total power depends on how well the task can be performed as parallel processes; in other words, whether the component tasks can be performed independently of each other or whether much communication is needed between the processes during the computation. 1996: Cray T3E supercomputer In 1996, CSC put to use the Cray T3E supercomputer, which had 192 processors and was a typical representative of massively parallel architecture. The number of processors in the final assembly totaled 544 and the top performance 384 GFlop/s, or 384 billion floating point calculations per second. The first supercomputers used special processors that required considerable cooling power. Liquid cooling was common. The assembly of supercomputers from components intended for office use started in the 1990s. 1994: IBMSP The IBMSP acquired by CSC in 1994 was based on air-cooled SP2 processors. The processors were upgraded later to make them faster and their number was increased. In the final assembly, the IBMSP had 512 processors and its top performance was 2.2 TFlop/s. 1989 and Finland’s first supercomputer CSC’s first supercomputer, Cray X-MP acquired in 1989, had a speed of 0.935 TFlop/s, or 935 million floating point calculations per second.