Weather forecasting was a stimulus to use electronic computing. John W. Mauchly wanted to use the then tremendous speed of the ENIAC to solve numerically extensive systems of equations describing the weather. John von Neumann also dedicated himself to this task, and played a major part in the development of computers in the USA. "If it takes longer than 24 hours to compute the weather for tomorrow, we will drop the project", he is alleged to have said.
High-speed computers were therefore required. At the beginning of the 1960s, Seymour Cray of Control Data proved with the CDC 1604 that it was possible to increase computer speed one hundredfold using available components. Shortly after this, he set up business on his own and launched the legendary CRAY-1 on the market. The super-computer was born. Multiple processors performed the unchanging arithmetic steps in parallel to solve systems of equations. For suitable problems, this so-called vector computer was able to achieve a speed that was several times greater than that of a single-processor machine.
When fast microprocessors became available at the beginning of the 1990s, various attempts were made to implement powerful multiprocessor systems. This part of the exhibition shows the vector computer family tree, as well as various structures or topologies on which multiprocessor systems were built.
Simulation was the application area for these high-speed machines. For example, to perform a meaningful car crash test in virtual mode, it is necessary to input a wide range of design data and experimental parameters, and to compare the computed data carefully with experiments that have actually been carried out. As soon as the procedure returns reasonable results, comprehensive trials can be carried out in the design phase. This allows engineers to test the impact of the design on the system as a whole in an early trial phase. Simulation then replaces reality.
We are all benefiting from a vast improvement in weather forecasting these days. Simulations on supercomputers are not only capable of making one-off calculations, but are frequently able to repeat their calculations using a variety of start parameters, in order to check the reliability of the results.