Compute enables insights to the future and to visualize the invisible. For this to be possible a mathematical model of the physical system is required. With compute the model is brought to life and this digital reality produces data that can be interpreted and analyzed.
When laptop or desktop computers do not provide the required resources one have to turn towards dedicated compute facilities. The best fit compute resources depends of the nature of the problem. Two main groups are High Performance Computing (HPC) and High Throughput Computing (HTC).
High Performance Computing
The characteristic of HPC calculations is that the solution needs to be found for a domain that needs to be handled as a single entity. Parallel computing resources handle such calculations, which are also known as tightly coupled problems. In general this case requires that compute resources exclusively dedicated to the calculation with a given number of allocated cores. The reason is all cores need to synchronize their state frequently and the slowest core sets the pace for the calculation. Furthermore, if a core stops the whole calculation stops.
High Throughput Computing
The characteristic of HTC calculations is that the solution requires many similar datasets are processed. As a result many similar and independent calculations needs to be executed. Distributed computing resources are well suited for such problems, which also is known as embarrassingly parallel problems. As there is no dependency between the cores it is easy dynamically to scale the numbers of cores contributing to the calculations. It is not fatal if a core crashes as the given dataset easily can be handed to another core.