This article provides a big picture description of what high performance computing (HPC) is, who are doing it, what they do with it, and where are we positioned in this big picture. We use a superfast train as the icon to express the speed of a HPC system.
What is HPC?
HPC is Parallel Computing which aggregates the computing power of multiple units of machines. There are two approaches for parallel computing. The first is homogeneous based on multiple units of the same type of processors. The second is heterogeneous based on a combination of CPU and GPU.
Who does HPC?
There is an organisation http://www.top500.org/ that compiles data on the top 500 performance systems in the world. This organisation publishes the Top 500 list every 6 months and measures system performance with a hardware benchmarking application called LINPACK. LINPACK is an open source application and a standard, and anyone can download it to measure the performance of his own system.
The single most important performance indicator produced by LINPACK is FLOPS standing for floating point operations per second. This is done on Double Precision (16 digits) basis as against Single (7 digits). On the Top 500 list, we can see the ranking of systems based solely on FLOPS. The list shows also the number of cores in the system and the total electricity power consumed by the system. We are able to rank systems based on FLOPS per core and FLOPS per Watt.
What is HPC for?
HPC is for problems that take a time period to compute longer than we desire. The faster the HPC system the more desirable would it be. Space, price and electricity consumption are the restraints. Such problems exist both in the academic/research world as well as in the commercial world. Compucon uses the terms Digital Concept Creation and Digital Content Creation to describe the 2 main camps of applications.
(Hover mouse over to enlarge images)
SuperMUC, Top 4 in the world as of June 2012, is being used to develop Cave Automated Virtual Environment for example. “Pirates of the Caribbean” was produced with HPC system amidst a US$180 million budget as another example. AMBER taking 1 day to simulate 64ns of bio-molecular behaviour is the 3rd.
Compucon could be the first local hardware technology company in New Zealand that has systematically attempted to master the eco-systems for heterogeneous parallel computing along the Nvidia CUDA direction.