What is a supercomputer?
A supercomputer is a machine built to throw an enormous number of processors at a single problem at the same time. The defining feature is not raw clock speed — it is scale, and the interconnect that lets all those parts work together without spending their time waiting.
What makes it different from a server
A data centre runs many independent jobs side by side; losing one machine costs you one job. A supercomputer runs one job across the whole floor, so every part has to stay in step with every other part. That is why the network between the nodes matters as much as the processors, and why these systems are measured on problems that force all of it to cooperate at once.
Modern leaders pair conventional CPUs with GPU-style accelerators, which is where most of the arithmetic now happens. El Capitan, the current number one, runs 11,039,616 cores at Lawrence Livermore National Laboratory.
How speed is measured
The TOP500 ranks machines by LINPACK, a benchmark that solves a large dense system of linear equations. The reported figure is Rmax, the performance actually sustained, in floating-point operations per second. It is an imperfect proxy — a machine tuned for LINPACK is not automatically good at every real workload — but it has been measured the same way twice a year since June 1993, which makes the series unusually comparable over time.
See what an exaflop is for the units, and why Rpeak figures should be treated with suspicion.
What they are used for
- Climate and weather modelling, where finer grid resolution needs more compute
- Molecular and materials simulation, including drug and battery research
- Nuclear stockpile stewardship — the reason several US machines exist at all
- Astrophysics and cosmology, simulating structures too large or slow to observe
- Engineering simulation, from aircraft airflow to combustion
- Training and running large AI models, increasingly the growth driver
Who runs them
The top of the list is dominated by national laboratories and government-funded research centres, because the machines cost more and last less time than most commercial buyers can justify. The current leader is operated by Lawrence Livermore National Laboratory in United States, at 1.742 EFlop/s.