December 16, 2020
We are used to describing a computer’s performance based on its processor speed and size of random-access memory, or RAM. Supercomputers, though, are particularly powerful mainframe computers that run computationally intensive applications and serve many high-performance computing (HPC) clusters. What other features make up a supercomputer?
If we were to give you a virtual tour of a supercomputer, you would notice these features that set a supercomputer apart from its commercial counterparts:
Main features of a supercomputer
A vast number of processing units
Today’s supercomputers feature several hundreds of thousands, some up to a million, of processing units, CPUs or GPUs depending on the architecture, working in unison using massively parallel processing. Often called capability computing, this is the one feature that sets supercomputers apart from the average computer. This particular feature rendered the old computer performance metric, processor speed, obsolete. The performance of a supercomputer is measured in Floating Point Operations Per Second, in short FLOPS. Mathematically speaking, floating-point operations are truly multiplications and divisions since they require significantly more computational power than additions and subtractions. As of late, HPC clusters are also boasting impressive scalability and processing power.
An immense collection of RAM-type memory units
The sheer size of a supercomputer’s random-access memory (RAM) is only the first piece of the memory puzzle. Supercomputers’ RAM modules, although distributed across many nodes (one can think of a “node '' to be a super performant computer with many processing units or cores), can be treated as a single pool. The ability of the RAM modules to work in unison is achieved with specialized software, usually called middleware. In many cases, supercomputers feature several terabytes of RAM. Another unique feature of the supercomputers’ memory, besides its size, is that it is addressable by all processing units, enhancing the ability of a supercomputer to solve large mathematical problems.
High-speed interconnect between nodes
The large number of nodes (think of nodes as a massive computer on a rack) that make up a supercomputer communicate via switches that operate at high speed. To ensure a high rate of data throughput, the nodes are usually connected in a non-blocking, fat-tree topology delivering not only up to 200Gb/s bandwidth between nodes but also in-network computing acceleration for communications frameworks (MPI). Specialized high-speed interconnect (bandwidth and latency) are quite the trademark of supercomputers and are critical to enabling effective use of their large number of CPUs and RAM.
High input/output and file systems speeds
High-speed computations that supercomputers feature is also complemented by equally powerful data writing and reading capability. This is achieved with parallel file systems such as Lustre or GPFS. Speaking of speed, another important supercomputer feature is quick read/write access to data.
Custom software and specialized support
Most of us, when imagining a supercomputer, visualize only the machine. One of the most underrated “features” of a supercomputer is the staff that serves it. Troves of brilliant programmers and IT staff work tirelessly to devise innovative solutions that ensure the best computing performance and achieve high production levels from these complex systems.
Effective thermal management
Innovative ways of cooling the processing units and auxiliary components help supercomputers maintain high processing efficiency. Thermal management of supercomputers involves intricate and efficient liquid cooling systems, hot water cooling, and immersion cooling. Keeping the processors below their throttle temperatures ensures that all CPUs from all nodes are always available at full capacity. Their processors’ performance is at its peak if their core is not overheating due to crowding, improper ventilation, or ineffective cooling.
Now that you know what features make a computer a supercomputer, learn what supercomputing is and what supercomputers are used for including weather modeling, pandemic spread, aerodynamics, protein denaturation, or global warming, making them extremely valuable, worth their weight in gold.