They are optimised for different things. A mainframe is built for throughput and reliability — processing enormous numbers of transactions without stopping. A supercomputer is built for raw calculation speed on a single large problem. One is measured in transactions per second, the other in floating-point operations.
The core difference
- mainframe — maximum input and output, maximum uptime. Built to process transactions continuously for decades.
- supercomputer — maximum computation. Built to run one enormous calculation as fast as possible.
What each is actually for
Mainframes run banking, insurance, airline reservations, payroll and tax systems. The workload is billions of small operations that must all be correct, with the machine never going down. Redundancy is built into everything, and components are designed to be replaced while the system is running.
Supercomputers run climate models, molecular dynamics, nuclear simulation and fluid dynamics. The workload is one vast problem split across many processors. If the machine stops, the job restarts from a checkpoint — irritating, but not catastrophic in the way a failed payment system is.
The measurement tells the story
Supercomputers are ranked by FLOPS, floating-point operations per second, and the TOP500 list has tracked them since 1993. Mainframes are not usually compared that way at all; the relevant figures are transactions per second and availability, quoted in nines — five nines being roughly five minutes of downtime a year.
Why mainframes did not disappear
They were predicted to be obsolete for decades and are not. A very large share of card transactions still passes through them, and much of the software is measured in decades of accumulated correctness. Replacing a system that has been right for forty years carries risk that is hard to justify, and the machines themselves have kept modernising.
Where the words come from
Mainframe originally meant the physical cabinet holding the central processing unit, as distinct from the peripheral units around it. The name for the frame became the name for the class of machine.
Supercomputer is straightforwardly descriptive, in general use from the 1960s. Modern supercomputers are clusters of many ordinary processors rather than exotic single machines, so the word now describes scale and interconnect rather than a special kind of chip.
Frequently asked questions
What is the difference between a mainframe and a supercomputer?
A mainframe is built for transaction throughput and continuous uptime. A supercomputer is built for raw calculation speed on a single large problem.
Are mainframes obsolete?
No. A large share of card transactions and core banking still runs on them, largely because decades of proven correctness are expensive to risk replacing.
How are supercomputers measured?
In FLOPS, floating-point operations per second. The TOP500 list has ranked them on that basis since 1993.