Do they have more/bigger CPUs or GPUs or a different kind of computer architecture or something
In: 5
A supercomputer is a rather vague term today. The days of a single huge machine being a supercomputer (you may have heard the name Cray in the past, a company in the business of supercomputers) are pretty much gone and now “supercomputers” tend to be built from many smaller, more ordinary systems but connected with high speed networking and running the same software/goal/project at once. In the high end space, 100 gigabit network speeds are common and maybe even considered too slow for some jobs.
So I would provide the definition as: a collection of computers collaborating with a shared/single purpose whose computational capacity vastly exceeds what a typical person needs. The measurement of computation is usually FLOPS (Floating Point Operations Per Second). It used to be that counting your power in “teraflops” (Trillions of FLOPS) could be a supercomputer, but now most modern GPUs push a teraflop fairly easily. So… PetaFLOPS now?
Exactly what kinds of supercomputers exist vary. Sometimes you need more RAM. Sometimes you need more CPUs. Sometimes you need more GPUs. Lately with AI, quantity of RAM on the GPU itself has become a concern. Supercomputers are usually built to the needs of the buyer, loading up on the specific resources required. That said, any typical PC or server could be a participant.
If you’ve ever heard of Folding@home, software run by Stanford University to allow any ordinary person to run scientific research on their own PCs for the benefit of real scientists, then your home computer participated in a massive supercomputer project. It received a massive spike in popularity back in 2020 when Covid-19 hit and science turned its attention to it very strongly. I was running a CPU and GPU on it, so I’m part of a supercomputer when I run it.
Latest Answers