What is the work of supercomputers

SupercomputersEconomical but complicated

Supercomputers are currently taking an evolutionary step. So far, they have often consisted of thousands or tens of thousands of processors, called CPUs. In the future, however, they are likely to be replaced by GPUs: graphics processors. These are particularly good at parallel computing and use less power than CPUs.

"We'd rather put money in our heads than in electricity bills," says Thomas Schulthess from ETH Zurich and the Swiss national supercomputing center.

He built a new supercomputer for the MeteoSwiss weather service, based on GPUs. The new computer is 40 times faster than its four-year-old predecessor, but uses 30 percent less energy. The operating costs are therefore reduced. Schulthess invests part of the money saved to solve another problem:

"Many computer systems and high-performance computers have a tendency to be very archaic in the way they are programmed. And that shouldn't be. You really have to work on it to change that."

Unnecessarily complicated programming environments

In his view, supercomputers make it unnecessarily difficult for scientists to run their simulations on them. The use of GPUs instead of CPUs requires extra work anyway: Programs that run on older computer architectures have to be laboriously adapted for the new architectures.

In addition, scientists first develop their simulations on a smaller computer and then have to scale them up for supercomputers. This is where complicated math comes into play. All of this cannot be avoided. But the fact that the researchers then have to deal with the uncomfortable programming environments of the supercomputers at the same time should be a thing of the past if Thomas Schulthess has his way. He sees the operators of data centers and supercomputers having an obligation:

"The goal is that you have these environments that you already have on your laptop today, that you can use the same environments on supercomputers. To make the transition as easy as possible. The goal must be that people do not use them fight the software environment, but fight with the algorithmic problems. Because: they need people's creativity. "

Little investment in user-friendly software environments

As with development environments on smaller computers, scientists should be able to fall back on building blocks, mathematical libraries, when they use supercomputers. So far, they have had to reinvent the wheel too often. But data centers would shy away from investing in software environments that are easier to use. Schulthess did it differently with his Swiss supercomputer. He invested part of the electricity costs saved by GPUs in software development.

He says, "the software investments we've made are high, but are still modest compared to the hardware investments."

Those who work with the computer can concentrate more on the actual science instead of dealing with the unfamiliar system. This not only makes the supercomputer more efficient, but also makes working with it more efficient.

Science in focus on Sunday, August 14, 2016 at 4:30 p.m. is about what knowledge scientists gain with the help of computer simulations. Title: "Research in the Matrix - How Simulations Explain the World".