Get organized to supercompute – Thot Cursus

Exascale supercomputers are being prepared in Europe. These calculators will be able to operate on the scale of exaflops, i.e. 10 exp18 calculations per second, a billion billion.

Exaflop: 1,000,000,000,000,000,000 calculations/second.

At this scale, the supercomputer can dissipate several million euros of electricity per year, which becomes a significant aspect to consider.

If researchers and engineers are salivating at the prospects these machines suggest, the problems raised by their operation are of several orders:

  • software programming;
  • computing architecture;
  • processors:
    • CPU – central unit processors, like those in our computers;
    • GPU – graphics card processors;
    • FPGA, Programmable logic circuit (Field-programmable gate array) – programmable cards on demand, adapted to HPC algorithms.

  • energy consumption.

“A supercomputer breaks down the tasks to be performed into millions of sub-tasks, each of which is then performed, simultaneously with all of the others”

How to organize all this to obtain the best energy yield and optimal efficiency is the objective of the Textarossa project (Towards EXtreme scale Technologies and Accelerators for euROhpc hw/Sw Supercomputing Applications for exascale).

The teams around this project bring together engineers, mathematicians, programmers, researchers, technicians… it seems obvious to them that the very way they organize themselves will have an effect on their progress.

“The project is also innovative in its conduct. The partners indeed work in a spirit of “co-design”, in which everyone brings skills or ideas, while adapting their own developments to the results demonstrated by others.

It is by distributing tasks and efficiently orchestrating their execution that supercomputers obtain extraordinary computational strength. The members of the Textarossa consortium know this perfectly well… and they intend to apply this principle and achieve the same efficiency, relying on agile project management! »

Inria’s contribution to the project, particularly with regard to the software part and the measurement of energy performance, is already tangible.

Illustration: Inria / C. Morel

For the full article: Textarossa: preparing for the high-performance computing revolution

Learn more about this news


See more news from this institution

Leave a Comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.