There is a global race to build the first supercomputer in the world capable of performing 1 million trillion operations per second (1 ExaFlops in our jargon!). Extrapolating the performance delivered by TOP500 list suggests that an Exascale system (a system that can reach 1 ExaFlops) could be built by 2018-2019. However, according to the literature of the topic, only China and Japan are focused on building an Exascale supercomputer by 2020. It seems that United States probably won’t build its first Exascale supercomputer until 2023 (Europe isn’t even in the race!).
The reason for the gap, between the extrapolated date from TOP500 list and the expected dates, is because there are big challenges that must be addressed before Exascale systems become a reality. The main issues include: Power consumption, Massive Scalability, Heterogeneity, Integration, Resilience, Programming methodologies and at last but not less important Skills for programming these systems.
For instance, in order to give the reader a magnitude of the challenges, one of the most efficient large-scale HPC system today is Tsubame 2.5 at the Tokyo Institute of Technology, with a peak performance of 5.6 Petaflops and a consumption of 1.4 MWatts. If the current system is scaled to Exaflops it would consume 250 MWatts, which is at least an order of magnitude too much. In my humble opinion the first Exascale system may use exotic technologies to hit the performance target and may be motivated by politics or national interests!
This is an interesting topic to debate and today my students did some presentations on this subject in the Supercomputer Architecture master course (Masters in Innovation and Research in Informatics, High Performance Computing) at UPC. Attached you will find the links to a couple of them that I want to share with you. In my opinion they are excellent presentations that could give you some ideas about the magnitude of the tragedy. Thank you to all students for today’s discussion, especially Andrea Ferri for his presentation “The Meaning of Exascale in 20th Century (link) “ and Aimar Rodriguez Soto for his presentation “When will we have a Exascale computer? (link)”.