Supercomputers – future predictions
Future predictions of supercomputing power are presented below. This graph shows how these machines have developed from the early 1990s through to the present day, with a predicted future trend to 2055.
Supercomputers are a type of very powerful computer, sometimes with vastly greater capabilities than a standard desktop machine. Although they can come in smaller sizes, they are probably more familiar to people as large wardrobe-sized cabinets, grouped into a network to combine their processing power.
Supercomputers perform a variety of tasks for researchers, governments, businesses, universities and other institutions. These include highly intensive calculations such as problems involving astronomy, climate research, molecular modelling and physical simulations, quantum mechanics and weather forecasting. Examples of physical simulations include aeroplanes in wind tunnels, nuclear weapons detonations, and research into nuclear fusion energy.
The first supercomputers emerged in the 1960s, designed primarily by Control Data Corporation. This firm led the market into the 1970s, until its co-founder Seymour Cray left to form his own company, Cray Research. He then took over the market with his new designs, holding the top spot in supercomputing for several years. In the 1980s, a number of smaller competitors entered the market, but many disappeared in the mid-1990s supercomputer market crash.
The preferred metric for supercomputing performance is called floating point operations per second (FLOPS). Intel's ASCI Red became the first machine to exceed one teraFLOPS (a trillion calculations per second) in 1997. It had a peak speed nearly 800 times faster than the Cray-2, released just 12 years earlier in 1985. IBM's Roadrunner achieved one petaFLOPS (1,000 trillion calculations per second) in 2008, or about 1,000 times faster than ASCI Red.
Since then, supercomputers have continued to improve by orders of magnitude, though at a somewhat slower rate. The slowdown is clear from the second half of 2013 onwards, according to the TOP500 list maintained by researchers in the U.S. (which uses the LINPACK benchmark) and on which the above graph is based.
Nevertheless, the very fastest machines will soon reach the next milestone: exaFLOPS. Or put another way – 1,000,000,000,000,000,000 floating point operations per second. Several contenders are waiting in the wings, one of which may claim the title in the near future.
Oak Ridge National Laboratory in Tennessee is developing a $600m machine called Frontier, with a targeted performance of 1.5 exaFLOPS, predicted to arrive by the end of 2021. Meanwhile, the Argonne National Laboratory on the outskirts of Chicago is developing Aurora, designed by Intel and Cray, comprising more than 200 cabinets with a combined 1 exaFLOPS. This had originally been predicted as the world's first exaFLOPS machine but is now delayed due to chip manufacturing issues with Intel, and will therefore arrive after Frontier.
Although China currently dominates the TOP500 list, with 188 machines (versus 122 for the U.S.), it only has two listed in the top 10. However, China has been developing at least three designs for exascale supercomputers, as it seeks to reestablish its leadership. Successors to the Sunway TaihuLight and Tianhe-2A are likely to emerge soon, with prototypes said to be at an advanced stage.
Japan is already home to the world's fastest supercomputer – Fugaku – which has 442 petaFLOPS of performance. The first exaFLOPS machine will therefore need at least 2.3 times that speed. Electronics giant Fujitsu and the RIKEN institute are collaborating on a successor to the K computer, which held the title of world's fastest computer in 2011 and became the first machine to reach 10 petaFLOPS. Their "post-K" supercomputer is predicted to arrive in 2021 or 2022.
Later this decade, and into the 2030s, it seems safe to predict that supercomputers with tens and even hundreds of exaFLOPS will emerge, since the trend in computing power is exponential. Using the above graph, we can also predict the milestone of zettaFLOPS, a supercomputer capable of a billion trillion (1,000,000,000,000,000,000,000) floating point operations per second. The most likely date appears to be 2036. Such a machine would be powerful enough to generate completely accurate weather forecasts over a two-week period.
Looking even further into the future, we arrive at yottaFLOPS, a computing milestone that may occur in the early 2050s. A yottaFLOPS supercomputer would be 1,000 times faster than a zettaFLOPS machine, and a million times faster than an exaFLOPS machine. To give some perspective – a large-scale simulation that might have taken six months to run during 2021 could be completed in just 15 seconds by a yottaFLOPS supercomputer.
The applications of such a machine could enable a whole new plateau of possibilities, which are part of the 'Singularity' that many futurists predict. We can only imagine the sort of world-changing discoveries that will flow from this technology and its exponentially greater capabilities. Perhaps we will recreate the earliest moments of the Big Bang in minute detail, for example. Or maybe a new material will be developed at the nanoscale that is radically more advanced than anything in existence today. Having already modelled the human brain in earlier decades, researchers may use a yottaFLOPS machine to run millions of these simulations together in real time, to study the dynamics of entire populations.
An even more exciting (and frightening) possibility is the emergence of the first super AI, with a supercomputer for its brain. In one scenario, predicted by futurists such as Life 3.0 author Max Tegmark, this entity might seek to escape its containment and proliferate around the web. In what sounds like the plot of a science fiction movie, it could then download itself into robot bodies. Whether this will be good or bad for humanity remains to be seen.
"I think if we succeed in building machines that are smarter than us in all ways, it's going to be either the best thing ever to happen to humanity, or the worst," said Tegmark, during the book's launch in 2017. "I'm optimistic that we can create a great future with AI, but it's not going to happen automatically. It's going to require that we really think things through in advance, and really have this conversation now."
Performance Development, TOP500:
Japanese supercomputer is world's fastest, Future Timeline Blog:
Life 3.0: Being Human in the Age of Artificial Intelligence, Future of Life Institute:
Posted: 9th August 2021. Last updated: 9th August 2021.