We live in a digital world where a country’s computing prowess is a key element of its power and progress, but India seems to have been left far behind in the fastest supercomputer race.
Meanwhile, a virtual race is on among developed countries like US, China, European Union and Japan for the past few years to build the next generation exascale supercomputer with computation power of, at least, one exaflop or a quintillion (a billion billion or 1,000,000,000,000,000,000) calculations per second.
Exascale supercomputers, combining artificial intelligence with simulation capabilities, will hopefully help scientists solve the grand challenges faced by humanity in fundamental science, be it in the field of particle physics, genomics or artificial intelligence.
By their high predictive power, the machines can simulate nuclear explosions, make far more accurate weather predictions and better models of ocean currents. As brain computation is supposed to take place at exascale, the machines are expected to give a boost to brain research too.
The fastest supercomputers in the world today give a petaflop performance, doing 1 quadrillion (1,000,000,000,000,000) calculations a second, much lower than what exascale computers will be capable of.
In the past few years, China has aggressively entered the field of supercomputers and now dominates the world, owning over 45 per cent of the world’s fastest supercomputers, leaving US far behind with just about 20 per cent of the fastest computers.
The US, however, holds the two top slots in the fastest supercomputers of the world, with Summit, an IBM-built system at the Oak Ridge National Laboratory (ORNL) in Tennessee, USA, having a performance of 143.5 Pflop per second and super computer Sierra, at Lawrence Livermore National Laboratory, having a speed of 94.6 Pflop/s.
Today’s fastest supercomputer, Summit, takes up 5,600 square feet of floor space and has nearly 200 miles of cable. It uses 4,000 gallons of water per minute to stay cool and consumes enough power to run 8,000 homes.
China comes third with Sunway TaihuLight, a system developed by China’s National Research Center of Parallel Computer Engineering & Technology (NRCPC), and installed at the National Supercomputing Center in Wuxi with a capacity of 93 Pflop/s.
A close fourth supercomputer is also from China, the Tianhe-2A (Milky Way-2A), developed by China’s National University of Defense Technology (NUDT) and deployed at the National Supercomputer Center in Guangzho, China, with 61.4 Pflop capacity.
China has been No 1, on the global Top 500 list since the launch of Tianhe-2 in 2013.
China, which held the top slot for five years, having been displaced by the US in 2018, wants to get back in the race and plans to invest billions of dollars for having an exascale (that can execute a quintillion calculations a second) supercomputer.
Three proto-type exascale computers in China are already being developed by three teams led by the National Research Center of Parallel Computer Engineering Technology (NRCPC), Dawning Information Industry, and the National University of Defense Technology (NUDT).
The US has its Exascale Computing Project with the goal of launching an exascale computing ecosystem by 2021.
Although the US has dominated supercomputing for many years, China has been No 1 on the global Top 500 list since the launch of Tianhe-2 in 2013. Located in the National Supercomputer Centre in Guangzhou, Tianhe-2 was built by China’s National University of Defence Technology.
In view of China’s plans and that of European Union and Japan towards building their own exascale supercomputers, the US, a few days ago, queered the pitch by coming out openly and revealing details of its plans to build Aurora, the exaflop performance computer of US, by 2021, with Intel and Cray.
Today’s supercomputers are made up of thousands of connected processors, and their speed has grown exponentially over the past few decades.
The first supercomputer, manufactured in 1964, was called the CDC 6600. It used a single processor to carry out 3 million calculations per second. While that may sound impressive, it is tens of thousands of times slower than an iPhone.
In the European Union, EuroHPC (European High-Perfomance Computing Joint Undertaking) is pooling efforts of different European Union countries to build exascale super computers (performance between 150 Petaflops and 1 Exaflops), at a total budget of around 1,4 billion Euros. The target is to build, by 2026, a world-class European supercomputing infrastructure and initially have at least two pre-exascale systems by 2020 at a total budget of around 500 million Euros.
A consortium of Finland, Sweden, Denmark and Norway Switzerland, Czech Republic, Belgium and The Netherlands will set up a machine at CSC – IT Center for Science, in Kajaani, Finland.
As one country may not be able to achieve the target, many countries are coming together with the common purpose of a common good.
Finland is chosen for competitive advantages, including energy and cost efficiency, societal and geographical stability and high security standards.
EuroHPC will be developing exascale supercomputers based on competitive European technology, that could be acquired around 2022-2023, and would be ranked among the top three supercomputers in the world.
Italy, Portugal and Spain are also contenders for building pre-exascale supercomputers in Europe. Germany and France are also looking at exascale computers.
The goal is a supercomputer capable of 1 exaflop — 1000 petaflops — five times faster than the current champion, the Summit supercomputer at Oak Ridge National Laboratory in Tennessee.
Japan’s Post-K exascale computer could be running in 2020. The United States aims at deploying its first exascale system at the Argonne National Laboratory in Lemont, Illinois, in 2021. The European Union is ramping up its own programme. China is aiming for 2020, but the date may slip.
Japan, which used to be at the top, has in a way given up the race for the top slot in building the fastest computer. Japanese companies feel that number crunching or how fast one can do calculations may not be the only criteria to judge a computer. They are trying to build supercomputers which may be slower in calculations but are more efficient in energy conservation and in their ease to use.
Riken officials reported development of a CPU that will be the brain of its Post-K, which will succeed Japan’s leading supercomputer K in 2021
Riken, which is developing the Post-K with Fujitsu Ltd, is looking to claim the world’s top spot overall, based on user-friendliness and other features – not in simple calculation speed.
The Post-K will be able to conduct calculations for practical purposes, such as simulations to study protein changes in the body for drug development, up to 100 times faster than the K supercomputer, while requiring only three times more electricity.
It will be useful in predicting earthquakes and tsunamis that would affect cities across an area measuring dozens of kilometers square as well as developing new materials and studying the creation of the universe.
The CPU for the Post-K has been designed in Japan, but manufactured by a non-Japanese company.
According to Riken and Fujitsu, it will be able to perform 2.7 trillion calculations per second, nearly double the speed of the world’s fastest CPU made by Intel Corp of the United States and will also be one of the world’s most energy-efficient.
The K supercomputer in 2011 topped the TOP500 global ranking, which is based on simple calculation speed, with CPUs produced in Japan.
A total of 43 supercomputers developed by NEC Corp., Fujitsu and Hitachi Ltd. made the TOP500 list in 2002, when NEC’s Earth Simulator topped the ranking.
But a little more than 10 Japanese supercomputers are now on the list, while Chinese models have increased their presence.
While the No 1 slot was long held by either US or Japanese machines, a Chinese supercomputer clinched the top slot in 2010 for the first time. Beijing also topped the list between 2013 and 2017.
Matsuoka said the Post-K, which will be able to perform several hundred quadrillion calculations per second, may top the TOP500 list when it starts operations in 2021.
The performance of supercomputers is said to increase in geometric progression, but the improvement rate has been sluggish.
The calculation speed of the supercomputer at the top of the TOP500 list improved 39 times from 2001 to 2006 and 29 times from 2006 to 2011. But the performance increased only 11 times over the five-year period through 2016.
The best supercomputer in 2021 is expected to perform several hundred quadrillion to 1 quintillion calculations per second, compared with 100 quadrillion for the top model in 2016.
Fujitsu will produce the Post-K CPU outside Japan, but the assembly process will be conducted at a factory in Ishikawa Prefecture.
Comments are closed.