Race to exascale: EU enters the HPC supercomputer battle
Wednesday, October 03 2018
The EU’s proposed €1 billion (US$1.2 billion) investment in expanding its high performance computing (HPC) infrastructure over the next five years is vital for the region’s scientific and industrial competitiveness. The United States, China and Japan have already committed to similar initiatives, as part of a global supercomputer arms race to build the first exascale system, capable of one quintillion operations per second. This has become strategically important, alongside the rollout of 5G and cloud infrastructure, given the need for technological supremacy. From an economic perspective, the next generation of HPC will accelerate advancements in research and development across multiple industries such as healthcare and manufacturing, as well as improve climate and natural disaster forecasting. The EU estimates HPC generates €867 in revenue for every Euro spent. It is increasingly critical for national security and the military, in terms of cyberwarfare and the design of new weapons and materials. Exascale systems will also play a key role in the development of new artificial intelligence capabilities, which is another area of competition between the world’s superpowers. The EU’s intentions highlight the growing demand for each power to have its own supercomputing capabilities. This means not only owning the systems, but also the hardware and software intellectual property required to build and operate them.
The EU is a leader in HPC consumption, but lags in capacity
The EU estimates it uses approximately one third of the world’s HPC resources, but only 5% of it is currently located in the region. None of the global top 10 supercomputers are based within the EU member states. This compares to four in 2012, which highlights the underinvestment. Overall, the EU has 19% of the current top 500 systems worldwide, measured by petaflops. Post Brexit, it will only have 14% based on the current systems. It lags behind China, the United States and Japan, and has become dependent on facilities beyond its borders. The risk of industrial espionage and commercial trade secrets, as well as the loss of scientific resources and failing to negate state-sponsored cyberattacks is high. To address this, it has committed to build four petascale HPC systems, capable of one quadrillion operations per second by 2020, and an exascale system by 2023. In addition, it plans to develop the first generation of EU designed and manufactured low power processors, as well as software, including the operating systems and applications that can utilize the processing capabilities. The investment is part of the EuroHPC Declaration, a multi-government agreement designed to pool resources and establish HPC as-a-service for the scientific community, government agencies and the private sector. Currently, 21 nations have signed it including France, Germany, Italy and Spain, but not the UK.
Government interest in the development of exascale systems centers around four key areas:
- Healthcare – Ageing populations present major challenges in terms of increasing healthcare costs and adding pressure to existing constrained resources. By 2050, the World Health Organization forecasts there will be more people over the age of 60 than people aged between 10 and 24. Next generation supercomputers will accelerate DNA sequencing that enables improved diagnosis and genetic screening of diseases. It will lead to new drug discoveries and rational drug design to target specific genes, as well as gene therapy and pharmacogenomics to develop custom drugs for individuals. This will be done at scale, as part of precision medicine. Predicting the best course of treatment based on modelled drug responses will be another key outcome.
- Industry – Innovation through research and development will accelerate industrial transformation, and drive job creation. Advancements in material and predictive science, as well as engineering, manufacturing and electromagnetic simulation will enable retooling and establish new processes to shorten development cycles. This is crucial given the demands on industry to consume less resources, reduce emissions, switch to higher value output and be more competitive on a global level. Industries such as aerospace and automotive will benefit. For example, each new Boeing aircraft put into operation, such as the 737 MAX and 777X, are typically 15% to 20% more efficient than those models being replaced. Continued advancements will enable further cost savings. Addressing energy shortages will be a key area of focus, in particular designing batteries and storage, biofuels and achieving nuclear fusion.
- Environment – Adapting to regional climate changes, in terms of predicting extreme weather events like flooding, droughts, as well as rising sea levels and natural disasters will be enhanced by the development of exascale systems. Improvement in speed and accuracy of climate modelling and weather forecasting will be key outcomes. This in part will be enabled by running multiple interconnected HPC models simultaneously. The efficiency of renewable energy will be optimized, as a result of better simulations. For example, a 1% improvement in wind turbine performance equates to more than US$100 million in annual cost savings. Urban planning will also benefit from this, with integrated building, transport, utilities and waste models producing more detailed data for planners. This is an important area, as the urban population is expected to exceed 60% of the worldwide total by 2030.
- Security – The evolution of warfare and increasing sophistication of cyberattacks has raised the importance of governments to have technological supremacy. Designing new weapons that have greater accuracy and impact will be accelerated with the next generation of supercomputers. Nuclear weapon development will advance, as more sophisticated simulation models can be implemented to offset the restrictions of the Comprehensive Nuclear Test Ban Treaty. Advancements in data processing capabilities will enhance predictions of the evolving nuclear threats of adversaries. Utilizing AI for automating fleets of air and sea drones, combined with enemy modelling, will also be enhanced. This will give nations military advantages and play a critical role in disputed territories. Exascale technology will be crucial in developing new data encryption, as well as identifying and responding to highly complex and coordinated cyberattacks, in particular from state-sponsored agencies.
China, the United States and Japan battle for supercomputer supremacy
China, the United States and Japan are expected to beat the EU’s 2023 target to build an exascale system. In the latest US government budget, funding for the Department of Energy’s (DoE) Exascale Computing Project increased to US$636 million, up from US$376 million in the previous fiscal year. President Obama’s 2015 executive order established the National Strategic Computing Initiative. The United States currently has six of the top 10 supercomputers, including the most powerful system, Summit. This was built for the DoE using IBM Power9 processors and NVIDIA Volta GPUs, with peak performance being approximately one fifth of an exascale system. The United States has a distinct advantage in the race to exascale, as it can utilize established US-based high-end processor and system vendors including AMD, Cray, Dell EMC, HPE, IBM, Intel and NVIDIA. Its first sustained exascale systems are expected in 2022.
The Chinese government had the two most powerful supercomputers up until this year, with its Sunway TaihuLight and Tianhe-2A systems. These were the culmination of its significant investment in building out its national HPC infrastructure. China currently has 41% of the leading 500 systems in production. Development in the country has been hampered by the trade embargo by the United States that came into effect in 2015 banning Intel, NVIDIA and AMD from selling high end CPUs and GPUs to the Chinese government for its supercomputer program. However, the ban stimulated the development of domestically designed and manufactured components, including the Sunway multi-core processors. Developing Chinese made semiconductors is a top priority, falling under the US$22 billion National Integrated Circuit Industry Development initiative, which secured a further US$47 billion investment earlier this year. The Chinese infrastructure vendors, including Huawei, Inspur and Sugon have also emerged. Lenovo has also emerged, as a key HPC vendor on a global level, with wins in each continent. The government has established a network of 17 HPC centers nationwide, and application communities for specific vertical software such as industrial product design, drug design and digital media. China is expected to be the first to have an exascale system by 2021.
The RIKEN Center for Computational Science is leading Japan’s exascale development, which is focused on the Post-K supercomputer. This is being developed in collaboration with Fujitsu, which is building an ARM-based processor rather than using the SPARC64 architecture used to create the original K supercomputer. Its first production exascale system is expected to come online by 2022. Japan has an established HPC sector, with NEC and Hitachi as well as semiconductor firm PEZY Computing.
Each superpower will eventually have its own domestically developed technology
There are major challenges players have to overcome in the development of the next generation of supercomputers, in particular making power consumption and cooling cost effective, as well as developing operating systems and applications that can handle the new capabilities. The EU faces further challenges in terms of the lack of EU-based technology compared to the domestic capabilities of the other economic superpowers. A key project is the €120 million European Processor Initiative (EPI), which aims to build low-powered processors capable of supporting exascale systems. The initiative is not starting from scratch, as it is likely to utilize the current Mont-Blanc 2020 project launched by ARM, the Barcelona Supercomputing Center and Bull, which is part of Atos. Bull is well established in HPC, and will be key the EU’s initiatives. European based processor companies like Kalray and Semidynamics could also play key roles. European efforts are vastly under resourced on a financial basis compared to China. Development of other hardware and software is part of the EuroHPC program. Coordinating all these projects to deliver on time will be challenging. It has a lot of ground to catch up.
The EU will be the last economic superpower to have an exascale system, which will put it at a disadvantage in the short to medium term, both economically and politically. It is unclear if the EU can complete the project independently, without technology from either U.S., Chinese or Japanese vendors. Completing by 2023 and within current costings is highly optimistic. If the project gets completed, the EU will have developed an ecosystem of new indigenous-developed hardware and software technologies that could eventually be widely deployed by governments across the region, giving it the chance to reduce dependence on foreign IT vendors in five years’ time. Given the increasing global trade and geopolitical tensions, the need to establish domestic technology capabilities is critical for each superpower.
For more information or to receive a copy: