Europe rides the exascale wave

Over the last few years, we have witnessed an unprecedented boost in the adoption of digital technologies. From high-performance computing (HPC) and big data analytics to artificial intelligence (AI) and machine learning (ML).

Supercomputers, also called computational or parallel clusters, process complex simulations by breaking computational problems into smaller “jobs.” These run simultaneously on multiple server nodes interconnected by a fast network. The price-performance ratio of these systems is constantly improving. Which helps make HPC more affordable. A computing problem that would take weeks to process twenty years ago, on multi-million dollar systems, can be solved in just a few hours on a single GPU-equipped server for computation.

HPC, once only available to educational institutions and large enterprises, has become more affordable. Because public clouds and collective data centers allow smaller organizations to reserve remote computing capacity on demand. They only have to pay for the hours of use per core and related services.

This high-performance computing enables organizations to solve a wide range of complex problems. It includes product optimization and electronic design, credit analysis and fraud detection, drug discovery, simulation of human studies, weather forecasting, etc.

A new era of supercomputing

From 2002 to 2009, supercomputing performance doubled almost every 12 months. However, this rate slowed to every 2.3 years between 2009 and 2019. This is attributed to various factors, such as slowing down of Moore’s Law and technical limitations such as Dennard scaling.

Supercomputers, also called computational or parallel clusters, process complex simulations by breaking computational problems into smaller “jobs.”

However, the technology offers innovative ways to overcome the limitations, ushering in what is called the era of exascale computing. Such a system is one that can perform a quintillion floating point operations per second (FLOPS). That’s a trillion trillion—or 1,000,000,000,000,000—meaning that exascale machines can solve calculations five times faster than today’s best supercomputers, and run more complex models with greater precision, too.

To reach these new heights of performance, engineers take a heterogeneous approach. This consists of the integration of CPU with GPU and iterative optimization of both hardware and software. The goal is to reach new levels of performance and efficiency at a lower cost per FLOPS.

There is no better example to demonstrate this than the Frontier supercomputer, which is under development at the Oak Ridge Leadership Computing Facility in the United States. It is set to make history as the world’s first operational exascale supercomputer when it goes live later this year. It integrates third generation AMD EPYC CPU and Radeon Instinct GPU. With what it will offer more than 1.5 exaflops of maximum processing power. There will also be an even more powerful AMD-based exascale-class system called El Capitan, scheduled to be built at Lawrence Livermore National Labs in the US by 2023.

Japan came in first, with its own top-performing 1.42 exaflops FUGAKU supercomputer, and China reportedly already operates a 1.32 exaflops Sunway “Oceanlite” system, though it is less well known. Where does Europe stand in this race?

The exascale European mission

As China and the United States seek to become world leaders in supercomputing, Europe is taking a more collaborative approach with the European High-Performance Computing Joint Undertaking (EuroHPC), funded by the European government. The initiative pools resources to fund a European HPC infrastructure, with world-class data, and support an innovative supercomputing ecosystem.

The continent’s supercomputing efforts are also bolstered by Horizon Europe. It drives discoveries and world firsts, including the development of exascale machines located in the EU.

Take the Hawk supercomputer as an example, currently ranked 24th on the TOP500 list of the world’s fastest supercomputers. It is installed at the University of Stuttgart (HLRS) and is an HPE Apollo 9000-based system with 5,632 nodes spread over 44 racks. They each run AMD EPYC CPUs and offer around 26 petaflops of peak performance. HLRS enables customers in the automotive industry to perform structural analysis and fluid dynamics simulations.

There is also Lumi, a pre-exascale machine located at the IT Center for Science (CSC) in Kajaani, Finland. With its custom AMD EPYC “Trento” CPUs and four AMD Instinct MI250X GPU accelerators per node, it will be capable of executing more than 375 petaflops or 375 billion calculations per second, with a theoretical maximum throughput of more than 550 petaflops per second. .

Lumi also offers an innovative ‘free cooling technology’, which enables waste heat to be used in the Kajaani district heating network. This further reduces costs and the CO2 footprint. This technology is expected to reduce the annual carbon footprint of the entire city by 13,500 tonnes, an amount equivalent to that produced by 4,000 cars2.

On its way to exascale, Europe already knows the new world of possibilities offered by this unparalleled level of performance. These systems will help solve the most complex questions of scientific research. They will allow scientists to create more realistic models of the Earth system and climate. In addition to promoting new studies of the Universe, from particle physics to star formation.

With access to technology developed by AMD and x86 vendors that’s already used in some of the world’s fastest machines, as well as access to a host of open source software tools to optimize and scale workloads of supercomputing, Europe is already capable of solving complex solutions and taking advantage of exascale computing. However, for Europe to get serious about exploiting its own Exascale-class systems, it will be necessary for several European countries to continue to invest in the development of scalable hardware, tools and software.

By Roger Benson, Senior Director, Commercial EMEA at AMD.