Supercomputing has become an indispensable asset for nations’ prosperity and security. In this timely Q&A, Hyperion Research Senior Adviser Steve Conway surveys the rise of so-called indigenous or sovereign HPC programs across the world. He examines the efforts of the most active regions, including the United States, which now finds its supercomputing leadership challenged by other countries and alliances. With the Russia-Ukraine war portending the end of globalization, the global state of domestic advanced computing has rarely been more relevant.
HPCwire: Initiatives to develop indigenous HPC technologies in pursuit of HPC sovereignty are gaining momentum around the world. Why?
Steve Conway: Governments have long recognized HPC’s importance for advanced research, but today governments also see HPC as crucial for economic competitiveness. That’s a major change. China, Europe and Japan, among others, are pursuing indigenous HPC technology initiatives because they believe HPC leadership is a prerequisite for economic as well as scientific and engineering leadership. Because political relations among countries can be uncertain, China and Europe don’t want to remain heavily dependent on the U.S. or other foreign sources for this critical technology. They want to attain HPC sovereignty by fostering homegrown technologies and supply chains. Japan’s situation is different because Japan has been an HPC technology leader for decades.
HPCwire: Can you summarize the main initiatives and the context for each of them?
Conway: I won’t do justice to these initiatives but let me try. The initiatives in China, Europe and Japan have some common elements, especially the central role assigned to developing indigenous processors, mostly based on industry-standard technology that can support broad sales by domestic HPC system vendors. Deep down, the global exascale race also turns out to be a processor race.
HPCwire: Let’s take the initiatives one at a time, starting with China.
Conway: China has made enormous progress in designing and deploying leading supercomputers featuring indigenous processors, and China may have deployed the world’s first exascale supercomputers but not yet reported them for Top500 listing. China’s semiconductor fabs are a generation or more behind global leaders, however, and the Made in China 2025 Plan, whose goals include 70 percent self-sufficiency in semiconductors by 2025, appears to be significantly behind schedule. U.S. government restrictions against importing foreign semiconductors have motivated China to accelerate indigenous development, but it will take some time for China to achieve semiconductor self-sufficiency in HPC or in other markets.
HPCwire: What about Europe?
Conway: Europe has also made great progress toward becoming one of the global leaders in producing and consuming HPC technology in the exascale era. Europe promises to be among the global leaders. Especially impressive are Europe’s achievements in organizing more than two dozen nations to pursue this goal through the EuroHPC Joint Undertaking, which among other things has already deployed eight or so pre-exascale supercomputers. Initiatives such as ETP4HPC and the European Processor Initiative have advanced indigenous technologies. Europe is poised to become the global leader in democratizing exascale technologies through its Centers of Excellence and SMB initiatives aimed at scientific and industrial researchers in participating countries. Europe doesn’t see HPC sovereignty as doing everything alone and is likely to continue collaborating with non-European HPC suppliers as the situation demands. Processor sovereignty will take a while but is well under way.
HPCwire: And finally, Japan.
Conway: Japan’s indigenous technology initiative is different. In the earlier, “hot box” era of proprietary vector supercomputers, Japan and the U.S. vied for global leadership. Japan was slower to shift to cluster architectures based on x86 processors and other industry-standard technologies. This caused Japanese system vendors to largely pull back from non-domestic markets. Japan’s government then followed a pattern of providing substantial funding about once a decade to develop a top-ranking supercomputer, such as the NEC Earth Simulator, first on the Top500 list from 2002 to 2004, and the Fujitsu K computer, first in 2011 and 2012. Fujitsu’s Fugaku system has headed the list since June 2020 and reflects a major shift toward higher-potential microprocessor technology, in this case Arm64, that could support significant adoption beyond Japan. Already multiple government sites in the U.S. and Europe are exploring Fugaku systems, and Riken is broadening its non-Japanese user community by awarding time on the number one supercomputer to international research teams.
HPCwire: Russia is one of the nations that was developing a native HPC ecosystem. What impact will Russia’s invasion of Ukraine have on indigenous HPC projects? What are the considerations there?
Conway: Russia’s government has acknowledged the importance of supercomputing but hasn’t made adequate funding available for a world-class exascale program or national HPC technology initiative, although innovation has continued at vendors such as RSC and others. Russia has four systems in the top 50 of the November 2021 Top500, all of them using processors from AMD and Nvidia, who along with Intel, TSMC and Global Foundries have suspended exports to Russia in compliance with U.S. restrictions following the invasion of Ukraine. Russia’s post-invasion disconnect from the SWIFT global financial messaging service is another problem, especially for selling outside of Russia. No quick workarounds seem available to Russian HPC vendors, who may need to revert temporarily to older process technology.
HPCwire: Some of these efforts were motivated by a desire or necessity to not be dependent on U.S. supply chains. Does the United States also have an “indigenous” HPC initiative even if it isn’t explicitly framed that way? Can you discuss some of the ways that the U.S. protects its interests?
Conway: In effect, the U.S. has had a continuous indigenous HPC technology initiative since the 1960s, driven by the defense and other government sectors, and more recently also by growing recognition of HPC’s importance for America’s economy. The 2021 CHIPS for America Act aims to secure the semiconductor supply chain, especially given that 88 percent of the U.S. semiconductor supply in 2020 was fabricated outside of the country. Related to this goal, TSMC and Samsung are building fabs in the U.S. as well as in Europe. Intel will now fab for outside customers and is increasing its footprint in Europe with plans for a large fabrication facility in Magdeburg, Germany. All of this says that the U.S. government recognizes that the global HPC and mainstream IT markets have become more competitive.
HPCwire: How realistic is the goal of HPC sovereignty? Doesn’t every country have some external supply chain dependencies? What about foundries specifically?
Conway: Today and for the foreseeable future, no country or global region can expect to have a fully independent HPC supply chain. Indigenous semiconductor production is typically seen as a cornerstone of HPC sovereignty, yet every major processor development initiative depends on non-domestic sources for some crucial supplies or capabilities, such as manufacturing in Taiwan and South Korea, lithium from China or advanced lithography equipment from the Netherlands. It follows that the goal should be supply chain security, ensuring that the supply chain consisting of domestic and non-domestic sources is as uninterruptible as possible.
HPCwire: Has the HPC community seen anything like this before? The ancient Romans said, “there’s nothing new under the sun.”
Conway: Yes, the world has seen this before. A concerted U.S. effort overtook the UK’s worldwide lead in the early computer era. Japan’s initiatives have challenged America’s HPC position since the 1980s. In the early 1990s, India began its Param supercomputer initiative that continues today, and China’s fast rise in HPC is due to a government push.
HPCwire: How successful have the indigenous HPC technology initiatives been so far? Are they on track?
Conway: HPC leadership takes years of continuous investment and is a never-ending effort. Chinese vendors are producing impressive leadership-class supercomputers but the effort to replace foreign semiconductors in China’s HPC and other IT markets is behind schedule, as I mentioned earlier. Europe, especially via the EuroHPC Joint Undertaking, has made very impressive progress and is positioning itself to become one of the global leaders in the exascale era. Given the enormous sums needed for new semiconductor facilities, Europe is wisely continuing to collaborate with non-European chipmakers while ramping up indigenous semiconductor innovation. Japan has long been an HPC leader and is taking important steps needed to increase global sales and research collaborations. The U.S. is responding to growing global competition through government initiatives such as the CHIPS Act and also by establishing more R&D facilities in Europe and elsewhere.
HPCwire: What are some of the challenges these initiatives still face?
Conway: I’ve mentioned some, but among the biggest challenges faced by some HPC vendors based in Europe and the Asia-Pacific region is gaining more global HPC market experience. To compete globally, you need access to as much of the global market as possible, because global revenue provides needed scale and the customer cultures and requirements can differ substantially from place to place. These varied requirements need to be designed into your products. Vendors that have sold HPC products only or mainly domestically are therefore at a temporary disadvantage. Related to that is the challenge of protective barriers, especially for European HPC vendors, because barriers are higher in the U.S., China and Japan. Addressing these so-called market asymmetries is a government-to-government matter.
HPCwire: How do you think these initiatives will play out in the global HPC market, especially with regard to competition and collaboration?
Conway: The next few years at least promise to be a time of transition, where these initiatives increasingly balance researchers’ need for uninterrupted access to the most capable supercomputers no matter where in the world they come from, and the goal of strengthening indigenous supply chains, which may mean favoring domestic suppliers in procurements. During this transition period, when some indigenous technologies and products under development may not be ready to compete with the world’s best, I would expect solutions that combine domestic and foreign elements as needed. Keep in mind that some non-domestic vendors, especially in the processor realm, are investing heavily in local facilities that should better enable them to qualify as domestic vendors in government procurements.
HPCwire: How can countries with smaller economies compete?
Conway: Countries compete and align themselves with other countries in certain areas such as defense, but where HPC’s main function is concerned, research, the HPC community is increasingly global. HPC’s main contributions, such as severe weather prediction, safer vehicles and better energy exploration, have benefited humanity as a whole. So, the more important challenge is ensuring that researchers in countries of every size have access to HPC resources. The EuroCC program will establish HPC competence centers in 33 participating European countries, and Japan’s RIKEN is also broadening access for researchers in other countries of all sizes. Over time, this expanded use of HPC will benefit both science and domestic economies.
HPCwire: Summing up, then, how do you see these indigenous technology initiatives?
Conway: They should ultimately increase competition and innovation in the global HPC market, giving buyers and users more choices. There may be temporary disruption as some governments learn to balance researchers’ needs with the goal of strengthening indigenous supply chains. In an ideal world, HPC competition would be free, fair and global, without the need for protective barriers, but the real world is politically divided. It’s reassuring that indigenous HPC technology initiatives aim to foster rather than restrict global research collaboration.