As China continues to prove its supercomputing mettle via the Top500 list and the forward march of its ambitious plans to stand up an exascale machine by 2020, one-to-two years ahead of other world tech leaders, there has not yet been a decisively competitive counter from the U.S. Many government and industry representatives we’ve spoken with are quick to express concern (usually off-the-record) about the United States’ positioning within the pack of global supercomputing powers, but official channels haven’t been very forthcoming.
A U.S. government report, published by the Networking and Information Technology Research and Development (NITRD) program, reflects a new willingness to speak openly about the increased global pressures that impact the competitiveness of U.S. supercomputing and the implications for the nation’s economic prosperity, scientific leadership and military security.
The report summarizes the findings of a joint DOE-NSA Technical Meeting held September 28-29, 2016, which brought approximately 60 experts and leaders from the U.S. HPC community together to assess U.S. plans in light of China’s standing up the number-one ranked 93-petaflops Sunway TaihuLight three months prior (in June 2016). As we know, that was just the latest show of strength. China had already held the Top500 spot with Tianhe-2 for six iterations of the list from June 2013 through November 2016. With the addition of TaihuLight, China now claims the number one and number two systems, which provide the list with nearly 19 percent of its total FLOPS. Arguments that question the utility of China’s FLOPS-centric approach in relation to the US, EU and Japanese focus on “sustained application performance” may have merit, but only to a point.
The Chinese machine is not a stunt machine; the report is explicit about this:
“It is not a stunt. TaihuLight is a significant step up in performance for China; indeed, its 93 petaflop/s is significantly greater than the aggregate flops available to DOE today – Titan, Sequoia, Mira, Trinity (Haswell), etc. More importantly, where previous Chinese HPC systems were unimpressive except for running benchmarks (i.e., LINPACK tests), TaihuLight is being used for cutting-edge research, with real-world applications running well at scale. This year, three of six finalists for the Gordon Bell competition (see Appendix A) are Chinese efforts; China has never been a finalist before.”
Meeting attendees were also impressed by China’s homegrown processor architecture and the innovative nature of the design, the report notes.
It terms of both Top500 system share and aggregate performance share, China is now on par with the U.S., which prior to 2016, the U.S. had a clear lead going all the way back to the first Top500 ranking in 1993.
The report reflects this sentiment: “These results indicate that China has attained a near-peer status with the U.S. in HPC. The U.S. asserted its intention to maintain a leadership position in HPC in the July 2015 Executive Order establishing the National Strategic Computing Initiative (NSCI). It is now clear that future U.S. leadership will be challenged by the Chinese. The 2012 Net Assessment of Foreign HPC noted the aggressive development of Chinese HPC capabilities, and, in particular, the accelerated rate of investment that China was making in these areas.”
National security was high-priority topic. “HPC play[s] a vital role in the design, development, and analysis of many – perhaps almost all – modern weapons systems and national security systems,” the report states, and concludes “[national security] requires the best computing available[;] loss of leadership in HPC will severely compromise our national security.”
But the report authors also provided a thoughtful statement on the nature of China’s motivations:
Participants [especially those from industry] stressed that their personal interactions with Chinese researchers and at supercomputing centers showed a mindset where computing is first and foremost a strategic capability for improving the country: for pulling a billion people out of poverty; for supporting companies that are looking to build better products, or bridges, or rail networks; for transitioning away from a role as a lowcost manufacturer for the world; for enabling the economy to move from “Made in China” to “Made by China.
Having said that, their focus on using HPC systems and codes to build more advanced nuclear reactors and jet engine suggests an aggressive plan to achieve leadership in high-tech manufacturing, which would undermine profitable parts of the U.S. economy. And such codes, together with their scientific endeavors, are good proxies for the tools needed to design many different weapons systems.
Recommendations
The NSCI, and by extension the Exascale Computing Program, is central to the United States’ plan to ensure its global economic, scientific and military competitiveness. The post-Moore’s Law challenge, led by IARPA and other agencies, is another key part of the U.S. strategy.
“Leadership positions, once lost, are expensive to regain. To maintain U.S. leadership in HPC, a surge of USG investment and action is needed to address HPC priorities,” the report states.
Further,
The 2012 Net Assessment of Foreign HPC noted a divergence in R&D investment between the U.S. (slowing) and China (accelerating). Objectives #1, #2, and #3 of the NSCI [refer to Appendix E in report] can be seen as identifying the USG investments necessary to support a healthy HPC ecosystem for the next 30+ years. A notional timeline for the impact of these investments is the following:
+ Today to 2025 – HPC ecosystem nurtured by USG investments to reach a capable Exascale system
+ 2025 to 2035 – HPC ecosystem takes advantage of USG leadership in architectural innovations (described below)
+ 2035 and beyond – HPC ecosystem endures because of USG investments in “Post Moore’s Law” era
At the same time that U.S. supercomputing experts and officials are calling for “an investment surge” and spelling out the consequences of status quo funding levels, the Trump administration has proposed cuts to the DOE that would roll back funding to 2008 levels, guided by a blueprint from right-wing think tank, the Heritage Foundation. New DOE head former-Texas Governor Rick Perry has professed support for DOE supercomputing but has not repudiated the Heritage Foundation plans.
Bottom line: “If these investments are not made, the U.S. can expect an HPC capability gap to emerge and widen in less than a decade,” the report asserts.
NSCI Objectives (source)
1) Accelerating delivery of a capable exascale computing system that integrates hardware and software capability to deliver approximately 100 times the performance of current 10 petaflop systems across a range of applications representing government needs.
2) Increasing coherence between the technology base used for modeling and simulation and that used for data analytic computing.
3) Establishing, over the next 15 years, a viable path forward for future HPC systems even after the limits of current semiconductor technology are reached (the “post- Moore’s Law era”).
4) Increasing the capacity and capability of an enduring national HPC ecosystem by employing a holistic approach that addresses relevant factors such as networking technology, workflow, downward scaling, foundational algorithms and software, accessibility, and workforce development.
5) Developing an enduring public-private collaboration to ensure that the benefits of the research and development advances are, to the greatest extent, shared between the United States Government and industrial and academic sectors.