While much attention has focused on China’s rising supercomputing prospects lately, a number of other countries, including Russia, are also quickly muscling up their HPC resources. With their eyes set on exascale, Russia is planning to invest over a billion dollars this decade to field at least one such system by 2020.
This week at the High Performance Computer and Communications Council (HPCC) Conference in Newport Rhode Island, T-Platforms’ chief products and technology officer Alexey Komkov, described the effort as well as some of latest supercomputing happenings in Russia.
Although the country is currently in 9th place on the TOP500 list in number of top supercomputers, it already claims a peak petaflop machine, the 18th-ranked Lomonosov at Moscow State University. Lomonosov is a T-Platforms system, but of the remaining four Russian supers that placed in the TOP500, three are from HP, with the remaining one from Russian HPC developer RSC SKIF. In fact, of the top 50 supercomputing systems in Russia, 6 were built by T-Platforms, 17 from IBM, 16 were from HP, and 11 from a variety of other vendors.
The mix of vendors helped support a growing HPC market in Russia, which has swelled from less than $20 million in 2005 to over $140 million in 2011. Although the rate of growth has slowed over the past couple of years, it’s still outpacing HPC growth on a worldwide basis.
Like China though, Russia is working toward developing more indigenous HPC capability. The program was begun in 2010, with an initial $37 million invested toward HPC research and application development. Since then it has expanded into a full-blown national exascale effort, with the government budgeting $1.5 billion toward the effort over the next several years.
The program encompasses everything from education and research to middleware and applications to interconnects and high-performance microprocessors. The latter is important to Russia because technology restrictions are still in place that prevents the country from importing the latest CPUs from the US or the European Union if they are destined for machines in the defense industry or for nuclear labs. Curiously, according to Komkov, there are no such restrictions for GPUs.
While these restrictions do not apply to systems in other areas, like academia, oil & gas, and bioinformatics, the country’s defense establishment is important enough that some of the $1.5 billion planned for exascale development will go toward a new national microelectronics center. The center, which is slated to open in the second half of 2012, will get an initial investment of $70 million, with the intent to secure another $180 million or so in the near future. It’s purpose is to develop Russian ASICs, some of which will be designed for high performance computing. That certainly mirrors what China is doing with its indigenous microprocessor initiatives, such as the Godson CPU program, although that effort is much further along.
Given that the Russian microprocessor program is in its infancy, the chips aren’t expected to be in production anytime soon. Komkov said Russia’s first 10 to15 petaflops system, which is targeted for deployment in 2014 or 2015 will likely stick with Intel CPUs and NVIDIA GPUs. But the first 100 petaflops machine in 2017 or 2018 could very well be equipped with native microprocessors, and the exaflops machine of 2020 is even more likely to be powered by Russian-designed chips.
For the exascale program as whole, the effort is being spread across non-profit and commercial organizations including the Russian Academy of Sciences, T-Platforms, and RosAtom, the state controlled Russian nuclear regulatory body. Exascale application areas are more narrowly focused than they are in the US or Europe, and are centered on strategically important industries in Russia, specifically the defense and energy sectors, but will also leak over into areas such as telecommunications, aerospace, and scientific research.
Even though a 2020 exascale deployment in Russia is not assured, at least the effort appears to be better focused than what is going on in the US at the moment. DARPA, the NSF, and the DOE do have exascale initiatives in play, but there is no agency to coordinate the disparate efforts. And according to Indiana University’s Thomas Sterling, DARPA’s funding for the principle program that was supposed to deliver exascale technology, UHPC (Ubiquitous High Performance Computing), will come to an end this summer.
UHPC, which was originally designed to be four-phase, seven-year program, was funded for just a couple of years before DARPA pulled the plug. There is a another DARPA program in the works, called Power Efficiency Revolution For Embedded Computing Technologies (PERFECT), which could help to facilitate some UHPC goals, but that initiative is focused on embedded systems.
Sterling, who delivered an exascale roadmap presentation at HPCC on Monday, believes the US effort is now probably behind schedule, especially in critical areas like the execution and programming models. Those software components, along with the new computing architecture that will be needed, will take several years to develop, and as of now, there is no consensus on any of these elements
As a result, Sterling thinks the 2020 timeframe for the deployment of a US exaflops supercomputer may now be optimistic. “The clock is already ticking,” he said. “Perhaps it’s already too late.”