China to Finish 'Indigenous' Supercomputer Soon, Experts Say
This is the view of some research and industry experts in the U.S., but most notably Steven Koonin, the undersecretary for science at the U.S. Department of Energy (DOE), who says China is now working on petaflop-class supercomputer "using entirely indigenous components that is expected to be complete within the next 12 to 18 months."
"It is clear that DOE will not be the only organization working to push the limits of computer performance," wrote Koonin, in a post published on DOE's Energy Blog on Friday.
Explainging how the 12-to-18 month estimate was made, an advisor in Koonin's office told Computerworld it was a collective assessment based on data coming from China and Chinese researchers and visits to China by several people.
Koonin isn't the only person to think such a time frame is possible. Jack Dongarra, a professor of computer science at University of Tennessee and a distinguished research staff member at Oak Ridge National Laboratory, makes a similar prediction, and cites China's work on microprocessors, which include chips based on MIPS architecture, and the Loongson or Godson processor.
China's supercomputing goals are becoming plain. Last month, China announced that its Tianhe-1A supercomputer had achieved 2.5 petaflops in a standard benchmark test, making it the world's most powerful supercomputer.
This standing is expected to become official when the Top500 supercomputing list is released in the middle of this month. (A petaflop is one thousand trillion sustained floating-point operations per second.)
Today, U.S. chip makers, in particular Intel , dominate supercomputing, and even the Tianhe-1A relies Intel Xenon processors and Nividia Tesla GPUs. Intel chips are used in 406 of the top 500 systems, followed by AMD at 49, and IBM Power at 42, according to Top500 list data.
However, China's Tianhe-1A shows how the country is gradually making its break from Western technology. Along with the Intel and Nividia chip, this system includes a third chip, the FeiTeng 1000, an eight core Chinese produced Sparc-based processor that is used to operate service nodes, such as log-ins.
China's latest supercomputer includes its own proprietary interconnect that is faster than InfiniBand but not as fast as the proprietary interconnects from U.S. companies, said Steve Conway, an HPC analyst for IDC. "[But] the most impressive thing about this is they developed most of the software themselves," Conway said.
China has also developed its own Linux operating system, called Kylin.
It's conceivable that in 12 to 18 months the Chinese could build a supercomputer based on all Chinese processors, but it would be for this specific supercomputer and not for the commercial market, Conway said.
"I would have a hard time believing that those would be processors that they could sell into the global marketplace," said Conway, who said competing globally may take China five to 10 years.
But China is clearly motivated to develop its own technology, Conway said. "For historical reasons China was unable for decades to buy the most powerful class of supercomputers from the West, and they were in a state of dependency, and presumably they want to escape that dependency," he said.
Nathan Brookwood, a chip analyst at Insight64, said doesn't doubt that China can design a homegrown system, but added: "I'm not sure where they can manufacture such a chip. Intel just opened the most advanced fab in China, but it operates at 65nm, two generations behind the facilities it uses in the rest of the world."
China could use chip makers outside of China, such as the Taiwan Semiconductor Manufacturing Co., and other makers as well, Brookwood said.
The challenge by China to produce supercomputers for scientific research and industrial development may be coming at an interesting time. This week's election is expected to bring a new push to cut federal spending.
The U.S. is the major source of funding of large systems that compete on the scale of those being built in China, Europe and Japan.
The U.S. has allocated just under $2 billion this year for high-performance computing in unclassified areas. Funding for classified systems is not disclosed.
"With the change in Congress there will be heavy pressure to cut discretionary spending, so yes, researchers in all disciplines should be concerned," said Peter Harsha, director of government affairs for the Computing Research Association.
But science funding does find some support from elected officials on both sides of the aisle, "who recognize that the country's ability to stay a competitive is tightly tied to its ability to remain a home for innovation, and that fundamental research is a huge driver of that innovation," Harsha said.
"Tianhe-1 is a great example of how capable and committed to research our economic competition has become," he said.
Patrick Thibodeau covers SaaS and enterprise applications, outsourcing, government IT policies, data centers and IT workforce issues for Computerworld. Follow Patrick on Twitter at @DCgov or subscribe to Patrick's RSS feed . His e-mail address is email@example.com.
Read more about mainframes and supercomputers in Computerworld's Mainframes and Supercomputers Topic Center.