Chinese Chip Closes In on Intel, AMD
China may finally have a processor to power a homegrown supercomputer.
At this year’s Hot Chips conference at Stanford University, Weiwu Hu, the lead architect of the “national processor” of China, revealed three new chip designs. One of them could enable China to build a homegrown supercomputer to rank in a prestigious list of the world’s fastest machines.
The Loongson processor family (known in China by the name Godson), is now in its sixth generation. The latest designs consist of the one-gigahertz, eight-core Godson 3B, the more powerful 16-core, Godson 3C (with a speed that is currently unknown), and the smaller, lower-power one-gigahertz Godson 2H, intended for netbooks and other mobile devices. The Godson 3B will be commercially available in 2011, as will the Godson 2H, but the Godson 3C won’t debut until 2012.
According to Tom Halfhill, industry analyst and editor of Microprocessor Report, the eight-core Godson 3B will still be significantly less powerful than Intel’s best chip, the six-core Xeon processor. It will be able to perform roughly 30 percent fewer mathematical calculations per second. Intel’s forthcoming Sandy Bridge processor and AMD’s Bulldozer processor will widen the gap between chips designed by American companies and the Godson 3B.
However, China’s chip-making capabilities are improving quickly. Intel’s Xeon processor uses a 32-nanometer process (meaning the smallest components can be formed on this scale), while the Godson 3B uses 65 nanometers, leading to significantly slower processing speeds. But the Godson 3C processor will leapfrog current technology by using a 28-nanometer process, although this will only increase its clock speed by about a factor of two, estimates Halfhill. With its eight additional cores, this should make the 3C about four times as fast as the Godson 3B.
Hu, lead architect of the Godson project, said via e-mail that China’s Dawning 6000 supercomputer, originally slated for completion in mid-2010, will instead debut in 2011, using the Godson 3B. Halfhill calculates that the Dawning supercomputer will use CPUs that are slower than fastest Intel chips. However, it could still rank on the Top 500 list of the 500 fastest supercomputers in the world–a significant coup for China’s fledgling electronics industry. “Just getting into the Top 500 with a native processor is a worthy accomplishment,” says Halfhill.
The Loongson processor is based on the MIPS instruction set, the basic commands that a microprocessor understands. In contrast, Intel and AMD processors are based on the x86 instruction set. Engineers at China’s Institute of Computing Technology (ICT) have added more than 300 instructions to the MIPS instruction set in the latest generation of the Loongson processor, and most are devoted to vector processing, a technique for processing data in parallel that can speed operations like graphics and scientific processing. The Dawning 6000 would mark the first time a MIPS-based supercomputer has appeared in the Top 500 list since 2004.
The ongoing development of the Loongson processor family is good news for Stanford-based MIPS Technologies, which licenses the MIPS instruction set and competes with the x86, ARM, and IBM Power architectures. “It’s our view that the ICT team and the MIPS instruction set are in a leading position for the [Chinese] government-driven national processor effort,” says Art Swift, vice president of marketing at MIPS.
At the low end of the Godson family of processors, the new 2H chip is an incremental improvement compared to previous chips in the Godson 2 series, says Halfhill. According to Hu, the chip is designed for netbooks, other mobile devices, low-powered PCs and embedded systems.
An important factor for the Godson 2 series has been the porting of Google’s Android operating system (used in smart phones, and in some tablets and netbooks) to the MIPS instruction set, says Swift, who adds that ICT engineers were very active in that effort. “The uptake of Android in China was phenomenal; they were way ahead of everyone else, and the whole rest of the field has followed,” Swift says.
Hu has emphasized in the past that a primary goal of ICT’s “national processor” effort is the creation of an affordable chip that can help bring China out of the industrial age and into the information age.
“I think what they’re really after is a national processor that is broadly used and displaces the Intel monopoly,” says Swift.
Displacing the Intel monopoly does not necessarily mean displacing the Windows monopoly, however. Despite ICT’s emphasis on Android and open-source software, the Loongson family includes many instructions designed to speed up emulation of the x86 instruction set, and the Microsoft architecture team attended Hu’s presentation at Hot Chips, according to Swift. “I wouldn’t rule out this being a great Windows processor at some point,” he says.
The Loongson family of processors may, however, face a fundamental challenge to its ability to compete with other architectures in terms of performance.
The Godson processor appears to have been designed primarily with automated circuit design tools, which is common throughout the microprocessor industry, but the processor has not been manually tweaked by engineers, which is not. This could mean unnecessary bottlenecks in the flow of data through the processor. “That’s always been a puzzle to me,” says Halfhill. “It’s not like there is a shortage of circuit designers in China.”
One of the most unexpected surprises of the Hot Chips presentation was the acknowledgement that if ICT’s current fabrication partner, STMicro, is unable to produce the Godson 3C in a 28-nanometer process by 2011, production could be moved to the Taiwan Semiconductor Manufacturing Company. Historically, China and Taiwan have had chilly political relations even as their economic interdependence has increased.
Government support for ICT’s national processor project was reaffirmed Monday when it was announced that the chip will be part of the country’s 12th Five-Year Plan. If the Godson 3B shows up in a supercomputer by 2011, it will be an important milestone in China’s billion-dollar effort to cultivate a homegrown CPU.
Become an MIT Technology Review Insider for in-depth analysis and unparalleled perspective.Subscribe today