They can deny it all they want
In the second week of April 2015, the U.S. Department of Commerce, the neo-economic arm of the DoD, DoE, DHS, NSA, and FBI, told Intel to take 50,000 of its Xeon processors out of the truck they were in and put them back in the warehouse—they weren’t going to get shipped to China after all.
Why?
Because the U.S. government has declared four technical centers in China associated with the massive computer known as Tianhe-2 to be acting contrary to U.S. national security or foreign-policy interests. Or stated another way, we’re not going to sell you chips to hack our agencies. To no one’s surprise, the developers of the Tianhe-2—or the Milky Way-2 in English—have said it is mostly used for scientific projects like genome research.
A few years back, China unveiled their Tianhe-1A supercomputer with 7,168 Nvidia Tesla M2050 GPUs and 14,336 Intel Xeon CPUs.
The Tianhe-1A finally got the attention of the U.S. government, and in November last year, the U.S. Department of Energy (DoE) awarded IBM and Nvidia $325 million to build the world’s fastest supercomputers by 2017, named Sierra and Summit. The two supercomputers are supposed to deliver more than three times the performance of the machines currently in operation, which include the behemoth Chinese Tianhe-2.
In between, the DoE ordered the Aurora and Theta machines from …. Intel and Cray—imagine that. So what the government taketh away it also giveth. Intel will do the chips; Cray will do the boxes.
It’s not been publicly stated how many Intel Xeon processors will go into Aurora, but 50,000 to 80,000 has been mentioned in hallways.
Now my friends at Lawrence Livermore Labs are adamant that the two events (China ban, Aurora award) are not related and that the Aurora project has been in the works for some time. Yeah, sure, I believe you, wink wink.
And Horst Simon, a supercomputer expert and deputy director of Lawrence Berkeley National Laboratory, said the U.S. restrictions in the long run will help Chinese chip makers and hurt U.S. companies.
China significantly lags behind the U.S. in chip design, though the Chinese government has been bankrolling research to improve the capabilities of local chip makers.
“The Chinese will be more incentivized to develop their own technology, and U.S. manufacturers will be seen as less reliable and potentially not able to satisfy foreign orders,” Simon said.
What do we think?
China, as you may know, has a special no-royalty deal for the MIPS architecture. There has been an on-again/off-again program in labs in China to leapfrog the U.S. and design a superprocessor based on MIPS. However, China doesn’t have, nor will it anytime soon, the fabs needed to build such a processor if it were designed. Also, regardless of the U.S. actions, China wants to have its own processor and be independent of the U.S., and if possible replace the U.S. as the number one processor supplier. As consultant Mark R. Anderson of the SNS Global Report on Technology and the Economy points out, China is not just a country; China is a country with a business plan. The U.S. doesn’t get that (even though Anderson is a major consultant to the government).
So will denying China a few thousand processors accelerate their plans? No. You can’t make a CPU design go faster just because you’ve been pissed off by a rival government, and you sure as hell can’t build a 10-nm fab even if you do have most of the dollars in the world. Unless, of course, you believe in conspiracies and think the Chinese already have all the technology from their hacking efforts, and are bringing home all the U.S.-trained engineers to build new fabs and design new chips.