China represents a big and growing market opportunity for IT vendors around the world. It’s huge population and market upside compared with the more mature regions across the globe is hugely attractive to system and component makers, and the Chinese government’s willingness to spend money to help build up the country’s compute capabilities only adds to the allure. In addition, it is home to such hyperscale players as Baidu, Alibaba and Tencent, which like US counterparts Google, Facebook and eBay are building out massive datacenters that are housing tens of thousands of servers.
However, those same Chinese government officials aren’t going to make it easy for US companies to bring their products into the country. China has been clear that they want system vendors in the country to buy more Chinese-built components, and over the past few years the government has invested billions in such areas as semiconductors – including spending $150 billion over 10 years to grow its chip-making capabilities – and exascale computing. For chip vendors like Intel and Nvidia, it became even more challenging when US lawmakers two years ago blocked them from selling products to four supercomputing centers in China for fear that the facilities are working on nuclear tests and that the efforts go against US national security interests.
All this has forced US chipmakers and others to become creative when addressing the Chinese market. One way has been to create joint ventures with Chinese organizations or partner with them to develop processors and other technologies. For example, AMD last year created a joint venture with China-based Tianjin Haiguang Advance Technology to develop x86 chips for servers in China. Also last year, Intel announced it was working with Tsinghua University and Chinese tech firm Montage Technology Global Holdings to develop a reconfigurable computing processor module and system software that will work with standard Intel Xeon server chips to offer capabilities that address requirements specific to the Chinese market in such areas as security.
Chinese chip designer Suzhou PowerCore has developed the CP1, a CPU based on IBM’s Power architecture through the OpenPower Foundation and aimed at Chinese systems, and a host of system vendors – including Zoom Netcom and ChuangHe – last year announced plans to roll out servers based on the new chip. A public-private partnership created by the Chinese government in 2014 called the China Power Technology Alliance was aimed at helping bring OpenPower technologies into local systems for projects in the country.
Chinese chip developers also are pushing products into the space. The Next Platform has detailed work that Chinese vendor Phytium Technology is developing ARM server chips dubbed “Earth” and “Mars” FT series.
For its part, Qualcomm officials in early 2016 announced a plan to create a joint venture with China’s Guizhou Province – a growing hotbed of technology development in the country – to build ARM-based server chips. Qualcomm President Derek Aberle late last month caused a stir when he briefly noted that the joint venture – called the Huaxintong Semiconductor Technology Co. – has begun creating a new CPU based on Qualcomm’s work with the low-power ARM architecture.
“In China, we also have made great progress with our data center joint venture, which is now busy developing a customized server CPU product based on our technology and designs for the China market,” Aberle said during a Jan. 25 quarterly earnings conference call.
It shouldn’t come as a surprise. As we’ve noted here at The Next Platform, Qualcomm has been among the most aggressive of the chip vendors that are pushing the ARM architecture – which is mostly found in mobile and embedded devices – into datacenter servers. The company late last year unveiled the Centriq 2400 line of SoCs with up to 48 cores and based on its “Amberwing” core. Qualcomm has made its billions pushing its low-power ARM chips into mobile and network devices, but it is aiming to challenge Intel’s dominant share of the server processor market. Given that Intel has well over 95 percent of the market, even a small dent could mean big money for Qualcomm.
Aberle’s statement regarding the customized ARM server chip being developed in the Chinese joint venture and the other efforts in China by other semiconductor vendors highlights the ongoing and growing competition between the US and China in the global datacenter space, particularly as we near the exascale era. China last year unveiled a supercomputer called the Sunway TaihuLight, a massive system developed by National Research Center of Parallel Computer Engineering and Technology (NRCPC) and installed at the National Supercomputing Center in China and armed with almost 10.5 million processing cores. The system runs at 93 petaflops but has a peak performance of more than 125.4 petaflops.
Powering the Sunway TaihuLight are more than 40,000 of Sunway’s ShenWei 26010 processors, developed in China. The supercomputer came fully on line last year, giving China the number-one and two spots on the Top500 list of the world’s fastest supercomputers. (Number two is Tianhe-2, housed at another National Supercomputer Center in China and running on Intel Xeon and Xeon Phi chips.)
Exascale computing is the next key goal, and is crucial for countries because of it could mean to them in terms of business, research and innovation. Chinese supercomputing officials upped the pressure last month when they said that a prototype exascale supercomputer that will be capable of running 1 quintillion calculations per second will be ready by the end of the year. The prototype initially was to be ready next year, but development is ahead of schedule. A fully enabled exascale system will be ready in 2020, the Chinese say.
The prototype won’t be running on the ShenWei processors, but apparently will get much of its compute power from a DSP coprocessor developed in China and called the Matrix2000 GPDSP that the folks at the Top500 organization say will deliver some 2.4 teraflops of double-precision performance for each chip and will be in the performance neighborhood of Intel’s Xeon Phi “Knights Landing.”
In the United States, the Department of Energy is operating the Exascale Compute Project, which is looking to develop an exascale system by 2023, though a new plan has developed recently to move that to 2021, backed by a push by the Obama Administration. However, the Trump Administration will have to support it, and there are worries that the new president and his team could slash federal funding and hamper exascale efforts.