The Inquirer-Home

China's supercomputers are super, just not all of them

Column Showing the right face to the world does matter, too
Fri Mar 30 2012, 15:38

novatheinqnewWHEN YOU'RE IN THE MOOD for a turbocharged time, spend a week in the gigantic metropolis of China's city of Shenzhen.

With close to 15 million souls - no one knows the exact number, but it's surely more than the official statistics - Shenzhen is just a hop from Hong Kong with its seven million inhabitants and is joined at the hip with Dongguan's six million people, with another 16 million in the Guangzhou-Foshan conurbation.

In 100 miles and close to 50 million people - 60 million if you count other nearly merged cities - of a building after building urban area, you will find the world's highest concentration of electronics firms.

You'll see shopping malls the size of the whole of Oxford Street with floor after floor of all possible electronics and computer components and assemblies - a hardware nerds' heaven of a sort - yet this is a clean, well developed city too, on a par with most Western metropolises, except for its lack of good Western and organic food.

It's quite a coincidence that while I was there a fancy story appeared in the Wall Street Journal of 23 March, intriguingly titled 'China's Not-So-Supercomputers' and focused on Shenzhen and its supercomputer centre. As you can see here, the story's message is that China's supercomputing is still quite immature, focused on mundane problems related to local politicians' wishes and based on machines that mostly rely on US technology - in short, not any threat to the good old US.

Since a certain individual named Wendy, or Win Di in Chinese, Deng is the 'lady boss' of the Murdoch empire controlling the WSJ, I guess she didn't have anything against this portrait of China either.

I took a bus number 36 on a lovely windy Friday afternoon to the Shenzhen Supercomputer Centre and went directly to talk with the marketing lady there. I went inside thanks to a kind researcher, as the marketing lady in question didn't want to meet, uttering in a loud voice that she was very busy, yet asking me before that, what did I think about the WSJ story?

But okay, that's life. Maybe the (anti) marketing lady scared off the poor American journalist too, and that's why that story ended up the way it is. So, the 'face to the world' was what the WSJ chose as the first problem.

Let's rely on the researchers there, and some are less shy than the others. I took a look at the Shenzhen Supercomputer Centre from the inside and had a quick chat with two researchers about the goings-on. That's where the second problem, troubles with software optimisations and such, came up.

Yes, China has less experience than the USA, Europe or Japan in coding software for large machines like this, but they are learning fast. So, they disagree with the WSJ's suggestion that software 'maturity' will remain a problem, since they are gaining ground there by leaps and bounds.

Coming to the third problem, I was privately told that the largest issue is actually the 'useless' architecture of the Nebulae machine, which owes its petaflop status to a bunch of Nvidia Tesla GPGPUs stacked on top of regular Xeon 5650 Westmere EP six-core processors.

The Nebulae machine that powers this supercomputer centre is actually the wrong example for assessing the status of Chinese supercomputer technology. It's basically a run-of-the-mill generic server cluster of lots of Intel machines made by Dawning, which is now Shuguang, using Nvidia Tesla GPUs and Infiniband interconnect.

There's nothing Chinese or generally unique in there, and it's something any system integrator or university can pull off with a bit of money and suitable space, power and cooling. And, as anyone in the field will tell you, including me, an HPC system insider, programming those GPUs to actually extract useful performance in more than a couple of codes can be a challenge. Even the Linpack benchmark didn't exactly shine, with only a 1.2PFLOPs Rmax actual rate obtained from a nearly 3PFLOPs Rpeak theoretical rate, and this is one routine that all supercomputer centres in the world optimise for in their bid to get a space on the TOP500 list.

Compare it to the current Chinese petaflop record holder, the Tianhe system in Tianjin Supercomputer Centre next to Beijing, which uses its own interconnect at twice Infiniband speeds with SPARC-like domestic I/O processors driving the connections. Yes, it also has a bunch of GPUs, but its interconnects and system design are world-class, and fully Chinese.

Then, of course, the Jinan petaflop supercomputer machine, also in the north, is based on thousands of 'Shenwei' 16-core Chinese SIMD-enhanced Alpha CPUs, coupled with eight-channel DDR3 memory controllers and RAM dies soldered on board, plus ultraslim liquid cooled blades for a supercompact layout. Oh yes, each of these Alphas delivers peak floating point throughput similar to that of a Xeon E5 at less than half the power consumption. Now, who wouldn't want to see Alpha again? Well, it was resurrected, but in China, like it or not.

I've seen the first system, and saw details on the second one, and China's mastery of hardware here is top-notch, just like their 500km/h capable maglev trains in Shanghai. If you want to know more about all these systems including the specifications, it's all at, so there's no point in repeating it here.

By the end of 2014, China will have at least three 100PFLOP machines that I know of installed - one each in the brand new supercomputer centres of Guangzhou, Changsha and Chongqing. The latter one, in the new industrial powerhouse city of 34 million in western China, where one fifth of all world's motorbikes and soon similar proportion of all laptops are made, will be fully based on Chinese 'Loongson' MIPS processors with massive SIMD extensions. That doesn't even cover what the northern provinces will have.

When creating a machine of this scale - remember it's over half a million high end Xeon chips, not cores, per machine - excellence in system design, interconnects, power, cooling and even software stack including management, scheduling and load balancing, has to be there. So, the Chinese will have it. And, the fight for the speed crown will also grow within each city. I also had a chance to talk to the competing local HPC centres in Shenzhen. Watch out for some exclusive takes on intra-Chinese supercomputer competitions too.

In summary, the WSJ article isn't right. China is rapidly maturing its supercomputing technology, especially from the hardware point of view, with its own Alpha, MIPS and SPARC processors, plus a dozen ARM licensees and a couple of its own advanced instruction set architectures out there as well, plus its own interconnects and I/O controllers. The Middle Kingdom can control its complete 'vertical stack' of hardware and avoid security risks or technological dependence on a potentially hostile foreign power.

Also, there's nothing wrong in having 'mundane, every day' uses of supercomputers. For instance, a detailed traffic simulation of a gigantic city like that, where every person, vehicle and road, and their behavioural patterns, are fully simulated, can help save billions in correct positioning or construction of new highways or rail lines. Do all the supercomputers have to simulate nuclear bombs? I guess no, so in that sense the civilian use benefit of Chinese supercomputers is just fine.

As for the example of the otherwise beautiful looking Shenzhen facility, as you can see in the WSJ picture, the choice of system architecture in its rather bland main machine and the choice of its 'window to the world' from the marketing point of view, are their own problems to solve. In that sense, the WSJ story is right in its quest for more maturity here.

We'll have more on China high technology adventures for you soon. µ


Share this:

blog comments powered by Disqus
Subscribe to INQ newsletters

Sign up for INQbot – a weekly roundup of the best from the INQ

INQ Poll

Happy new year!

What tech are you most looking forward to in 2015