The other issue, I guess, is that the west decided this was just a pissing context not worth playing. Heck, when I was in grad school (U of Utah) they would out the SGI supercomputer they had together to do world rating benchmarks and then promptly divide it afterwards into multiple computers because it wasn’t very useful in that mode. If you are simulating your nuclear explosions well enough already, the demand for ever faster super computers simply isn’t there when a cluster of GPUs will give you much better bang for the buck for newer tasks like machine learning.
Part of the reason is likely a lack of true demands, like the example you gave. BAT have shown that they don't lack talents to run computing infrastructure. However they probably don't have enough properly funded long term basic research projects that truly require HPC.
HPC != super computer, and hasn’t for 10+ years now. Just like mainframes (for most tasks) have been replaced by clusters, so have super computers (for most tasks).
This needs to come with an asterix: profiles of the rise of the Chinese tech industry or its supposed move from "imitation to innovation” often forgo how Chinese companies are entirely protected from foreign competition in their domestic market and that the Chinese government is forcing foreign companies wanting to do business there to share intellectual property with Chinese competitors.
It's a story about unsavory government behavior going unchecked not about private industry achievement.
Which are exceptionally narrow and target too narrow of technologies to make much difference from a GDP perspective.
I used to work CFIUS issues for the DoD and can tell you that the government preventing tech from being exported when both parties (developer and acquirer) want to share basically never happens.
Governments think of technology as products, but of course it is people that possess knowledge. Unless you cut off connections amongst people, eventually knowledge will diffuse.
That is like me saying something like this: they may be 1.4 billion strong, but most me are considered unhappy and sexually frustrated by policy design.
>China has many millions more men than women, a hangover of the country's one-child policy, which was overturned in 2015, though its effects will last decades more. The gender imbalance is making it hard for many men to find a partner – and the gap is likely to widen.
>By 2020, it’s estimated there will be 30 million more men than women looking for a partner. In his book, The Demographic Future, American political economist Nicholas Eberstadt cites projections that by 2030, more than a quarter of Chinese men in their 30s will not have married.
So 2.14% (30 million / 1.4 billion) of the males are potentially sexually(?) frustrated as monogamy is not an easy option for them, how is this relevant to the discussion exactly?
yet when the US practices protectionism it's the end of the world. Reality is that our 'allies' get rich by abusing trade agreements to act as middlemen for China
What's unsavory about it? Why should the Chinese care about American IP laws? Also I encourage you to look up how much the US has respected IP law when the shoe was on the other foot.
When Obama banned Intel from selling chips to China, it was assumed it would take years for them to catch up. They invested heavily in research and built their own to occupy spots #1 and #2 on TOP500. They also plan on building only ARM-based supercomputers in the future.
One thing important to Chinese is that the Sunway supercomputer uses entirely China-made CPUs. While these CPUs may never outcompete Intel or ARM in a commercial market, they could be used in computers of the military, nuclear stations and other critical infrastructure. China will then be the sole country free from the threat of US mandated backdoors in CPUs.
This might be a key factor in the drastic rise. China is annually graduating about 8 times as many STEM degree holders as the US, as of 2016:
"...In 2013, 40 percent of Chinese graduates finished a degree in STEM, over twice the share in American third level institutions."
"The World Economic Forum reported that China had 4.7 million recent STEM graduates in 2016. India, another academic powerhouse, had 2.6 million new STEM graduates last year while the U.S. had 568,000."
An honest question, are these degrees well regarded? In the main are the Chinese qualifications comparable to those from the US in terms of depth, breadth and rigor?
I'm only somewhat familiar with their primary and secondary education (based on observation, not experience). If that is any indication, then their university courses are likely harder or at least as hard at a comparable level of institution prestige. (Harder doesn't always mean better. It's a characteristics of education that may/may not help with real-world applications.)
One reason for that: It could be harder for a Chinese student to get into Tsinghua than a US student to get into MIT. The same likely holds for a range of other institutions as well. This is based on corresponding admission rates. Also, on average, Chinese students have better math background than American students.
Obviously, MIT, Stanford, etc. currently have higher prestige in the world's eyes. It doesn't mean though that the education there is always more rigorous than top-level institutions in China.
It really depends on your hukou given China’s residency system. Each province has a small quota, though some get much more than others, it isn’t has hard as a Beijinger to get into Qinghua as a person from Henan. Many richer Chinese choose abroad for better educational opportunities, even those that are obviously going to get into Qinghua or PKU (e.g. Beijing’s top gaokao scorer a few years back went to HKU).
On average, Chinese from the big cities have better math background than all Americans, not on average for the entire country. Compulsory education in china stops at 9th grade, and small city/rural schools are incredibly underfunded compared to PISA-record beating shanghai.
I'm working as a manager in Shenzhen. I have a guy on my team who's studying to do his IELTS so that he can get accepted into a foreign university to do his masters in computer science. He says that he can't get into Tsinghua (or any Chinese university at that tier) because the competition's too fierce, and all the universities a step below are not well enough respected. Hence studying for IELTS. He's targeting University of Toronto or something at that level. For what it's worth, it's only one anecdote from one Chinese guy.
The admission standards for master's degree for the same school is usually lower than that for an undergraduate degree. At least in US, Canada, Britain and China itself.
Would guess that it is at MIT/Stanford. A key thing to remember is that the top minds of the world aren't really emigrating to go to China to study or bring their family and that isn't going to change anytime soon. All roads still do not lead to China, and for good reason. Another important thing to take into consideration is having your trig identities memorized slightly better doesn't equivocate even remotely to higher intellectual output.
By harder, I meant problems more intellectually challenging to solve. Compare hard questions in Chinese Gaokao exam to the hardest questions in SAT Math:
By the way, SAT Math questions are generally considered a piece of cake by most good students in East Asia (and they often get 800 or close without much extra preparation), which says something about the level of average American high school math education.
The SATs are simple math. The GREs are as simple, which is presumably four years farther along than the SATs. However, it's hard answering all those questions correctly in the given time frame (presumably). You'd need a randomly sampled set of chinese kids taking the SATs to see where they fit against our distribution of scores to draw further conclusions. Bear in mind, too, top american students are studying advanced math in high school (advanced meaning, enough to reasonably answer the questions on the quora page, with time to practice and get used to things like the proof questions)
Note: I'm not trying to justify the american education system, just providing commentary and trying to level the comparison a bit.
Check out PISA 2015 which tests Chinese students from 4 provinces (200+ million population). They are on average above Massachusetts, the top scoring American state, in math.
Note that fewer percentage of Chinese youths attend high school than American, so that might skew the comparison.
I think I've heard of that. Went ahead and looked at the charts and it looks like the only mention of China were Taipei, Hong Kong, Macau, and Singapore. That's kinda biased. For example, we could equivalently grab all the students from stuyvesant high school or phillips exeter academy, make sure they know some set theory and basic calculus and what not, then make them take the exam.
Anyway, don't mean to argue - it does appear on the news a lot "the chinese are out mathing us! America sucks!" with some political finger pointing, and my reaction is usually a facepalm coupled with wondering how controlled the comparison is.
> the only mention of China were Taipei, Hong Kong, Macau, and Singapore
0/4, none of these are really under the aegis of the Beijing regime - and while it is flattering of you to compare the entirety of the population of four countries with 40 million people to the single most selective school in the US, it is rather inaccurate to do so.
If you look closer, PISA 2015 also had China in there, where the four provinces Jiangsu, Guangdong, Beijing, and Shanghai participated. Now poor illiterate migrants' kids might not get hukous to Beijing schools, so you might be right that this might be slightly biased - but even these four richer provinces have nearly the population of the US, so even if only 200 million Chinese are "outmathing us", it still makes a massive difference.
The key factor is that US cloud systems are not being included in supercomputing calculations at all. The most powerful supercomputers are Amazon's AWS or Google's systems. Alibaba's cloud systems are still tiny by comparison.
Cloud compute systems are the next evolution of supercomputing. Or put a better way, the traditional supercomputers being measured, are history.
It'd be like holding up 56k modem market leadership today as a sign of technological dominance.
Indeed much scientific computing these days is "high-throughput" rather than "high-performance" computing [1], and is done on commodity hardware, including commercial clouds. However there are certain classes of problems that still need highly-connected leadership systems [2,3], but that distinction is definitely blurring.
High performance computing (HPC) is a specific field. Fast communication between compute nodes is very important and you won't get that on GCP or AWS where you may get 10gbs at the max. Most of the fast simulation codes are bounded by communication time between nodes.
Although, Azure is currently doing some HPC offerings.
The point of the chart is that the rise in China's tech capabilities is quite broad-based. The tagline about supercomputers is just an illustration for non-tech readers.
Quantum communications and certain NLP applications/technologies are some areas that China has likely surpassed the US. These are definitely not comparable to 56k modem.
There's zero evidence China has surpassed the US in quantum. In fact all the evidence is that the US is leading quantum by a significant margin, from IBM to Microsoft to Google. Your NLP reference is meaninglessly vague.
Traditional supercomputers, except in isolated use cases, are a thing of the past. Nearly all supercomputing work will be done in the cloud in the not very distant future, supercomputing will continue to shrink.
China's use of supercomputing is because on the whole they're still a decade behind on adopting cloud computing (which is why Alibaba, their leader in the segment, has such a small business compared to the US cloud giants).
For real-world NLP: Xiaoyi (Little Doctor) was the first (and currently the only) AI that passed a medical licensing exam that human doctors need to pass (and rote learning is not enough to pass): http://sites.ieee.org/futuredirections/2017/12/02/congrats-x...
That probably includes CS degrees one can get at places like Hunan Tourism Univeristy (I would joke about Beijing Forestry University but their graduates are better than you’d expect). Finding good programming talent in China is still very hard, there are a lot of low quality degrees to sift out, so that everyone is chasing a very limited supply of new graduates from just a few well regarded schools.
Does anyone know more about the Chinese "Sunway" chips used in the TaihuLight? The wikipedia description sounds really interesting, sort of like the old Transputer.
It looks difficult to code for (although the paper mentions they have their own parallelizing compiler & have successfully compiled several real applications).
It is poorly understood that the strategy in Chinese side is set to position their industry orthogonal to global economy/markets.
They take upon something niche and obscure, but with great promise to give tons of offshoot tech and give breakthrough boosts to existing industry: renewable energy, batteries, rare earth metals, FPGAs, lights-off factories at scale... etc...
They play quiet for few years, before quickly locking down everything when there is zero change anybody catching up to Chinese companies.
Chicoms manage their economy as a single Japanese style conglomerate.
See how it went with renewable, China has a lockdown on the tech,
And today they circulate articles bragging with the number of supercomputers. It's the same as publicly listed companies increase headcount to look like they are thriving ("X is profitable and hiring"). Microsoft, for example, created a specialised AI department. Does it mean we'll have AGI in a couple of years?
In the age of cloud computing, does the number of supercomputers even indicate anything except the ability to run simulations on climate and nuclear fallout?
Personally, I wish someone created a more intelligent, less cronyist, less wasteful, but as willing to experiment alternative to the US tech industry. But it's definitely not China today that is even more reliant on meaningless KPIs than the corporate America. Of course, when the dear leader has an ambition to make the country a superpower, it makes sense to order more supercomputers.
Why are you baiting another pointless US vs China nationalism faceoff?
China has had a huge population and economy for decades. Almost by definition they will play an important role in the future of the internet. Anyone who said otherwise was wrong then and is wrong now.
> Why are you baiting another pointless US vs China nationalism faceoff?
That was not my intention. Instead, my implication was that the dismissal of China's progress that you see in HN was entirely motivated by nationalism, not objective analysis of the facts. If you were trying to objectively analyze of the facts the rational conclusion should be "this is really complex, who knows how it's going to turn out."
HN has a very strong USA (and in particular silicon valley) bias but I take offense at your generalization. I don't recall ever seeing a thread where the comments were unilateral china-bashing.
Things change. Back then it was largely true. It still is actually IMHO for the most part but we are approaching an inflection point in the next few years.
China spent almost all of the last 30 years copying developed nations to try to catch up. They're still aggressively doing it across every field, from reverse engineering Russian jet engines, to stealing IP & copying US fighter jets, to using CRISPR knowledge & technology developed in the West to try to leap frog, to the Landwind clone of the Land Rover. Their phones remain copies of what Apple and Samsung are doing; nearly all of their domestic phones run versions of Android, because they can't build their own OS yet. Copying permeates their entire economy.
That isn't in fact a criticism or a negative. It makes sense for any poor or undeveloped nation - which China absolutely was until recently - to do that to attempt to rapidly improve.
No no, the kind of talking point that I was sarcastically referring to was one that kept surfacing in reaction to news suggesting that China might surpass the USA in technology. Back then the HN reaction was "China just copies therefore they can't innovate". It's no the "they copy" bit that I'm talking about, it's what implication was being drawn from there.
That "therefore" is a negative and in my opinion more than anything shows that American has been leading the world for so long that they can't even conceive that might not be the case at some point in the future. Even people in tech who should be paying attention dismiss it by saying "nah, they can't really innovate, they're just copying, and therefore will never surpass us".
Even in this thread, you can see more of the same insecurities. I personally think the Kübler-Ross model of grief is a surprisingly accurate model of this phenomenon.
In general, I think we're still in the denial/anger phase, with a trend towards the latter. You can literally feel the seething anger just reading some of the comments here.
We used to say the same thing about Japan back in the 50s. And the British said the same thing about America in the 1850s. And it was totally true both times but successful imitation is hard and the skills to do that lead to the skills needed to innovate.
The measurement basis of TOP500 is effectively how good you are at solving a dense LU factorization of a matrix. This is a metric that is really easy to game, since the computation/communication ratio is very high (at least linear in problem size).
What I've heard is that most of the Chinese supercomputers do really well on the LINPACK benchmark and rather much more poorly on real workloads.
http://www.globaltimes.cn/content/1094092.shtml
The other issue, I guess, is that the west decided this was just a pissing context not worth playing. Heck, when I was in grad school (U of Utah) they would out the SGI supercomputer they had together to do world rating benchmarks and then promptly divide it afterwards into multiple computers because it wasn’t very useful in that mode. If you are simulating your nuclear explosions well enough already, the demand for ever faster super computers simply isn’t there when a cluster of GPUs will give you much better bang for the buck for newer tasks like machine learning.