At the 2010 International Supercomputing Conference researchers announced the annual list of the world’s most powerful supercomputers, noting that China now has the world’s second most powerful
supercomputer, and is well positioned to overtake the United States for the number one slot.
But more important than whether or not a particular country has the world’s “biggest iron” is whether or not scientists have access to and support in using the country’s supercomputing infrastructure. In other words, while the bragging rights that come with owning the world’s fastest supercomputer are substantial, the total (and per capita) access to high performance computing resources is a better measure of how prepared a country is to leverage this kind of computing power for innovation and national defense, argues John West of insideHPC.
To get an idea where the world’s countries stand under such a measure, here’s a visual representation of the total peak theoretical performance (Rpeak) of all of the supercomputers possessed by a country, as of 2010, via this nifty Many Eyes visualization of the list of the world’s top supercomputers. Compare it to 2009, below.
2010 Countries By Total Peak Supercomputing Processor Power
2009 Countries By Total Peak Supercomputing Processor Power
In a single year China (light green circle in 2010, dark green circle in 2009) nearly tripled its total installed supercomputing capacity, moving ahead of Germany and pulling away from Japan, France and the UK. Meanwhile, the U.S., which already has an enormous capacity for high performance computing, in absolute numbers only added about half the number of floating point operations per second (Flop/s) to its installed base as China.
The explosive growth of China’s supercomputing capacity, witnessed nowhere else in the Top500 list of the world’s most powerful supercomputers, demonstrates the country’s commitment to high performance computing, and no wonder: HPC is used for everything from basic research and weather prediction to design and (simulated) testing of nuclear weapons.
Interestingly, if we calculate the ratio of Flop/s (performance) to the total number of processors (chips) in all the supercomputers in both China and the U.S., we find that China is getting more than twice the performance per processor as the U.S. That’s because China’s capacity is the product of a recent build-out, whereas the U.S. has been dominant in supercomputing for some time, and has many more aging machines.
So, on the one hand, as relatively late adopters, the Chinese have the advantage of being able to pay less for more overall performance, whereas in the U.S. there is considerably more experience with supercomputing in general.
There is every indication that the trends evidenced in recent data on supercomputing performance by country will continue–and if the Chinese do take first place in terms of performance of a single computer, it will be a triumph not only in terms of raw engineering horsepower, but also in terms of breaking free of the hegemony of U.S.-based semiconductor firms. It’s rumored that the machine that will break the record will be based on chips of native Chinese design, as previously reported by Technology Review.
When designing an embedded system choosing which tools to use often comes down to building a custom solution or buying off-the-shelf tools.