What is a supercomputer?

  In November 2016, the “International Supercomputer TOP500” announced the latest world supercomputer rankings in Frankfurt, Germany. The “Shenwei·Light of Taihu” supercomputer independently developed by my country has achieved 93 quadrillion floating-point operations per second. Become the fastest supercomputer in the world.
  What is a supercomputer?
  As the name suggests, compared with ordinary small computers, supercomputers are machines with super-computing capabilities. Our personal computers seem to be powerful, but if they deal with huge amounts of data and specific computing purposes, their computing power will be far behind. How can we solve this dilemma?
  A very logical method is that if one machine does not work, then connect tens of thousands of machines, “how many machines are powerful”. So, in short, a supercomputer system is a cluster that organically superimposes the computing power of many computers to meet specific computing applications.
  So, where is the specific role of supercomputers?
  Don’t worry, the following three short stories will take you into the unknown world of supercomputers.
  Alan Mathison Turing (1912-1954), a British mathematician and logician, was known as the “father of computers.” The movie “Imitation Game” tells his story. The main line of the movie’s story is how to crack the Enigma code of the complex encryption machine used by the German army. The Enigma code automatically switches the password every night, with 1016 different changes. It is too late to be cracked before the next switch by manual calculations. This prevents the Allied forces from knowing the German bombing and submarine hunting plans in advance and suffers heavy losses.
  In order to crack the Enigma code of the German cryptographic system and end the war as soon as possible, Turing was ordered to work in distress. After analyzing a large number of German telegrams, he discovered that many telegrams have fairly fixed formats.
  With this as a breakthrough point, Turing thought of using the “candidate word” method to decipher Ennig’s messages, and proposed the theory that only machines can defeat machines. He led the MI6 scientists to secretly develop a code-breaking machine, and successfully cracked the German code two years later.
  Turing’s invention of the computer reversed the situation on the Atlantic battlefield and brought World War II to an end at least two years earlier.
  The reason for mentioning this story is that this is the earliest example of using a computer to break a code. How long will it take to brute force the Enigma code with today’s fastest supercomputer? Roughly based on the calculation speed of “Shenwei·Light of Taihu Lake”, it is about 0.11 seconds!
  In March 2016, during the man-machine battle, the world champion of Go, Lee Sedol, lost to “Alpha Dog”, and the loss was ugly.
  Before the start of the game, both the science and technology circles and the Go world were not optimistic about the “Alpha Dog” defeating Li Shishi. “Chess” Nie Weiping also asserted that “Alpha Dog” was not Li Shishi’s opponent at all.
  This battle slapped many people in the face, but it also made the trendy term “artificial intelligence” appear in news headlines every now and then. However, what you don’t know is that artificial intelligence, a fascinating and eye-catching red flower, is in vain without the green leaf of a supercomputer.
  Do you think it was just the little computer you saw on the live TV that you were fighting against Li Shishi? wrong! It’s a behemoth that you can’t see behind.
  The stand-alone version of this computer is equipped with 48 CPUs and 8 GPUs. It can be regarded as a small supercomputer, with floating-point computing power of 11 trillion operations per second. But in order to achieve better computing performance, “Alpha Dog” has also connected to 1920 networks composed of CPUs for cloud computing during training and battle. This remote computing power can also be transmitted to “Alpha Dog”, so that Its performance is 40 times higher than the stand-alone version, which has reached the threshold of challenging human masters.
  You may often hear the computer guy next door talk about the CPU’s computing speed or frequency. This combat value is called floating-point operations per second on the supercomputer, which is the speed capability of computing floating-point data.
  Take a Lenovo computer equipped with the i7-5500U fifth-generation Core processor as an example. Its main frequency is 2.4G Hz. According to the system analysis and evaluation tool Wheelstone benchmarks, its floating-point operation speed is roughly 14.2 billion operations per second.
  To make a rough comparison, the computing power of a “Shenwei·Taihu Light” is equivalent to 640,000 ordinary Lenovo computers. The workload for 1 minute of booting is equal to the workload of ordinary Lenovo computers for 444 days. The efficiency is high. It’s not tough.
  The artificial intelligence in the man-machine war story is composed of software and hardware. Algorithms are software, and supercomputers are irreplaceable hardware. In the race against top masters like Li Shishi, even if the software is beautifully written, if the supercomputing speed can’t keep up, Li Shishi will be fine. Your machine may not be finished!
  Therefore, computing speed is the lifeblood of supercomputers.
  ”Tiangong-1″ is China’s first target aircraft and space laboratory, as well as a passive target in space rendezvous and docking experiments. It was launched on September 29, 2011. It has been in space for more than 5 years and is expected to return to Earth in 2017.
  The problem is that during the process of returning to the earth, human aircraft will have a “black barrier” phenomenon. The violent friction with the atmosphere will cause it to rise to more than 1,000 degrees Celsius, and the communication signal will completely disappear. This is a very dangerous stage. Therefore, it is necessary to predict the detailed flight data and safety factor during the landing process in advance to be foolproof.
  But like this kind of high-end space flight, which is costly and requires various forces to complete, R&D personnel simply cannot do real experiments on the ground, and the results of traditional wind tunnel tests are not 100% absolutely reliable.
  At this time, supercomputers came in handy.
  The National Computational Fluid Dynamics Laboratory wittily borrowed the “Shenwei·Taihu Light” supercomputer from the National Supercomputing Wuxi Center to conduct computer experiments on the return process of the “Tiangong No. 1” to further strengthen the judgment of the reliability of the data.
  The R&D staff simplified the two cabins of the “Tiangong-1” aircraft (more than 10 meters in length, nearly 3.5 meters in cross-sectional diameter), falling flight (altitudes of 65 kilometers and 62 kilometers, speed of Mach 13), and the state of the bypass flow A massively parallel simulation was performed. The experiment used 16,384 processors, and completed the conventional 12-month computing tasks within 20 days. The efficiency was increased by 18 times. The calculation results are more consistent with the results of another wind tunnel experiment, which is “Tiangong No. 1”. The flight test provided important data support.
  For a long time, scientific research, including the aerospace field, has been the hardest hit area for complex computing. Due to the low computational efficiency of traditional computers, scientists have to abandon a lot of scientific research. The addition of supercomputers now seems to have taken off the curse on scientists, making the world of science refreshed.
  After telling the three stories, I believe you also have a better understanding of supercomputers.