It's 2022, and the E-class supercomputing that was cheered on the Internet in previous years has not been manufactured yet?
The so-called E-level supercomputing is a supercomputer that can perform tens of billions of mathematical operations per second (1EFLOPS), which is a milestone in the development of human supercomputing, the next "crown", and is expected to play a huge role in solving major problems such as the global energy crisis, climate change, and environmental pollution.
Supercalculation test
At present, the world's fastest supercomputer is Japan's Fugaku, whose peak computing speed reaches 537212.00Tflops, that is, 0.53Eflops, which has been counted as a foot into the threshold of E-level supercomputing.
It needs to be explained that the internationally recognized E-level supercomputing refers to the computing power of double precision floating point (FP64), if it is for low-precision operation speed, the supercomputer with tens of billions of operation speeds has long come out. Tesla's supercomputer Dojo, released in 2021, has a computing power of 1.8 Eflops, but it is based on the low accuracy of FP16, and If Fugaku in Japan uses FP16 tests, the speed can reach more than 2Eflops.
Frontier supercomputer
The real E-class computer has not really been manufactured at present, but the United States, Japan, Europe and other E-class computers are under development. Among them, the Frontier supercomputing in the United States has made the fastest progress, and the latest news says that it has been completely built and almost put into use publicly.
The mainland is not lagging behind in E-level supercomputing, Shenwei E-class, Shuguang E-class and Tianhe-3 have long been manufactured prototypes, and these 3 E-level supercomputers are expected to be put into use this year.
Tianhe-3 E-class supercomputer prototype
A few years ago, it began to be hyped, and the E-class computer that will come out in 2020 has not come out yet? Personally, I think there are several reasons:
First, the demand side
Demand is the main driving force to promote the development of science and technology, because supercomputing is almost applied in all cutting-edge scientific and technological research, large-scale engineering, mutual (material) networking and other fields, its demand has been very strong, which is also a reason for the rapid development of supercomputing.
Tianhe-2 supercomputer
Now the demand for supercomputing is just as strong, but there are also some different angles of thinking, that supercomputing should not be the pursuit of peak speed mode. Jack Dongla, founder of the Top500 list and supercomputing expert, said: "Speed does not mean everything, although the car reaches a speed of 300 km / h, but the real value of most cars is to drive safely and reliably within the speed limit." ”
Therefore, the new generation of supercomputers, they will consider more factors, including better economy, reliability, environmental protection and so on.
Second, the problem of energy consumption
In addition to the fast computing speed of the E-class computer, it also has strict requirements for energy consumption. If only the processor is continuously added, then the volume and power consumption will continue to increase, which is not a sustainable way out of supercomputing, and the current industry-recognized E-level supercomputing energy consumption standard is below 40MW to reach 10 billion times per second.
Inside the computer
The new generation of supercomputers in various countries are very concerned about energy consumption, the United States "Aurora" E-class supercomputing design power of 40MW, the first E-class supercomputing in the mainland is reported to not exceed 30MW, and Europe is committed to its first set of E-level supercomputing energy consumption control within 10MW, these goals will make the development of E-class supercomputing need to take more time.
Third, there is a shortage of chips
Although the chip shortage that has emerged since 2020 is mainly in the fields of automobiles and smart phones, the price of these chips continues to rise, resulting in a "siphon effect". Chip manufacturers focus their production capacity on the production of consumer-grade chips, which also affects the production speed of supercomputing chips to a certain extent.
The difficulties faced by the mainland E-class computer are even more numerous, and in order to block the development of continental supercomputing, the United States has exhausted all kinds of rogue means. For example, many of the companies listed on the so-called "entity list" by the US Department of Commerce last year are in the supercomputer industry chain.
U.S. Department of Commerce's Entity List
4. Other reasons
The global top 500 computers are released twice a year, the distribution is in June and November, if not launched in these two months, the world supercomputing ranking list has not been updated, generally not too many people to pay attention.
Supercalculate rankings
Finally, there is another reason, that is, the self-media is chaotic. Many self-media are catching wind and shadows, seeing a little vague information, they began to open the imagination and add oil and vinegar mode, and then the whole network followed the hype, just like a hen has not yet laid eggs, but in their pen is already a group of "Oh Oh" big chickens.
However, from the perspective of the speed of development and the latest situation, the probability of a real E-class computer appearing in 2022 is very high, but who will be the first to release it? China, the United States or Japan? Let's wait and see!