GPU is the image processor. A simple way to understand the difference between GPU and CPU is to compare how they handle tasks. CPU is composed of several cores optimized for sequential serial processing, while GPU has a large parallel computing architecture composed of thousands of smaller, more efficient cores (designed for simultaneous processing of multiple tasks). The GPU provided by NVIDIA provides 95% computing power for the "vertex".
Since China's supercomputer became the strongest supercomputer for the first time in June 2013, the United States no longer has the world's most powerful supercomputing. The supercomputer official ranking of the global supercomputing TOP500 organization will be updated at the end of this month, and the vertex is expected to end this situation.
Although supercomputers are slightly fading in the age of cloud computing and big data centers, many of the thorny computing problems still require large machines. A US government report said last year that the United States should invest more in supercomputing in order to catch up with China in defense projects such as nuclear weapons and hypersonic aircraft, aerospace industry, oil exploitation and pharmaceutical industry.
Us Energy Secretary Rick Perry Twitter screenshots
Rick Perry, the US energy secretary, believes that the "superposition" will have far-reaching implications for energy research, scientific discoveries, economic competitiveness and national security.
The Oak Ridge National Laboratory, founded in 1943, was originally built as part of the Manhattan plan for the production and separation of uranium and plutonium, formerly known as the Clinton laboratory. Their mission is to overcome the most serious scientific problems facing the United States.DevelopmentNew technology will create a better life for mankind and protect mankind.
27648 Nvidia GPU 9,000 IBM traditional processors
According to the data released by the Oak Ridge National Laboratory, the "vertex" covers the size of two tennis courts, and the circulation system needs to consume 4000 gallons (about 15142 litres) per minute to cool the 37000 processors. The results of the standard measurement of supercomputers show that the floating point operation of vertex can reach 20 billion billion times per second. This is 100 times faster than the supercomputer "Titan" that the United States made.
IBM CEO, U.S. Energy Secretary Rick Perry, Nvidia CEO
The "vertex" strength of the computer is 95% from the GPU, which is provided by the NVIDIA, that is, the Volta graphics processor that carries a 27648 block of Tensor Core, allowing the "vertex" to carry out a hybrid precision calculation of over 30 billion times per second (3.3 exaops) per second.
Eliu Huerta, a researcher at the University of Illinois, Urbana-Champaign and the American National Supercomputing application center, describes the "vertex" giant GPU pool as "like a dream."
GPU, the image processor, is different from the general computer's central processor (CPU): CPU is composed of several cores optimized for sequential processing, and GPU has a large parallel computing consisting of thousands of smaller, more efficient cores (designed for multiple tasks at the same time). Framework。
Several years ago, researchers found that the parallel operation capability of GPUs is very suitable for running deep learning algorithms. In the wave of AI, GPU launched the market quickly, and even affected Intel's position in the computer chip market.
Thomas Zacharia, head of the Oak Ridge National Laboratory, says that such a large scale of GPU use is not common for supercomputers, and it helps machine learning to make breakthroughs in solving scientific problems.
In addition to the help of GPU, IBM provides more than 9000 traditional processors.
"Apex" becomes the "playground" of cutting-edge scientists.
The supercomputing power of "vertex" also makes scientists believe that it can be a miracle in the future.
"Vertex" supercomputer appearance
The Oak Ridge National Laboratory has become a "playground" of sophisticated scientists, and many university scientists are eager to use the computer to complete the previously impossible tasks.
Huerta said he hoped that the "apex" could help analyze the 15TB images obtained from the "large integrated telescope" every night.
According to the introduction of Oak Ridge National Laboratory, the "trip" arrangement under the "vertex" is very intensive.
One is cancer research. The US Department of energy and the US National Cancer Research Project are working on a distributed learning environment for cancer. Their goal is to develop research tools that can automatically extract, analyze and classify health data to reveal the relationship between hidden disease factors, such as genes, biomarkers and the environment.
The two is the study of fusion energy. Fusion energy has long been a representative of clean energy and abundant energy. Scientists have been hoping to simulate fusion reactors and their magnetic confinement plasmas and speed up business development.
The three is the study of disease and addiction. In this study, researchers will use artificial intelligence to identify functions and evolutionary patterns of human proteins and cell systems. These patterns can help us better understand Alzheimer's disease, heart disease or addiction, and inform the drug discovery process.
In addition to the above items, the "vertex" can also provide more help in chemical and biological problems. Zacharia believes that the supercomputer can use medical records from 22 million veterans to contribute to the US Department of energy projects, including about 25 million genome sequences.
The "vertex" provides an unprecedented opportunity for the integration of artificial intelligence (AI) and scientific discovery, enabling researchers to apply the technology of machine learning and deep learning to the research of human health, high energy physics, material exploration and other fields.
5 facts about the "vertex" supercomputer
1.20 billion floating-point operations: this means that if each person on the earth does a single second calculation, it takes 1 years to reach a "vertex" of one second.
2.30 billion times mixed precision calculation: this means that if each person on the earth does a single second calculation, it takes 15 years to reach a "vertex" of one second.
3. in an early test, a gene research group took a "vertex" supercomputer to take 1 hours to solve the problem, but it could take 30 years if the problem was put on a traditional computer.
4. the "vertex" area is equivalent to the size of 2 tennis courts.
5. the weight of the top and bottom is equivalent to the weight of a commercial jet.