The U.S. regains its first supercomputer: Nvidia GPU provides 95% computing power

Five years later, the United States temporarily regained the top spot in the world supercomputer field.

On June 8, US local time, the Oak Ridge National Laboratory, subordinate to the US Department of Energy, announced that they have developed a supercomputer 'Summit' with a peak floating point speed of 2 billion billions per second, which is close to 'Shen Wei'. The light of Taihu Lake is twice that of the supercomputer. It is reported that this supercomputer is manufactured by IBM Corporation and is equipped with nearly 28,000 NVIDIA GPUs and over 9000 supercomputers with IBM traditional processors.

GPUs are image processors. A simple way to understand the difference between GPUs and CPUs is to compare how they handle tasks. The CPU consists of several cores optimized for sequential serial processing, and GPUs have a A massively parallel computing architecture consisting of a smaller, more efficient core (designed for simultaneous multitasking). This time, the GPU provided by Nvidia offers 95% computing power for 'vertex'.

Since China's supercomputer became the strongest supercomputer for the first time in June 2013, the United States no longer has the most powerful supercomputer in the world. The official ranking of supercomputers of the global supercomputer TOP500 will be updated at the end of this month. 'Vertex' is expected to be End this situation.

Although supercomputers are eclipsed by the era of cloud computing and big data centers, many thorny computing problems still require large machines. Last year, a report from the US government stated that the United States should invest more in supercomputing in order to achieve nuclear weapons and superiority. The soyspeed airplanes and other defense projects, aerospace industry, oil exploration and pharmaceutical industry are catching up with China in terms of commercial innovation.

The United States Energy Minister Rick Perry believes that 'very' supercomputing will have a profound impact on energy research, scientific discovery, economic competitiveness, and national security.

The Oak Ridge National Laboratory was established in 1943. It was originally built as part of the Manhattan Project of the United States. It was built with the main purpose of producing and separating uranium and plutonium. It was formerly known as the Clinton Laboratory. Their mission was to conquer the United States. The most severe scientific problems facing the world, and the development of new technologies to create a better life for humanity and protect humanity.

27648 NVIDIA GPU+9000 IBM Legacy Processors

According to data published by the Oak Ridge National Laboratory, the 'apex' footprint is equivalent to two tennis courts, and its circulatory system consumes 4,000 gallons (approximately 15142 liters) of water per minute to cool down 37,000 processors. The supercomputer's standard metrics show that floating-point operations for 'vertex' can reach speeds of 2 billion billions per second. This is 100 times faster than the US supercomputer 'Titan'.

And 95% of the 'verified' super computer power comes from Nvidia's GPU, which is a 27,648 Tensor Core Nvidia Volta graphics processor, which allows 'vertex' to exceed 30 billion times per second (3.3 exaops). ) Mixing precision calculations.

Eliu Huerta, a researcher at the National Center for Supercomputing Applications at the University of Illinois at Urbana-Champaign, described the huge 'GPU' GPU pool as 'like a dream'.

The GPU is the image processor, which differs from the general-purpose computer's central processor (CPU) in that the CPU consists of several cores optimized for sequential serial processing, while the GPU has one more than a thousand. Small, more efficient core (designed for simultaneous processing of multiple tasks) composed of massively parallel computing architectures.

A few years ago, researchers found that the parallel operation capability of the graphics processor was very suitable for running deep learning algorithms. Nvidia is also using the wave of artificial intelligence to launch the GPU, rapidly occupying the market and even affecting Intel’s computer chip market. In the place.

Thomas Zacharia, head of the Oak Ridge National Laboratory, said that such a large-scale GPU use is not common for supercomputers. It helps machine learning to make breakthroughs in solving scientific problems.

In addition to the help of Nvidia's GPUs, over 9000 traditional processors from IBM also contributed.

'Vertex' becomes a cutting-edge scientist' playground

Vertex's powerful computing power also convinced scientists that it could create miracles in the future.

The Oak Ridge National Laboratory has become a 'playground' for cutting-edge scientists. Scientists at many colleges and universities are eager to use this computer to accomplish tasks that were previously impossible.

Huerta said that he hopes that 'Vertex' will help analyze about 15TB of images each night from the 'Large Integrated Survey Telescope'.

According to the introduction of the Oak Ridge National Laboratory, the 'verything' of the next 'travel' schedule is very intensive, and projects that have already been on the agenda include:

The first is cancer research. The US Department of Energy and the National Cancer Institute of the United States are working on a project for a distributed learning environment for cancer. Their goal is to develop research tools that can automatically extract, analyze, and classify health data to reveal hidden disease factors. Relationships, such as genes, biomarkers, and the environment.

The second is fusion energy research. Fusion energy has long been a clean, energy-rich representative of energy. Scientists have been hoping to simulate fusion reactors and their magnetically constrained plasmas to accelerate commercial development.

The third is disease and addiction research. In this study, researchers will use artificial intelligence to identify the functions and evolutionary patterns of human proteins and cell systems. These models can help us better understand Alzheimer's disease. Illness or addiction, and inform the drug discovery process.

In addition to the above mentioned items, 'Vertex' can also give more help on chemical and biological issues. Zacharia believes that this supercomputer can use the medical records of 22 million veterans to contribute to the US Department of Energy program. Including about 250,000 whole genome sequences.

'Vertex' provides an unprecedented opportunity for the integration of artificial intelligence (AI) and scientific discovery, enabling researchers to apply machine learning and deep learning techniques to issues such as human health, high-energy physics, and material exploration.

5 facts about the 'vertex' supercomputer

1.20 Billion Floats: This means that if every person on earth does a one-second calculation, it will take a year to reach the 'vertex' of one second.

2.30 Billion Mixed Precision Calculations: This means that if everyone on Earth does a one-second calculation, it will take 15 years to reach the 'vertex' of one second.

3. In the early tests, a genetic research group took advantage of the 'very' supercomputer and spent an hour solving the problem, but if this problem were placed on a traditional computer, it may take 30 years.

4. The footprint of 'Vertex' is equal to the size of 2 tennis courts.

5. The weight of 'top and bottom' is equivalent to the weight of a commercial jet.