The multinational company Meta Platforms, in partnership with Nvidia, has announced the launch of a super-powerful AI-based computer. It has been named the AI Research SuperCluster (RSC) and was designed to advance and implement neural networks.
Meta Platforms’ Latest Development: The Most Powerful AI Computer in the World
Meta, known for its technological projects, has unveiled the world’s most powerful AI-based computer built on Nvidia GGX A100 series GPUs. The system, named AI Research SuperCluster (RSC), is designed for scientific research in artificial intelligence, neural network development, and machine learning.
“Meta’s developments have allowed us to create the world’s most powerful computational device, capable of executing deep learning algorithms for artificial intelligence,” said Meta CEO Mark Zuckerberg. “RSC is designed using advanced technologies and algorithms that can change the world as early as the beginning of 2022,” he stated.
In 2017, Meta Platforms announced a business partnership with Nvidia. Meta developers utilized video cards with the GV100 hardware architecture and assembled a high-tech system consisting of 22,000 such processors. During this time, corporations like Microsoft and Nvidia had already released their own supercomputers, but Meta aimed to create something entirely new. By the end of 2022, their primary goal was achieved.
Research found that the device can process 16 terabytes of data per second, significantly outpacing the speed of other supercomputers. This is due to the fact that the GV100 processors, which form the backbone of Meta’s computer, can operate at a frequency of 1,455 MHz. The system operates on 6,080 accelerators, and the linking element is the NVIDIA Quantum-2 InfiniBand platform.
The Purpose of RSC Development
According to the company, the innovation is unparalleled in the world. The machine is designed to solve complex scientific and industrial challenges. The super-powerful computer is intended for conducting research in AI, including:
- Creating and training thousands of virtual machines in the cloud;
- Identifying potential solutions related to system AI and machine learning algorithm errors;
- Utilizing open-source AI systems provided by Open Source;
- Data processing, forecasting, and decision-making related to cybersecurity;
- Supporting Data Science, Data Analytics, Computer Vision, and other data-driven activities;
- Addressing tasks related to image recognition, data analysis, simulation, and much more.
Specifically, it will be used in fields like computer vision, robotics, and other areas requiring deep data processing. Additionally, the computer will enable work with multiple data sets for AI training, including data from banks, insurance companies, retail networks, and other industries.
RSC will also be used for other types of computations, such as virtual reality, evaluating internet publications, or analyzing user behavior on social networks like Facebook, Instagram, and Twitter.
“This opens new opportunities for using machine learning in various industries, from medicine to energy, to solve complex applied and general-purpose tasks,” said Facebook CEO Mark Zuckerberg. The company expects that solutions based on RSC will speed up the development and implementation of new technologies, such as voice translation for business conferences or virtual reality.
How the Super-Powered RSC Computer Was Created
Development of the computing machine began in 2020. After project preparation, engineers started assembling the first prototype. At its core are 760 NVIDIA DGXA100 systems containing 6,080 accelerator modules. Each processor node is equipped with Tensor cores (Tensor Float). The device is designed for 80 GB of RAM and features PCI Express interfaces.
Even at the initial assembly stage, the supercomputer had performance 20 times greater than other high-performance computing machines, accelerating work with large volumes of industrial automation data, image and video processing, image and speech recognition tasks, and scientific research.
Over the course of the year, the computer’s power will double, and by the beginning of 2021, it will hold the leading position on the list of the world’s most powerful supercomputers. Currently, the computer’s performance reaches 5 exaflops. It was named AI Research SuperCluster and was developed as part of the Open Compute Project (OCP). The computer is capable of solving tasks related to artificial intelligence, machine learning, big data processing, and more.
By the end of 2022, after the completion of testing, Meta will introduce a new version of the ultra-powerful computer, equipped with 16,000 Nvidia processors. The performance of the components will enable the processing of data sets up to 1 exabyte in size or a trillion floating-point operations per second. For comparison, the power of Microsoft’s OpenAI-based supercomputer, which ranks fifth among the world’s most powerful computing systems, is 12 teraflops (1 exaflop = 1 million teraflops). Meta will use Nvidia A100 technology to accelerate GPU computations.
What is an AI-Based Supercomputer?
The capabilities of modern computers are impressive, but the concept of artificial intelligence (AI) and how it operates is still not entirely clear. In theory, it’s simple: AI is a computing system that can learn and self-learn. It can not only predict events but also simulate them.
A supercomputer, for instance, can analyze and forecast a country’s economic situation and recommend actions for the government to improve it. Unlike a human who thinks within the scope of their profession, the supercomputer can solve tasks in various fields.
The high-speed computer can execute trillions of operations per second, and its computational power increases exponentially with its memory volume—meaning the more RAM it has, the faster it works.
Currently, the world’s most high-tech computer is the American supercomputer Summit (OLCF-4), with a power capacity of 122.3 petaflops. It operates on two IBM POWER9 processors and has 250 petabytes of RAM. The computer was launched in June 2018.
What’s the Difference Between Supercomputers and AI Supercomputers?
Both types of computing equipment are similar in terms of power, but the main difference lies in their purpose. Multifunctional supercomputers are typically used for scientific research, such as space exploration, calculating the effects of weapon systems, earthquake prediction, etc. They are key components in many large-scale research projects.
Machine learning does not require high precision from an AI supercomputer, while traditional supercomputers are used to solve problems with more complex algorithms. This allows AI-based supercomputers to process more operations per second, making them suitable for tasks requiring significant computational power but not high precision.