Meta is well known for building the "metaverse", but with the continuous development of Meta, what foundations should be used to stabilize the rapid development of the "metacosm" has become a major problem.
As we all know, VR/AR is considered to be the entrance to the "metaverse", but another important foundation in addition to this is the AI supercomputer. Within Meta's vision, users can work, socialize, etc. within the "metaverse", which will undoubtedly increase the internal processing pressure.
Pictured: Mainland Tianhe No. 1
AI supercomputers are no longer unusual, but Meta hopes to build the world's fastest AI supercomputer.
supercomputer
What is a supercomputer?
Supercomputer refers to a computer that can perform a large amount of data that cannot be processed by ordinary personal computers and high-speed calculations, and the processing power of ordinary personal PCs and supercomputers is equivalent to the speed of family cars and supercars.
In 1960, UNIVAC built the Livermore Atomic Research Computer (LARC) for the U.S. Naval Research Center, which is considered one of the first supercomputers. In the 2010s, China, the United States, the European Union and other countries raced to be the first to create a 1 exaFLOP (capable of achieving exaFLOP) supercomputers.
As of November 2018, all TOP500 supercomputers are all 64-bit, mostly x86-64-based CPUs (Intel's EMT64 and AMD's AMD64 instruction set architecture), and only a few are based on reduced instruction sets (RISCs).
According to the June 2021 TOP500 list, our country's Sunway Taihu Light ranks fourth in the world, and has 188 supercomputers in the top 500 list, ranking first in the world.
What a supercomputer represents is not only the ability to process data, but also a way for a company, a country, to show its capabilities to the outside world. The development of mainland supercomputers was relatively late, starting in the 60s of the last century, and the Galaxy-1 supercomputer was successfully developed on December 4, 1983. Subsequently, the galaxy supercomputer of galaxy two, galaxy three and galaxy four was successfully developed.
Until 2016, the mainland successfully developed the light of Shenwei Taihu Lake, and once won the world's first place.
The supercomputer is done, so what is an AI supercomputer?
AI supercomputer is a computer focused on artificial intelligence, and everyone understands the beginning of artificial intelligence computers, most of which are "deep blue" to defeat chess champion Kasrov in one fell swoop. At the same time, this also makes many people realize the capabilities of AI supercomputers.
How strong is the Meta AI supercomputer?
The supercomputer developed by Meta is called the AI Research SuperCluster, or RSC for short. The company claims that it will be the fastest in the world when completed by mid-2022.
RSCs are high-speed computers specifically designed to train machine learning systems. Meta will use this AI supercomputer to train content moderation algorithms and augmented reality capabilities in business systems.
According to reports, RSC will have a variety of uses: to accelerate Meta's own artificial intelligence research in areas such as embodied intelligence, self-centered and multimodal artificial intelligence, to improve content moderation on platforms such as Facebook and Instagram by evaluating images, videos and texts faster and more accurately, to use more complex language models for training translation and speech recognition, and to advances in augmented reality functions such as object recognition, which is very important for the development of AR glasses.
Kevin Lee and Shubho Sengupta, also said: "We hope that RSC can help us build completely new AI systems, for example, to provide real-time voice translation to a large group of people, each speaking a different language, so that they can work together seamlessly on research projects or play AR games together." ”
Meta engineers began designing AI supercomputers a year and a half ago. The RSC is already in its first phase, with more than 760 Nvidia GGX A100 systems with 6,080 connected GPUs. These types of GPUs are great for solving machine learning problems.
The second phase of the RSC will be completed by the end of 2022 and will contain more than 16,000 GPUs in total. Compared to the 10,000 GPUs of an AI supercomputer built by Microsoft with openAI, a research lab, the RSC can surprise Microsoft.
Pure Storage provides FlashArray and FlashBlade storage for RSCs and says that "RSCs will have unmatched performance to quickly analyze structured and unstructured data." Pure says it helped design Meta's first-generation AI research infrastructure in 2017. This release has 22,000 Nvidia V100 Tensor Core GPUs in a single cluster.
Summary of the full text
The construction of the "meta-universe" is not just a company and a development team.
The "meta-universe" needs more computing power in the future, more talents are needed to join, the current "meta-universe" primary stage has not yet been completed, as a world-renowned enterprise Meta is still building the foundation, only the foundation can be built to build a tall building. The development of the "metaverse" can not rely solely on the power of Meta, Meta is to help everyone to create an open and free world, but the development of this world requires the efforts of tens of millions of people.
The AI supercomputer RSC is a blockbuster, the emergence is the world's first, although the ranking has been changing, but it is believed that RSC will help Meta build a "metaverse" more quickly in the future.