Facebook Parent Meta says its new AI computer will be the fastest in the world in mid-2012. Uses the research supercluster of AI to develop new meta over sensations. Based on preliminary benchmarks, the company claims that RSC might use a 20x faster machine vision workflow and more than 9x faster NVIDIA crowdsource library compared to the previous company’s configuration.
Meta says the RSC can train large-scale natural language processing models three times faster. Therefore, AI models that determine if “the action, sound, or image is dangerous or inappropriate” (e.g. stop hate speech) can be trained very quickly. As per the company release information, this research will help protect people on services such as Facebook and Instagram, as well as the metaverse.
In addition to creating tangible infrastructure and operating systems for the RSC, Meta said it needs to ensure security and privacy controls are in place to protect the real-world training data it uses.
The company claims that by using real data from its production systems instead of publicly available datasets, it can use its research effectively, for example, to identify malicious content.
This year, Meta plans to increase the number of GPUs in the RSC to 16,000 to improve AI training performance by more than 2.5x. The company, which began work on the project in early 2020, wanted RSC to train AI models on exabyte-sized datasets (equivalent to 36,000 years of high-quality video).
Some exascale systems were being built in the US: The Aurora supercomputer at the US Department of Energy’s Argonne National Laboratory is slated to hit 2 exaflops, while the Supercomputer El Capitan, which will transport nuclear material throughout the country, affects two Eaflops next year.
In terms of storage, the supercomputer features 175 petabytes of Pure FlashArray, 46 petabytes of cache on Penguin Computing Altus systems, and 10 petabytes of Pure FlashBlade.
This big AI computer isn’t complete yet, but benchmark tests show it’s 20X faster than its legacy infrastructure on computing workloads. It also leverages Nvidia’s Integrated Communication Library (NCCL) nine times faster, training 3D models and processing native language.
The company also emphasized that security and privacy are at the core of this great computer, which allows researchers to train their models with encrypted, user-generated data “that is only specified before training begins.
The RSC also has no direct internal or external internet connection and traffic flows only from Meta’s production data centers and has encrypted encryption between endpoints and GPUs.
” It will also measure the final system, which currently operates at 16 TB / S bandwidth for data formation, up to 1 Exabyte. The company claims it will be the “world’s fastest AI supercomputer”, with around 5 exaflops of precision computing.
What more Facebook Parent Meta shared about the fastest computer in the world!
Built by Fujitsu and the National Research Institute of Japan RIKEN, the Fugaku supercomputer justifies its title as The fastest computer in the world, beating its competitors in China and the United States for points per second.
In second place was IBM’s Supercomputer, which received just 148 petaflops. The standard, prepared by an international team of experts, is published in June and November last year.
Fugaku also falls into three other categories, which include high-performance intelligence and data processing capabilities.
The next generation of supercomputers will follow the Japanese supercomputer, which ranked # 1 in 2011. The $ 130 billion plan ($ 1.22 billion) went into effect in March.
The Japanese Automobile Manufacturers Association plans to use Fugaku.
Ricken concluded that by developing Fukuoka, companies could easily create software. Earlier this month, Fujitsu Japan began researching chemicals used to treat COVID19 using the high counterfeit power of fuka.
Although Fuga is also prevalent, the development of supercomputers in recent years has become a competition between the United States and China. Both countries hope to use them not only for industrial purposes but also for nuclear weapons as well as a military research.
The United States is using the next generation of machines that can perform ex-scale calculations. 1 Xflop equals 1000 petaflops, which is at least twice as much as Fuga. Oak Ridge National Frontier Laboratory and Argonne Aurora National Laboratory are set to make their debuts this year.
Development efforts in China remain a mystery, but there are rumors that it could inherit the world’s fastest supercomputer. It could launch its exascale machine this year. The United States and China are also in the race to develop quantum computers.
The race to develop computers has become a matter of national recognition due to its impact on industrial competitiveness and national security. In April, the US Department of Commerce suspended exports and added seven Chinese supercomputing companies to its list.