Demanding machine learning applications not only require massive amounts of training data, but also powerful computer hardware that can be trained. An analysis by the San Francisco based research firm OpenAI has shown that the computational effort in the largest training runs has increased exponentially since 2012 with a doubling time of 3.4 months and has increased more than 300,000 times over the same period
The trend was buoyant the development of supercomputers such as Sierra and Summit of the US Department of Energy, which use dedicated accelerator chips to accelerate the AI calculation. Now, the IBM Hardware Center, in collaboration with the State of New York, the SUNY Polytechnic Institute, and other members of the AI Hardware Center of IBM, has supplied a new machine for the Department of Computer Science at the Rensselaer Polytechnic Institute (RPI). of-State is optimized for machine learning workloads.
It is referred to as the Artificial Intelligence Multiprocessing Optimized System (AiMOS) (in honor of Rensselaer's co-founder Amos Eaton) and focuses primarily on projects in the fields of biology, chemistry, humanities, and related fields at the new IBM Research AI Hardware Center at the SUNY campus in Albany. It is based on the same IBM Power Systems technology as Sierra and Summit, including IBM Power9 processors and Nvidia Tesla V1
IBM claims that AiMOS is currently the most powerful supercomputer. It is housed in a private university, the 24th most powerful supercomputer in the world and the third most efficient.
"Rensselaer is home to one of the best high-performance computing systems in the US, and the world is pleased to advance our ongoing research on AI, in-depth learning, and on a broad intellectual level," said Shirley Ann Jackson, president of Rensselaer. "Creating new paradigms requires proactive employees, and we look forward to working with IBM and the state of New York to tackle global challenges in ways not previously possible."
AiMOS will provide modeling, simulation, and computation According to John E Kelly III, executive vice president of IBM, is expected to drive the development of AI-optimized computer chips and systems to make AI systems 1000 times more efficient over the next decade. "[AI] … will help us solve some of our most urgent problems, from health care to safety to climate change. To exploit the full potential of AI, the next big opportunity is special computer hardware. "
As fast as AiMOS may be, it's behind Intel's frontera, which can reach a peak performance of 38.7 quadrillion floating point operations per second. Frontera is the fastest computer designed for academic tasks such as modeling and simulation, big data and machine learning, a distinction he received for the second time in November, placing fifth in the Top500 list of AMDs Frontier, a Machine with more than 1.5 Exaflops of theoretical power expected to be delivered to the Oak Ridge National Laboratory in 2021. Intel, for its part, plans to deploy the Aurora cluster of the "Exaflop class" at the Argonne National Laboratory over the next two years, as part of the Department of Energy's Exascale Computing Project, a scholarship program designed to accelerate US exascale computing research National Supercomputing center in Wuxi (built on the Sunway SW26010 processor architecture) and Tianhe-2A in Guangzhou – 3rd and 4th place, respectively, at about 125 top petaflops and 100 top petaflops. Crays Piz Daint is sixth ahead of Trinity at the Los Alamos National Laboratory, Fujitsu's AI Bridge Clouding Infrastructure in Japan and Lenovo's SuperMUC-NG in Germany.
The race between China and the USA is tough. In the Top500 ranking, China surpassed the United States for the first time with 202-143 in the total number of ranked supercomputers two years ago. This trend accelerated the following year. According to the Top500 report from autumn 2018, the number of US supercomputers has dropped to 108, while China's total has risen to 229.
China and the United States ranked in the largest number of supercomputers in Japan, with more than 30 systems. the UK with over 20; France with almost 20; Germany with over 15; and Ireland with just over 10.