The Rensselaer Polytechnic Institute in New York has just revealed that it is treating itself to state-of-the-art technology to accelerate research and development in the field of artificial intelligence.
The Institute unveiled AiMOS, IBM's latest eight-petaflop supercomputer – a device capable of running eight quadrillion calculations per second. That's roughly the same as if every person on the planet ran one million calculations in their heads at the same second.
IBM said that AiMOS is the most powerful supercomputer currently housed at a private university. The device has been running at Rensselaer since October, and it was just announced that AiMOS will be used to advance the development of stronger hardware that can support new AI applications.
Last November, the supercomputer made its debut on the Top 500, a bi-yearly list that ranks the most powerful non-distributed computer systems in the world. AiMOS ranked 24th worldwide.
IBM, however, said that AiMOS uses the same Power Systems technology as the two most powerful devices in the world, Summit and Sierra, which are also built by IBM and are used by the US's Department of Energy.
In Rensselaer, IBM's supercomputer will be used to build new computers that can perform better and handle the ever-heavier workloads that are required by artificial intelligence.
John Kelly, IBM's executive vice president, said: "Our collective goal is to make AI systems 1,000 times more efficient within the next decade."
As AI finds more applications and algorithms grow increasingly complex, so does the need to build hardware platforms that can process sophisticated training models while also being cost-efficient.
Consulting firm McKinsey predicted that the demand for computing hardware will increase by 10 to 15% by 2025, and that opportunities to innovate will emerge particularly for semiconductor companies.
New hardware will also have to come in the form of GPUs, CPUs and accelerator chips. In other words, the better AI software gets, the more sophisticated the hardware it runs on will have to be.
According to IBM, AiMOS will be working on the computing chips and systems for next-generation AI workloads, by providing "the modeling, simulation and computation necessary to support the development of this hardware".
"In order to realize AI's full potential," said Kelly, "special purpose computing hardware is emerging as the next big opportunity."
The new supercomputer was delivered to Rensselaer by IBM's Hardware Center, which opened in partnership with the State University of New York (SUNY) Polytechnic Institute this year.
The Hardware Center focuses on finding new devices and architectures to improve algorithm processing efficiency, and creating hardware that can handle deep neural networks in a cost-effective way.
But AiMOS will also be available to support projects already under way for faculty members and staff at Rensselaer.
For example, the supercomputer will assist the Jefferson Project, an environmental-monitoring initiative that gathers more than nine terabytes of data annually from sensors and IoT devices distributed around Lake George in New York.
Although no exact figures were disclosed by IBM nor Rensselaer, the deployment of AiMOS was part of a $30 million investment from New York's chief economic development agency, the Empire State Development Corporation (ESD).
IBM's supercomputer, however, is not the fastest of them all within the academic world. The publicly-owned University of Texas this year revealed Frontera, a supercomputer built with Dell and Intel, and which comes in fifth on the Top 500 list.
Frontera can achieve 38.7 quadrillion operations per second, which places the device well ahead of AiMOS and effectively makes Intel's supercomputer the most powerful one in academia.