IBM unveils a server for artificial intelligence applications

Systems Power Systems The news provides 4 times higher performance in the field ofDeep learning compared to systems x86* Google and the supercomputers of the US Department of Energy use processors POWER9 * Three Israeli teams were partners in the global development

IBM unveils the new generation of Power Systems servers, which incorporate the new POWER9 processors developed by the company. The new system, which was developed especially for heavy workloads in the fields of artificial intelligence, is able to improve and shorten the training time of deep-learning based systems by up to four times - in a way that allows organizations to build more accurate deep learning applications, in a faster process than ever before.

Servers AC922 Power Systems The new ones based on POWER9 processors are the first to incorporate technologies such as the next generation of NVIDIA NVlink, PCI Express 4.0 and OpenCAPI, which accelerate data traffic up to 10 times compared to x86-based servers. All of these provide provable performance improvements around the popular AI frameworks such as Chainer, TensorFlow and Caffe, as well as advanced databases such as Kinetica. As a result, data scientists can build applications in a faster process - from generating deep insights in scientific research and medical imaging tasks, through real-time detection of fraud and risks in the operation of credit card systems.

Three teams from IBM's hardware development labs in Israel were involved in the development process of the new system. The teams took part in building the dedicated memories of the POWER9 processor and focused on high-density memories. In addition, they took part in the development of the OpenCAPI, which enables fast interfacing between the processors and the hardware accelerators and between them and the operating system, which results in the particularly high performance of the system.

POWER9 is installed in the core of the US Department of Energy's "Summit" and "Sierra" systems, which will soon become the fastest supercomputers in the world. In addition, they are used by a number of leading organizations, including Google. "Google is excited about IBM's progress in developing the new Power technology," said Bart Sano, vice president of platforms at Google. "The capabilities of OpenCAPI and the large memory capacities create opportunities for further innovation in Google's computing centers."

According to Moshe Leibovitch, director of the hardware development laboratory in Israel: "We had the honor to be partners in the development of the groundbreaking systems, together with IBM's global teams. This is a computing powerhouse that changes the rules of the game in the world of artificial intelligence and cognitive applications. IBM POWER9 systems will allow organizations all over the world to generate insights on an unprecedented scale, promote scientific discoveries and bring about business change in all areas of industry."

Race to the future with POWER9 processors

Deep learning is a rapidly developing machine learning methodology that extracts information by processing millions of processes and data to identify and rank the most important aspects of the data.

To meet these evolving industry needs, four years ago, IBM set out to design POWER9. This design started from the ground up - and is intended to build a completely new architecture that will allow managing free-flowing data, as well as data collected by sensors continuously, and processing advanced algorithms used in artificial intelligence and deep learning applications on Linux.

According to IBM's announcement, it is the only supplier in the market that currently offers a comprehensive infrastructure package for the world of artificial intelligence, which combines advanced hardware systems of its kind and software tools based on the latest open source innovations.

With the help of the PowerAI platform for artificial intelligence and deep learning applications, IBM optimizes and simplifies the tasks of setting up and deploying deep learning frameworks and software libraries that run on top of the Power architecture together with application acceleration tools - so data scientists can set up and run such new applications in just a few minutes.

IBM makes investments in cognitive systems, and the company's research laboratories develop a wide range of technologies for the POWER9 environment. IBM researchers have already managed to cut the time required to train deep learning systems from the days it took in the past - to only a few hours, with the help of the company's dedicated toolkit for distributed deep learning, PowerAI Distributed Deep Learning toolkit.

Building an open ecosystem to accelerate innovation

The era of artificial intelligence requires much more than tremendous processing power and unprecedented speeds: what is also required is an open ecosystem of innovative companies that will provide appropriate technologies and tools.

IBM promotes a community that already numbers more than 300 companies within an association OpenPOWER Foundation and a consortium OpenCAPI. The list of members in these bodies includes names such as SuperMicro, Inspur, Xilinx, Nutanix, NVIDIA, Inventec, Wistron, the Israeli Mellanox Technologies, Amphenol, Rackspace, Molex and Alpha Data, Toshiba, Google, Rambus, MicroSemi, Tektronix and Western Digital. Many of these partners are expected to launch POWER9-based products and services soon.

See more on the subject on the science website:

• For the first time: IBM introduces a quantum computer with a 50-bit processor
• The three questions that artificial intelligence will not be able to answer
• What did you want her to understand?
• Tomorrow's medicine in the era of data mining and Big Data