AiMOS, Son of Amos

Original article was published on Artificial Intelligence on Medium

The sophistication of Artificial Intelligence has been long been identified as the gateway into the next frontier of computing innovation. It’s what brought us the internet-of-things (IoT), it help’s us create shopping lists, find directions, or a good place to eat nearby. It’s also the fringe-science behind self-driving cars, speech and facial recognition, deep-fakes, optimized advertising and countless other applications that are still so new to us.

The location of AiMOS’ birth, The IBM Research AI Hardware Center, has only just opened at the SUNY Polytechnic Institute early last year. New York State, along with IBM, SUNY Polytechnic Institute, and a vast network of other founding partnership members have conglomerated to create a global AI research powerhouse. Collectively, their bravura is supercharging AI’s true potential in next generation’s hardware and software capabilities.

The roster behind IBM’s Research AI Hardware Center is filled with juggernauts. Some of it’s partners include: Samsung and IBM in the realm of software, prototyping, IP and research, Applied Materials and Tokyo Electron Limited (TEL) in the hardware and materials sector, and the recent infrastructure expansion at host SUNY Polytechnic Institute alongside its neighboring Rensselaer Polytechnic Institute (RPI) Center for Computational Innovations (CCI) that has opened the door for increased academic collaboration. This unprecedented collaboration of industry leading trailblazers has brought the most powerful supercomputer ever housed at a private university into existence.

In a statement made by IBM Executive Vice President, Dr. John E. Kelly III:

“IBM is proud to have built the most powerful and smartest computers in the world today, and to be collaborating with New York State, SUNY, and RPI on the new AiMOS system. Our collective goal is to make AI systems 1,000 times more efficient within the next decade.”

The $2 Billion program is starting to look like money well spent. In November of 2019’s edition of Top500 and Green500 supercomputer rankings, AiMOS was ranked the 24th most powerful computer in the world, and the third-most energy efficient.

AiMOS in short stands for Artificial Intelligence Multiprocessing Optimized System, and it cost nearly $300 million to build and operate. Its cross-vendor Heterogeneous System Architecture (HSA) enables AiMOS to perform at a capped rate of up to eight quadrillion calculations per second. HSA specifications allow for a reduction in communication latency by integrating the central processing units (CPUs) and graphics processors (GPUs) on the same bus and sharing the same tasks and memory.

Most modern GPU’s are calibrated for single instruction, multiple data (SIMD) and single instruction, multi threads (SIMT). HSA on the other hand, originally championed by AMD as “FSA”, extends all instruction processing to all processing units regardless of vendor. It effectively offloads instructions over the span of the system, combining the power-spread of its internal units. This is great for neural networks, as they are made up of multitudes of processing components.

Under the hood, AiMOS houses IBM POWER9 CPUs and NVIDIA GPUs. The eight petaflop state of the art POWER9 chip is the most performance-hungry I/O processor ever designed. It’s Ultra-high bandwidth CPU-GPU interconnection is the optimal specification for training AI models and maintaining low latency communications between accelerators and storage controllers.

AiMOS is open to private and public partnerships that is critical to the technological state of the globe. One initiative, the Rensselaer-IBM Artificial Intelligence Research Collaboration (AIRC), is a network that connects institutions to explore the frontiers of AI. Another joint collaboration between Cognitive and Immersive Systems Lab (CISL), and The Jefferson Project, combines the expansive network of IoT with expert analytics to generate blueprints for preserving bodies of freshwater around the globe.