Although the company's primary market is hyperscale data centres, Tachyum's CEO Dr. Rado Danilak recognized that government agencies - notably military and intelligence - are early adopters of Prodigy. Its ultra-low-power, ultra-high-performance enables and vastly improves next-generation defense systems such as unmanned aircraft and underwater systems, cybersecurity, communications, analytics, and more.
Prodigy's "unique attributes set the conditions to create a computing architecture fully aligned with the operational and strategic imperatives of our national strategy", stated retired Army Lieutenant General Richard Zahner. He is a career electronic intelligence specialist who headed signals intelligence at the National Security Agency (NSA) and ended his career as Army deputy chief of staff for intelligence. "The new microchip could help restore America's technological edge. This sort of re-levels the playing field potentially in our favour", Richard Zahner stated.
"AI and HPC are strategic to US, EU and NATO members' government, military and intelligence environments, and as demand grew rapidly, we saw the need to organize a separate unit around these projects", stated Dr. Radoslav Danilak, Tachyum founder and CEO. "Supercomputers, defense, national security, government and intelligence agencies are interested in Prodigy because it will execute human brain-scale AI 15 years ahead of any other technology, achieving breakthroughs other governments cannot."
Prodigy excels in technologies such as edge computing, IoT, HPC, convolutional AI, explainable AI, general AI, bio AI and spiking neural networks. For example, translating between English and Chinese requires a neural network with a capacity of more than 11 terabytes and is an arduous task on currently available "supercomputer" GPU processors of 20GB each. In contrast, Tachyum's Prodigy fits 8TB per chip which is 32TB in coherent DRAM per node.
Tachyum's team already includes several personnel carrying top-secret clearances, providing appropriate security levels for developing federal projects. Team members, including Tachyum founders, serving government customers have worked on classified technologies at entities such as Sandia and Skyera, and worked for the U.S. government and/or military.
Demanding AI/ML applications are currently run on specialty, single-purpose chips such as GPU and TPU, while CPUs power conventional workloads - resulting in high costs and inefficiency due to dual infrastructures. Instead, Tachyum's Prodigy supports both on a single homogeneous processor platform with its simple programming model. Prodigys ability to seamlessly switch among various workloads dramatically changes the competitive landscape and the economics of data centers.
Prodigy outperforms today's fastest Xeon processors while consuming 10x lower power on data centre workloads, and outperforms NVIDIA's fastest GPU on HPC, AI training and inferencing. A mere 125 HPC Prodigy racks can deliver 32 tensor EXAFLOPS. Prodigy's 3x lower cost per MIPS and 10x lower core power requirements translate to a 4x lower data centre Total Cost of Ownership (TCO), enabling savings of millions to billions of dollars.
Prodigy will enter volume production in 2021.