Conquering AI through sheer processing power

This chip has 256 cores and can be stacked with thousands of siblings to deliver a tremendous amount of processing power. The goal: conquer intelligence with sheer power. Credit: KnuEdge

A company, KnuEdge, has been working in the shade over the last 10 years with the goal of creating systems that can crack intelligence by leveraging on tremedous processing power.

The company has been founded by Daniel Golding, former NASA Director, has already raised 100M$ and has now announced two products: 

  • KnuVerse, a chip that can perform secure voice authentication even in very noisy environment (take a look at the clip) using neural networks, and 
  • Knupath, a chip that sports 256 cores to sustain massive parallel processing and specifically designed to support deep learning. 

The first chip is military grade and it is a product that, according to KnuEdge, can change the way we interact with machines. Voice interaction is now becoming more and more common but it often fails, particularly in presence of ambient noise (which is what you have in everyday life!). What KnuEdge claims is that KnuVerse can overcome ambient noise and accurately authenticate a person by having it voice a few words. This is the starting point to develop a system that can identify, and then understand, our talking (the idea is to go beyond speech recognition and move to talk recognition... which is quite more complex).

This requires plenty of AI, Artificial Intelligence, and here comes the second chip, KnuPath, that can deliver massive amount of processing power.

The chip itself contains 256 Cores. It is not a first, NVidia released its Tegra X1 GPU processor with 256 core in January 2015, but it can be stacked with up to 500,000 of its siblings to deliver a staggering processing power.

The idea behind this chip is that one can conquer AI by sheer processing power (and software of course). The neural network based chip has been designed to support deep learning, a technology that is now leading the effort in AI systems.

Whether Intelligence is a matter of processing power or not is not yet clear and there are different opinion. It is a give that you need significant processing power and significant complexity to capture the nuances of intelligence. Whether this is enough or not is open to debate.

Of course, part of the debate is tied to the fact that we do not have a clear, accepted, definition of intelligence...

Author - Roberto Saracco

© 2010-2020 EIT Digital IVZW. All rights reserved. Legal notice. Privacy Policy.

EIT Digital supported by the EIT