Communications of the ACM
IBM Claims Its Neural Computer Achieves Record AI Model Training Time
IBM have developed a reconfigurable parallel processing system to research and develop emerging artificial intelligence algorithms and computational neuroscience.
Credit: IBM
Researchers at IBM have developed a reconfigurable parallel processing system, called the IBM Neural Computer, in order to research and develop emerging artificial intelligence (AI) algorithms and computational neuroscience.
The first application demonstrated on the Neural Computer a deep "neuroevolution" system that combines the hardware implementation of an Atari 2600, image processing, and AI algorithms in an optimized pipeline.
The IBM Neural Computer is made up of 432 nodes based on field-programmable gate arrays. The nodes are arranged in a three-dimensional mesh topology, interconnected vertically with electrical connections called through-silicon vias that pass completely through silicon wafers or dies.
The team used 26 out of 27 nodes per card within the Neural Computer, completing experiments on a total of 416 nodes.
From VentureBeat
View Full Article
Abstracts Copyright © 2020 SmithBucklin, Washington, DC, USA
No entries found