Neural networking is on the rise. So is the need to process these workloads. By default, this requires compute architectures that can efficiently compute neural network graphs. So, how do traditional CPU-, GPU-, and DSP-based solutions fare in this context?
According to representatives from Blaize (formerly Thinci), not well. Why? Because, among other things, their need to constantly access memory consumes significant time and energy. Dinakar Munagala, the ceo of Blaize, explains how the company?s graph streaming processor (GSP) architecture has been optimized for neural network applications like smart vision and autonomous driving.
Here, two demos of Blaize technology show how the company?s solution can be applied in autonomous driving and facial recognition use cases.
WATCH NOW