MN-Core
Custom deep learning processor
Overview
To speed up the training of deep learning modeles, PFN is developing the MN-Core™ chip. It is a dedicated accelerator optimized for the matrix computations needed for deep learning. MN-Core is expected to achieve a world class energy efficiency of 1 TFLOPS/W (half precision). By focusing on the functions needed for deep learning, the dedicated chip can boost effective performance in deep learning as well as reduce costs.
Features
Optimized for the training phase in deep learning
Extremely densely integrated matrix arithmetic units
In spring 2020, we’ll launch MN-3, a new large-scale cluster based on MN-Core.
