Skip to content

Latest commit

 

History

History
11 lines (10 loc) · 849 Bytes

PURPOSE.md

File metadata and controls

11 lines (10 loc) · 849 Bytes

Numerical examples

The Universal library contains parameterized number systems to aid in the research to create custom algorithms that are tailored to the arithmetic requirements of the applications. In high-performance and real-time applications, this tailored number system approach has been a common approach to maximize performance and/or performance per Watt. More recently, Deep Learning has rekindled the interest in tailored number systems as training DNNs is a lengthy affair and improving computational performance by minimizing hardware, memory or network bandwidth, had immediate commercial benefits. As Deep Learning AI is getting embedded in stand-alone devices, the benefit of multi-precision number system optimizations is broadening to any application that uses deep computes to deliver knowledge generation or decision making.