Digital System Research

DSR releases revealing whitepaper proposal for an RNS based Tensor Processing Unit (RNS TPU)

June 9, 2017 – DSR is proud to announce the release of a highly revealing whitepaper which clearly demonstrates that RNS based arithmetic processing has a place in applications requiring wide precision, high efficiency product summation operations.  These applications include neural networks, deep learning, web search, cloud computing and a host of other numerically intense applications where speed and power efficiency are paramount.  Eric Olsen, president of DSR states ” The Google TPU is a great example of the performance leap that results using a small data width.  Researchers in RNS have been talking about this for years.  The problem with binary is this leap cannot be sustained if the precision or width of the data increases.  With our new RNS arithmetic, this is not a problem, since RNS processing can be broken into distinct and separate digits, with each digit processed using a circuit of least possible arithmetic precision.”  Mr. Olsen adds, “Its more than that, the digit operations are modular so conventional computer wisdom doesn’t apply”.  Eric Olsen further states: “So when we extraploate using our latest test data, we see the RNS TPU architecture represents a huge leap in processing of wide precision numbers.”  Click here to download the whitepaper.

Leave a Reply

Your email address will not be published. Required fields are marked *

Delivering the next leap in computation™