TECH SUPPLIER Oct 2021 - Technology Assessment - Doc # US48284521

The Skyrocketing Computational Demand of Deep Neural Networks

By: Fangwen Xia, Peter RuttenResearch Director, Infrastructure Systems, Platforms and Technologies Group, Performance Intensive Computing Solutions Global Research Lead

Abstract

This IDC study discusses some of the application areas that deep neural networks cover today, the computational needs of several of these application areas, and the economic cost associated with that computational need. We are experiencing incredible insights and even joy from fast advancing deep learning applications including interactions with smart devices. But ongoing research and published data have shown that due to the costly nature of deep learning, development will soon reach a point of stagnation.

"We expect that for certain deep neural networks, the future computational demand will exceed the capabilities of today's hardware," said Fangwen Xia, research associate at IDC's Infrastructure Group. "Software developers should gain more familiarity with the hardware for their program and reflect on its resiliency in error baring, while vendors must keep pushing for greater compute capacity with lower precision."


Coverage

Content


Get More

When you purchase this document, the purchase price can be applied to the cost of an annual subscription, giving you access to more research for your investment.



Related Links

Do you have questions about this document
or available subscriptions?

Contact Us