Wave Computing announced that it will contribute its variable fixed-point training technology, called Versipoint, to the TensorFlow open source development community. Wave asserts that Versipoint enables data scientists to quickly train neural networks without the need for energy-consuming floating point hardware, such as in a GPU.
Versipoint helps Wave’s systems deliver greater compute efficiency and better overall deep learning performance.
TensorFlow is an open source software library released in 2015 by Google to make it easier for developers to design, build, and train deep learning models.
“Wave Computing’s generous offer to release its Versipoint technology to the TensorFlow community will bring more efficient machine learning training to a wider group of researchers and practitioners,” said Kevin Krewell, Principal Analyst at TIRIAS Research.
Wave explains that its Versipoint technology allows more efficient, non-floating point hardware to be used for training neural networks by reducing storage and processing requirements by 2x or more. Enabled by the programmable and reconfigurable nature of Wave’s dataflow architecture, Versipoint technology allows for a wide variety of mode changes to occur dynamically at runtime.
Versipoint technology is part of Wave’s complete system solution that includes runtime software, schedulers, pre-developed and pre-compiled software modules and more, all of which is “under the hood” and ships as part of a Wave Computing AI system for either on-premise or data center environments.