Wave Computing accelerates deep learning by using dataflow technology to eliminate the need for a host and co-processor in the processing of a neural network.
In November, 2018, Wave Computing closed their Series E funding round at $86 million. Total investment in Wave is now over $200 million. Oakmont Corporation led the round.
Open.ai has shown Deep Neural Networks are doubling their performance requirements every three and a half months compared to the traditional Moore’s Law rate for central processing units (CPUs), which have historically doubled every 18 months.
Wave Computing recently acquired MIPS Technologies for CPU IP, has begun initial testing of their first-generation custom systems at end users’ sites for evaluation and feedback.
Wave Computing has asynchronous data flows. This enables faster clock rates and the first Wave chip will have 10 gigahertz clock rate. The company believes the typical throughput will be approximately two-thirds this peak rate.
Wave Computing is not making performance claims yet. They do say they will manufacture 7-nanometer modified MIPS chips through a deal with Broadcom. Wave Computing describes an architecture with more parallelism and higher bandwidth.
WaveFlow™ deep learning systems exploit data and model parallelisms present in convolutional and recurrent neural networks to provide high-performance, high-efficiency training and inferencing computing solutions that scale for any implementation.
Wave Computing’s dataflow-based systems each include Dataflow Processing Units (DPUs) that contain more than 16,000 Processing Elements per chip, high-speed memories, terabytes of storage.
The company’s full dataflow software stack is as follows:
the WaveFlow SDK,
the WaveFlow Agent Library,
WaveFlow Execution Engine, and
the Wave Machine Learning Framework Interface.
The benefits of Wave Computing’s dataflow-based solutions include fast and easy neural network development and deployment using frameworks such as ONNX, TensorFlow, Keras and more.
SOURCES- Wave Computing, Youtube
Written By Brian Wang.