Nvidia dominates the market for machine learning chips because its graphics processors can process huge amounts of information across thousands of cores in parallel. But to hold onto that early lead, the company has made major investments in its CuDNN software, adding support for every major environment for training neural networks, like TensorFlow and PyTorch.
The company’s rivals are taking pages from its playbook. That includes Intel, which is trying to boost its reputation in the machine learning market. Last week, the Santa Clara, California-based company said that it had purchased artificial intelligence startup Vertex, which builds tools for running algorithms more efficiently inside data centers and embedded devices. The terms of the deal were not disclosed.
“There’s a large gap between the capabilities neural networks show in research and the practical challenges in actually getting them to run on the platforms where most applications run,” Choong Ng, founder and chief executive of the seven-person startup, said. “The lack of portable, developer-friendly tools prevents most organizations from realizing the power of deep learning for their business.”
The Seattle, Washington-based Vertex developed a software tool to cut down on the complexity of running these algorithms in hardware. The company said that PlaidML would be maintained as part of Intel’s nGraph tool, which allows code from TensorFlow or any other machine learning environment to be translated and used by Intel’s hardware. Intel is trying to win over more customers by easing programming.
The company introduced another tool, OpenVINO, targeting the development of computer vision applications that identify product defects on manufacturing lines, manage inventory in warehouses or handle other tasks regardless of the hardware. That includes Intel’s field-programmable gate arrays – commonly known as FPGAs – and vision processing units, like the Myriad X. The software is used by Honeywell, General Electric and Amazon, among others.