Intel today announced the launch of the Edge AI DevCloud, a way to prototype and test AI with the OpenVINO toolkit for edge devices like drones and cameras. Developers can use existing tools and frameworks to test and optimize models in OpenVINO for Intel hardware like CPUs or FPGAs for free.
“With the launch of the Edge AI DevCloud, customers will now be able to model and simulate in the Deep Learning Workbench tool that we launched over the summer, then deploy it for free in the dev cloud on a variety of different hardware configurations to find out what the best hardware landing zone is for them,” Intel VP of IoT Jonathan Ballon said onstage today at the AI Summit 2019 conference in San Francisco.
Ballon offered no specific metrics but in a conversation with VentureBeat today called OpenVINO the fastest growing tool in Intel history.
Intel first made OpenVINO, or Open Visual Inference & Neural Network Optimization, available to developers and manufacturers in May 2018 for deep learning inference with Intel hardware. OpenVINO supports a range of machine learning accelerators from CPUs, GPUs, and FPGAs to the Intel Movidius Neural Compute Stick. The toolkit was updated earlier this year to extend beyond computer vision applications and support voice and NLP models.
Deep Learning Workbench was added for OpenVINO this summer to allow developers to simulate how a network will perform on various hardware configurations and tweak to optimize for power consumption, accuracy, speed, or cost.
“They don’t have to know the architecture of each of these products. All they have to know is what problem they’re trying to solve and the tool does it for them,” he said.
The Edge AI DevCloud beta launched six months ago and is currently being used by 2,700 customers, a company spokesperson told VentureBeat. The Intel DevCloud for data center users was first introduced in 2017.
The company today also introduced the NNP-T pod for data centers, which can hold up to 480 chips in 10 racks, and shared plans to launch the Keem Bay chip for edge deployment in the first half of 2020. Ballon said Keem Bay is able to achieve on-par performance with 1/5th the power of Nvidia’s Jetson Xavier chip for AI on the edge.
Intel also introduced an Edge AI for IoT nanodegree program today with Udacity.
“70% of data is being created is at the edge and only half of that will go to the public cloud. The rest will be stored and processed at at the edge and that requires a different kind of developer,” Ballon told VentureBeat in an interview.
Intel general manager of AI Naveen Rao said that Intel has seen $3.5 billion in revenue from AI so far this year, and that Intel has seen 20% year-on-year growth in edge computing in devices like drones, cameras, robots, and autonomous vehicles.
In other Intel news, CEO Bob Swan vowed the company would deliver 7-nanometer chips by 2021. Advanced Micro Devices (ARM) said they hit the 7-nanometer milestone this summer, a statement Intel refutes by saying the Navi graphics chip actually has 10-nanometers circuitry.