MWC22 | Intel launches a new version of OpenVINO to enable developers to accelerate AI reasoning

MWC22 | Intel launches a new version of OpenVINO to enable developers to accelerate AI reasoning

The Intel OpenVINO toolkit has undergone a major upgrade to easily accelerate AI inference performance.

Latest news: Since launching OpenVINO™ in 2018, Intel has helped hundreds of thousands of developers significantly improve AI reasoning performance and expand their applications from edge computing to enterprises and clients. Intel launched a new version of the Intel® Distribution of OpenVINO toolkit on the eve of the 2022 Mobile World Congress in Barcelona. The new features are mainly developed based on developer feedback over the past three and a half years, including more deep learning model choices, more device portability options, and higher reasoning performance with fewer code changes.

Adam Burns, vice president of OpenVINO Developer Tools at Intel Network and Edge Group, said: “The latest version of OpenVINO 2022.1 was developed based on feedback from hundreds of thousands of developers over the past three years to simplify and automate optimization work. The latest version adds hardware auto-discovery and auto-optimization capabilities, allowing software developers to achieve the best performance on any platform. The software, combined with Intel silicon, can achieve significant AI return on investment benefits and can be easily deployed to Intel-based solutions in user networks.”

About OpenVINO: The Intel Distribution of OpenVINO toolkit for high-performance deep learning is developed based on oneAPI to help users deploy more accurate real-world results to production systems faster on a variety of Intel platforms from edge to cloud. Through a simplified development workflow, OpenVINO enables developers to deploy high-performance applications and algorithms in the real world.

Why it matters: Edge AI is transforming every industry, enabling new and enhanced use cases, including manufacturing, health and life sciences applications, as well as retail and security. According to Omdia, global edge AI chipset revenue will reach $51.9 billion by 2025, driven by increasing enterprise demand for edge AI inference. Edge inference reduces latency, reduces bandwidth requirements, and improves performance, meeting the increasing demands for timely processing from emerging IoT devices and applications.

At the same time, developers’ workloads are growing and changing, requiring simpler, more automated processes and tools that have comprehensive intelligence to optimize performance from build to deployment.

About OpenVINO 2022.1 features: With these new features, developers can more easily adopt, maintain, optimize, and deploy code across a wider range of deep learning models. Highlights include:

Easier update API

• Reduce code changes when converting from frameworks: Conversion is now reduced by preserving exact formatting; models no longer require layout conversions.

• A simpler way to accelerate AI: The Model Optimizer API parameters have been reduced to minimize complexity.

• Train with inference in mind: OpenVINO training extensions and the Neural Network Compression Framework (NNCF) provide optional model training templates that can further improve performance while maintaining accuracy for action recognition, image classification, speech recognition, question answering, and translation.

Wider model support

• Support for a wider range of natural language programmable models and use cases such as text-to-speech and speech recognition: Dynamic shape support better enables the BERT series and Hugging Face transformer.

• Optimization and support for advanced computer vision: The Mask R-CNN series is now further optimized and introduces support for double-precision (FP64) models.

• Direct support for PaddlePaddle models: Model Optimizer can now directly import PaddlePaddle models without converting to another framework first.

Portability and performance

• Smarter device usage without code modifications: AUTO device mode can automatically discover available system reasoning capabilities based on model requirements, so applications no longer need to understand their computing environment in advance.

• Expert optimization built into the tool suite: Improve device performance through automatic batch processing, automatically adjust and customize system configurations and throughput settings for deep learning models to suit developers, thus enabling developers to achieve scalable parallel processing and optimized memory usage experience.

• Built for 12th Gen Intel® Core™: Supports hybrid architecture, providing enhanced capabilities for high-performance inference using the CPU and integrated GPU.

About edge adoption: With a “write once, deploy anywhere” approach, developers only need to write applications or algorithms once and then deploy them to a wide range of Intel architectures including CPU, iGPU, Movidius VPU and GNA. As data explodes, Intel develops software that enables developers to process data more intelligently to solve challenges and transform business models. As a result, new and unique AI inference technologies are increasingly being adopted at the edge and extended to enterprises and clients.

As Zeblok's AI platform as a service, AI-MicroCloud is a cloud-to-edge MLDevOps platform that allows customers to mix and match AI independent software developers and vendors on a large scale to deliver edge AI applications while supporting full lifecycle deployment. After integrating Intel OpenVINO software into AI-MicroCloud, the use of Intel processors will greatly enhance AI reasoning performance and minimize the cost of each insight. Zeblok's AI-MicroCloud platform is currently being evaluated to support specific network topologies in cities around the world.

“Our mission is to think about cost per insight,” said Mouli Narayanan, founder and CEO of Zeblok. “By using Intel processors, we have enabled cost-effective and energy-efficient AI inference and generated a very high return on investment. This new version of OpenVINO will create even greater value for our ecosystem.”

American Tower has built six edge data centers and will build more in the future. The company recently acquired CoreSite to accelerate 5G edge deployment. They are working with Intel and Zeblok to provide customers with a complete turnkey solution.

“With American Tower’s edge infrastructure, Intel’s OpenVINO deep learning capabilities, and Zeblok’s AI platform as a service, we can deliver a complete intelligent solution to the market,” said Eric Watko, vice president of innovation at American Tower.