Huawei has officially launched the most powerful Artificial Intelligence (IA) processor – the Ascend 910 – and the IA framework for all scenarios, MindSpore.
“We have been making steady progress since we announced our AI strategy last October,” said Eric Xu, Rotating Chairman of Huawei. “Everything is evolving as planned, from R&D to product launch. We promise a full-stack AI portfolio for all scenarios. And today we have done it with the release of Ascend 910 and MindSpore. This marks a new step in Huawei's AI strategy. ” Huawei's Rotating Chairman Eric Xu announces the launch of the Ascend 910 IA processor and the MindSpore IA framework on August 23, 2019.
Ascend 910: More computing power than any other AI processor
The Ascend 910 is a new IA processor that belongs to Huawei's Ascend-Max chipset series. The company had announced planned processor specifications in 2018 at the Huawei Connect event. After a year of continuous development, test results show that the Ascend 910 processor meets performance targets with lower power consumption than originally planned.
For half precision (FP16) floating point operations, the Ascend 910 offers 256 TeraFLOPS. For integer precision calculations (INT8), the processor provides 512 TeraOPS. Despite its unrivaled performance, the Ascend 910's maximum power consumption is only 310W, far below the planned specifications (350W).
“The Ascend 910 works better than we expected,” said Xu. "It has undoubtedly more computing power than any other AI processor in the world."
The Ascend 910 is used for IA training model. In a typical ResNet-50 based training session, the combination of Ascend 910 and MindSpore is about twice as fast in training AI models than on other conventional training boards using TensorFlow.
Going forward, Huawei will continue to invest in AI processors to provide better, more affordable and adaptable computing power that meets the needs of a wide range of scenarios (eg edge computing, standalone vehicle computing and training). ).
MindSpore: IA framework for all scenarios
Huawei also today launched MindSpore, an AI framework that supports the development of AI applications in all scenarios.
AI frameworks are essential for making AI application development easier, making applications more widespread and accessible, and ensuring privacy protection.
In 2018, Huawei announced the three development goals for its AI framework:
- Easy development: Dramatically reduced training time and costs;
- Efficient execution: Utilizing the least amount of resources with the highest possible OPS / W;
- Adaptable to all scenarios: Including device, edge and cloud applications.
MindSpore marks significant progress toward these goals. As privacy protection becomes increasingly important, support for all scenarios is essential to enable secure and widespread artificial intelligence. This is a key component in the MindSpore framework that can adapt to different implementation needs.
MindSpore helps ensure user privacy by only dealing with gradient and template information that has already been processed. It does not process the data itself, so private user data can be effectively protected even in cross-scenario environments. In addition, MindSpore has built-in model protection technology to ensure models are safe and reliable.
MindSpore's AI framework is adaptable to all scenarios – across all devices, edge and cloud environments – and provides on-demand cooperation between them. Its “AI Algorithm as Code” design concept enables programmers to easily develop advanced AI applications and train their models faster.
In a typical natural language processing (NLP) neural network, MindSpore has 20% fewer lines of core code than the industry's leading frameworks and helps programmers increase their efficiency by at least 50%.
Through framework innovation as well as optimization of MindSpore and Ascend processors, Huawei's solution can help programmers more efficiently address complex AI challenges and the need for a diverse range of computing resources for different applications. This results in better performance and more efficient execution. In addition to Ascend processors, MindSpore also supports GPUs, CPUs and other types of processors.
In introducing MindSpore, Xu emphasized Huawei's commitment to helping build a more robust and vibrant AI ecosystem. "MindSpore will be open source in the first quarter of 2020. We want to promote broader AI adoption and help programmers do what they know best."
Huawei's strategy for Artificial Intelligence
Before announcing the release of Ascend 910 and MindSpore, Xu revisited Huawei's AI strategy:
- Invest in IA Research: Develop fundamental skills in machine learning, natural language processing, decision and inference, etc. With focus in:
- Data and energy efficiency (e.g., use less data, computation and energy)
- Security and credibility
- Automation / Autonomy
- Build a full-stack AI portfolio
- Adaptable to all scenarios including standalone and cooperative cloud, edge and device scenarios
- Abundant and affordable computing power
- Efficient and easy-to-use full service AI platform
- Cultivate talent and an open ecosystem: Collaborate broadly with academia, industries and partners globally
- Strengthen Existing Portfolio: Bring an AI mindset and techniques to existing products and solutions to create more value and increase competitive strengths.
- Boost operational efficiency: Use AI to automate high volume repetitive tasks for better efficiency and quality.
Huawei's AI portfolio covers all deployment scenarios, including public and private cloud, edge computing, IoT industry devices, and consumer devices. The portfolio is also full-stack: it includes the Ascend IP and chip series, the CANN chip enable layer, the MindSpore inference and training framework, and the ModelArts application enable platform.
Huawei defines AI as a new general-purpose technology, such as railroads and electricity in the nineteenth century, or cars, computers and the Internet in the twentieth century. The company believes artificial intelligence will be used in almost every sector of the economy.
According to Xu, AI is still in its early stages of development and there are a number of gaps to be completed before AI can become a true general purpose technology. Huawei's artificial intelligence strategy is designed to bridge these gaps and accelerate adoption on a global scale. Specifically, Huawei wants to drive change in ten areas:
Widespread adoption of Ascend 310 and ModelArts
At Huawei Connect 2018, the company announced its AI strategy and its full-stack AI portfolio of all scenarios, including the Ascend 310 AI processor and ModelArts, which provide full channel model production services.
The Ascend 310 is Huawei's first commercial AI chip on an Ascend-Mini series (SoC) chip. With a maximum power consumption of 8W, the Ascend 310 offers 16 TeraOPS full precision (INT8) and 8 TeraFLOPS half precision (FP16), making it the most powerful AI SoC for edge computing. The processor also has a 16 channel FHD video decoder.
Since its launch, the Ascend 310 has had wide adoption in a wide range of cloud products and services. For example, Huawei's Mobile Data Center (MDC), which uses the Ascend 310, has been used by many leading manufacturers of shuttles, new energy vehicles and autonomous steering.
The Ascend 310 processor series Atlas acceleration card and server are now part of dozens of industry solutions (eg smart transport and smart grid) developed by dozens of partners.
The Ascend 310 also enables Huawei Cloud services such as image analysis, optical character recognition (OCR) and intelligent video analysis. There are over 50 APIs for these services. Currently, the number of API calls per day has exceeded 100 million and this number is estimated to reach 300 million by the end of 2019. More than 100 companies are using Ascend 310 to develop their own AI algorithms.
Huawei ModelArts provides model development services spanning the entire pipeline, from data collection and model development to training and model deployment. Today, more than 30,000 programmers are using ModelArts to handle over 4,000 training tasks per day (for a total of 32,000 training hours). Among these tasks, 85% are related to visual processing, 10% are to audio data processing and 5% are to machine learning.
With today's release of Ascend 910 and MindSpore, Huawei has unveiled all the key components of its full-stack AI portfolio for all scenarios. “Everything is evolving as planned. We promise a complete AI portfolio with all scenarios. And today we fulfill”Said Xu. This release marks a new milestone in Huawei's AI path; It is also a new beginning.
Huawei is working closely with its partners to make AI more widespread and affordable, and help bring the benefits of digital technology to all people, homes and organizations.