Home > News content

Intel AI Conference: New AI chip and AI road map for this giant

via:网易智能     time:2018/5/24 11:51:43     readed:1363

Talk about chips: push new AICloud chipPerformance increase 3-4Times

At the WSJ D.Live conference on October 17, 2017, Intel officially released a neural network processor (NNP) series chip designed for machine learning. The chip code was named LakeCrest.

DevelopmentThe developer will be able to develop new AI applications and maximize data processing capabilities, including health care, social media, automotive, weather, etc. He also mentioned that the development of Nervana products will be Intel's AI business in 2020. To achieve the goal of increasing the growth by 100 times.

Among them, Intel's $350 million purchase of Nervana Systems has played a key role, and this time in the seven months after the release of code-named LakeCrest, the Intel Neural Network Processor (NNP) series of chips was upgraded again, the new code-named Spring Crest.

According to Naveen Rao, the new version of the NNP chip will bring a number of updates, which will be Intel's first commercial NNP product, and will support bfloat16, an industry-wide digital format widely used in neural networks, over time. Intel will expand support for bfloat16 on AI product lines, including Intel Xeon processors and Intel FPGAs. He also stated that the Intel Nervana NNP target is a true model parallel that can achieve high compute utilization and support multi-chip interconnects.

In addition, he stated that the performance of the Nervana NNP-L1000 (Spring Crest) released this time was 3-4 times higher than the previous generation, and it is planned to be officially opened in the second half of 2019.

Developers: The ubiquitous AIAnd more open Intel

"This is an exciting week. We have gathered the smartest AI big cattle at AI DevCon. We realize that the full promise of artificial intelligence is not something that Intel can do alone. Instead, he needs the whole industry to join together, including Developer community, academia, software ecosystem, etc.” Naveen Rao said in the article.

At the first Intel AI Conference, Netease Intelligence felt Intel's attention and enthusiasm for developers and partners. Naveen Rao further introduced Intel's tools for developers and new results with more partners.

Naveen Rao pointed out that we learned from a survey that more than 50% of U.S. companies are turning to Intel Xeon processor-based cloud solutions to meet their artificial intelligence needs. This certainly includes Intel Xeon processors. The Intel Nervana and Intel Movidius technologies and Intel FPGA approach meet the unique needs of artificial intelligence workloads.

As shown in the above picture, this is the OpenVINO toolkit that was presented to us by Naveen Rao at the conference. This kit was released by Intel in mid-May. It mainly integrates computer graphics and deep learning reasoning into leading-edge visual applications. Its full name is. For open visual reasoning and neural network optimization, developers can help create and train AI models in the cloud, such as TensorFlow, MXNet, and Caffe, and deploy them to various products.

OpenVINO may expand to various markets in the future, including enterprise, retail, energy and medical. Forbes magazine commented that Intel’s move will help unify its multiple products CPU, GPU, VPU (Movidius) and FPGA (Altera). To achieve rapid development of the visual processing market.

Next, Naveen Rao introduced us to another developer-oriented tool, BigDL, which is a distributed deep learning library for Apache Spark. With BigDL developers, deep learning applications can be written as Scala or Python programs. You can take advantage of the capabilities of a scalable Spark cluster.

Knowing that AlfredXXfiTTs users likened BigDL to the potential of becoming a "big killer" in many scenarios, including but not limited to: 1. Large-scale distributed clusters (such as: Hadoop clusters); Large-scale Inference, such as: recommendation system, search system, advertising system; 3, (upstream and downstream) rely on the Spark / Hadoop ecology; 4, mild deep learning users, such as: data development engineers / data mining engineers; Scala/JVM enthusiasts.

Of course, there is an open source nGraph compiler for developers. This is a deep neural network model compiler for various devices and frameworks. Data scientists can focus on data science and development without worrying about how to deploy DNN models to various Different equipment to do effective training and operation. According to NaveenRao, nGraph currently directly supports TensorFlow, MXNet, and neon, and can indirectly support CNTK, PyTorch, and Caffe2 via ONNX. Users can run these frameworks on different devices, including Intel Xeon, GPU, Nervana, NNP, etc. .

Talk about partners: Let AIEmpower and embrace more people

In addition to the love for developers, Intel also demonstrated a variety of scenarios for collaboration under its AI empowerment, including AI composing, AI rendering animations, and more.

The AI ​​composer used Intel’s previously launched Neuron Computor Stick. The full name is Movidius Neural Compute Stick (NCS), which is a neuron computing stick. In general, it can “add blood” to the deep learning of machines. Provide localized code to run the support of the product, the price of about 500 yuan RMB, can be said to buy a loss can not buy fooled.

From the live demonstration, the effect of AI composing is rather surprising. Today, AI started to engage in more artistic work, which may indicate that the curse that artificial intelligence cannot engage in artistic creation is being broken.

In addition, the conference also demonstrated the use of Intel AI for 3D animation rendering of lions, lifelike lions really appear true and false, this is the ZIVA company based on Intel Xeon (Xeon) research.

It is worth mentioning that Intel also announced a heavyweight partner this time, they will become the artificial intelligence (AI) cooperation platform for the 2020 Tokyo Olympic Games, provide artificial intelligence technology for the Olympic Games, and launched the "Intel artificial intelligence AI Contest", the winning developer bonus is 10,000.

Talk about strategy: build a comprehensive hardware and software ecology

We noticed that in the past few years, Intel has successively acquired many companies to follow the footsteps of artificial intelligence, including Nervana, Movidius, MobileEye, and Altera. In addition, according to media reports, Intel has invested 140 in China since 1998. A number of technology companies total more than $1.9 billion.

Today, various science and technology giants are sharpening their muscles, and Intel is also the time to show true technology. In Naveen Rao's view, Intel wants to create an artificial intelligence strategy that integrates software, hardware and ecology.

They want to use their strengths and capabilities as a starting point to empower the entire industry and allow more companies and developers to join in. Current partners include Google and AWS.MicrosoftNovartis, C3 IoT, Baidu, etc., Naveen Rao said that we need to provide comprehensive solutions and more supporters in accelerating the transition to artificial intelligence driven future computing.

Talk about thinking: at AITimes take the lead AIEthics have not yet been perfectly solved

"Every company in the future is an artificial intelligence company." With the evolution of artificial intelligence, this consensus is still fermenting, and more and more companies are self-sustaining with AI. But after delving into the industry, you can discover old chips. Manufacturers always play a decisive role. With the participation of chip makers such as Intel and Nvidia, the heat of artificial intelligence may usher in a reshuffle.

After all, snoring also has to use fists to talk about things. Intel’s potential will help them to take the initiative in the AI ​​era. They also hope that they will not only be themselves, but also call the entire United States to act.

This is their thinking on artificial intelligence strategies, and Intel has its own understanding of AI ethical issues.

Before many problems can be solved, we must perfect our technology as much as possible. Humans must learn to live with artificial intelligence.

One more thing

Intel's artificial intelligence is basically here, right from the second is the AI ​​steer Naveen Rao.

China IT News APP

Download China IT News APP

Please rate this news

The average score will be displayed after you score.

Post comment

Do not see clearly? Click for a new code.

User comments