Ali Launched The "FPGA Version Of Taobao", A New Generation Of AI Highlights.

Recently,

A new generation of heterogeneous acceleration platform for AI is launched, which comprises 6 heterogeneous instances, including AI acceleration cards, including GPU and FPGA.

The example of the AI GA1 computing heterogeneous acceleration platform launched by Ali cloud includes GA1, GN4, GN5, GN5i, FPGA F1 and FPGA F2, which are suitable for mass data computation such as machine learning, graphics and image and scientific computation.

First, what is the heterogeneous computing of artificial intelligence? When will it happen? What is the opportunity?

At present, the way to achieve the complete AI computation is the collaborative computing of different architectures of CPU+AI chips, i. e. heterogeneous computing.

At present, there are mainly 3 kinds of chips as AI chip: GPU (graphics processor), FPGA (field programmable logic array), ASIC (ASIC).

After 2000 years, the demand for high-performance computing (HPC), such as scientific computing, numerical analysis and financial analysis, spawned the heterogeneous computing of CPU+GPU.

Data + computing power + algorithm is gradually mature, AI is rising again, and CPU + GPU/FPGA/ASIC heterogeneous computing is developing rapidly.

Intel expects that in 2016 to 2020, the computing power that focuses on AI will increase by 12 times.

Heterogeneous computing, as the main way of AI computing, has great room for development.

Two, Ali cloud has released 4 GPU heterogeneous instances since the beginning of the year, specializing in AI computing and scientific computing.

GA1, GN4, GN5 and GN5i are all

The example (which can be understood as a specific cloud computing product) can be purchased by the user (rented the product) for deep learning and scientific computing.

GA1 is a flexible GPU visualization example of AMD GPU using ECS GPU, which was launched at the beginning of 2017, and focuses more on image processing and high performance computing (HPC).

GN4, GN5 and GN5i are NVIDIA GPU, which are suitable for AI computing and scientific calculation. GN4 and GN5 are all launched in the first half of the year, and GN5i is the latest release.

In the second half of the year, Ali cloud also plans to launch high-performance GPU heterogeneous instances equipped with the most powerful V100 (Volta architecture, NVIDIA GPU's latest architecture).

Three, Ali cloud FPGA example the biggest highlight: "FPGA version Taobao" launched

At the end of 2016, AWS announced the testing of FPGA heterogeneous instances for encryption, finance, machine learning and so on. Ali cloud followed closely, and in early 2017, FPGA instance F1 was launched.

FPGA instances usually require four processes, namely, mirroring, developing, verifying, and deploying.

Combined with the characteristics of FPGA hardware reprogramming, Ali cloud has launched the mirror market, and users can purchase or sell FPGA IP.

Ali cloud FPGA example saves costs, improves efficiency and promotes pactions. It is expected that more AI start-ups and universities will use Ali cloud to develop AI.

Four. Key targets

"AI chip +" industry chain investment opportunities, AI chip is the foundation, based on AI chip data, scene application prospects are very broad.

We select the stocks from the upstream and downstream of the AI chip industry chain from the aspects of technology, data and scenario application. The first is: Zhongke Shuguang, wave information and Zhongke Chuang da.

Focuses on: four dimensional graphics, Hanwang Technology, Jia Du science and technology, Jing Jiawei, HKUST, Yishui, Hikvision, Dahua shares, flush, Dongfang net power, and Si Hui medical benefits.

Five, risk warning

1., the first thing to close stocks is short-term gains, and there may be a demand for callbacks in the short term.

2.AI chip technology and application progress is less than expected, competition is increasing, gross profit margin slipped.

More interesting reports, please pay attention to the world clothing shoes and hats net.