Creating a new height of AI: Zhaoxin dual-core 64-core domestic x86 chip released

According to news on May 14, Zhaoxin and Computing Energy have joined forces with the SOPHON.TEAM ecosystem to build a set of integrated AI training and reasoning solutions that can be implemented efficiently based on their own processors. It can solve the problems of high labor cost, high technical difficulty, long development cycle, and poor iteration ability for enterprises to deploy deep learning applications.

The hardware layer includes Zhaoxin Kaisheng KH-40000 series server processors with x86 architecture, as well as self-developed computing power products such as TPU processors, inference microservers, and smart computing cards.

The software layer deploys a training platform and an inference platform, integrating algorithm training and inference deployment capabilities.

The data layer performs data training and reasoning through operations such as video collection, data cleaning, and data annotation.

Users can independently produce and deploy the required algorithms according to actual needs, and automatically iterate the accuracy according to business scenario requirements, which is significantly better than the traditional solution of binding hardware and fixed algorithms.

Related Products--

1. Computing smart card SC7: A computing card for deep learning inference, it has completed the adaptation certification with Zhaoxin Kaisheng KH-40000 series server processor, and can efficiently adapt to the mainstream deep learning algorithms on the market. Applications like video structuring, face identification, behavior analysis, and status monitoring can be useful in smart cities, smart transportation, smart energy, smart finance, and other fields.

2. Lenovo Kaitian KR722z G2 server: equipped with dual-channel Kaisheng KH-40000/32 processor, a total of 64 cores, supporting 12 front 3.5-inch or 24 2.5-inch hot-swappable hard drives, and 4 rear hot-swappable hard drives Pluggable hard drive, memory capacity supports up to 2TB. The whole machine adopts redundant heat dissipation function and optional power supply function design, which not only provides powerful computing power, but also ensures the continuous and stable operation of the business.

3. Lianhe Donghai XRS 302 series server: equipped with dual-channel Kaisheng KH-40000/32 processor, 32 memory slots supporting a maximum of 2TB, and multiple specifications of PCIe interfaces to support hardware expansion such as GPU computing cards. Suitable for OA/mail services, database services, storage services, cloud platform computing nodes, AI computing services, inference services and other application scenarios.

This solution is oriented to real business scenarios and is divided into hardware layer, software layer and data layer.

Through the intelligent monitoring system, the operation data of the infrastructure is collected, and the AutoML automated training platform is used to analyze and model the infrastructure in scenarios such as computer rooms and base stations, thereby realizing the prediction and early warning of faults and safety hazards, and further predicting occurrences in advance. Determine and eliminate the possibility of faults and risk events.

Application scenarios--

1. Construct a modern three-dimensional video image inspection system: adopt a data enhancement strategy to generate targeted deep learning algorithms such as moving small target recognition, multi-frame filtering, and multiple detections to make up for the problem of too small moving targets and detection problems from the drone's perspective. It has shortcomings such as high difficulty, and accurate identification can quickly and accurately identify specific targets in the inspection area.

2. Improve the intelligent management of network infrastructure and fault risk early warning capabilities.

In addition, the AI education solution based on Zhaoxin processor platform has been successfully deployed and launched in the AI laboratory of the Engineering Training Center of South China University of Technology. This program further improves the school's AI teaching model and can help students simulate and explore the application of AI in real scenarios, and practice machine vision, natural language processing, deep learning, etc.

This solution is based on the AutoML zero-code automated training platform, which can help users independently complete the entire process of model creation, data annotation, model training, model testing, and model deployment through simple operations to achieve fast and efficient deep learning engineering. Landed.