Jinpin KG4224-G8 Pushes the Implementation of Autonomous Driving Technology

Background

A map company in Beijing is one of the top three map vendors in China and holds 25% of the market share in the front-loading car navigation market. To adapt to the market development trend, the company is transforming from a traditional map data provider to a location service platform provider. The company has Grade A surveying and mapping qualifications in navigation electronic map production, Internet map services, geographic information system engineering, and has obtained international quality management system ISO9001 and international automotive industry quality management system (IATF 16949) certification. The company's navigation map products have covered mainland China, Hong Kong, and Macau, with a road mileage of more than 6 million kilometers and more than 26 million POIs. The company is a neutral map vendor and has established strategic cooperation relationships with many well-known enterprises in the ecological chain such as ESRI, Yachuang, Mapbox, Sogou, and Neusoft, adhering to the concept of "cooperation, openness, and sharing", and is committed to providing high-quality solutions for customers in different fields.

Challenge

With the continuous development and progress of technology, autonomous driving technology has become the development direction of AI companies and traditional automobile manufacturing companies. After years of experience in the map and navigation field, the map company in Beijing keenly perceives the huge market scale and value of autonomous driving in the future, and thus vigorously develops autonomous driving technology and high-precision map field. With years of deep cultivation and accumulation in the map and navigation field, the development of autonomous driving technology has unique advantages because it has one of the core elements of autonomous driving, "high-precision maps", but the training of another element, "deep neural network", relies on large-scale computers. At present, the training is mainly based on CPU clusters, and one iteration takes 2-3 weeks, which seriously affects work efficiency and lags behind the product launch time. Insufficient computing power will become the bottleneck of the current situation.

Solution

Based on the user's pain point of insufficient computing power and a deep understanding of the application, Jinpin recommended that the customer use the Jinpin KG4224-G8 AI Server. This system is specially developed and customized for artificial intelligence and deep learning, and is a powerful tool for "deep neural network training" in autonomous driving. It supports 2 Intel Xeon E5-2600v4 processors and 8 Nvidia Tesla series GPU accelerators, with an output of 80 billion deep neural network training operations per second, equivalent to the computing power of about 200 CPU servers working simultaneously, which reduces the original calculation time from 2-3 weeks to within 2-3 hours.

Result

The development and implementation of autonomous driving technology cannot be separated from deep neural networks and high-precision maps, and it also cannot be separated from a supercomputer with supercomputing power. CPU-based computers cannot meet the requirements of autonomous driving and deep learning for computing power due to design architecture and core limitations. The Goldpeen KG4224-G8 AI Server was born out of the need for powerful computing power. The KG4224-G8 adopts a CPU+GPU coprocessor architecture (CPPC), which has tens of thousands of stream processors and Tensor cores, bringing rapid growth to the training of deep neural network models and pushing the implementation of autonomous driving technology.