2023 Author: Bryan Walter | [email protected]. Last modified: 2023-05-21 22:24
NVIDIA presented a specialized Xavier computer designed for machine learning and control of self-driving cars at the international exhibition of electronics CES 2017. Also, representatives of the company demonstrated the movement of an unmanned vehicle driven by Xavier, Engadget reported.
With the development of self-driving car technology in the past few years, there has been a high demand for specialized equipment that is responsible for processing data and directly driving a car, so some manufacturers, including NVIDIA, are developing specialized hardware platforms. The concept of a system-on-a-chip called Xavier was announced back in the fall of 2016, but NVIDIA representatives spoke about the real capabilities of the platform just now, at CES 2017.
The system-on-a-chip consists of seven billion transistors, 512 CUDA Volta cores and a proprietary eight-core processor. Xavier performance is measured by the manufacturer in special units of TOPS DL (trillions of deep learning operations per second) and it is 30 TOPS DL - for comparison, the previous generation of specialized computers NVIDIA Drive PX 2 claimed performance was "only" 20 DL TOPS.
In addition to the Xavier presentation, NVIDIA representatives also demonstrated the computer at work - the unmanned prototype BB8, built on the basis of Lincoln and controlled by the new computer, drove almost the entire route, including passing the section along the expressway. Previously, this prototype has already demonstrated the ability to drive on dirt roads, move on roads with no markings, take sharp and blind turns, and also avoid obstacles on the roadway.
Xavier is not the first dedicated computer from NVIDIA to be positioned as a platform for self-driving cars. At CES 2016, the company unveiled the Drive PX 2, which has eight teraflops of computing power and is capable of processing data from 12 cameras, lidar, radars and other sensors. The Drive PX 2 was reportedly used in Roborace self-driving cars during the 2016-17 Formula E season.
Earlier, the American company Movidius, specializing in the development of software and hardware in the field of machine learning, presented a "neural network flash drive" - a portable computing module for neural networks with an interface for connecting via a standard USB port.