Nvidia Unveils Supercomputer For Self-Driving Cars

Nvidia Corp. unveiled details about its new supercomputer for self-driving cars at the Consumer Electronics Show in Las Vegas. CEO Jen-Hsun Huang explained the company’s latest automotive system, the Drive PX 2, is about the size of a lunchbox and is powered by an array of Nvidia graphics chips that offer the computing power of approximately 150 MacBook Pro laptops.

The result is a computer that can work through 24 trillion deep learning operations per second. This means it can process up to 2,800 images per second using a neural network-based algorithm, which should provide the ability for a car to accurately orient itself in the world and determine how to handle road hazards, such as bad drivers, debris, and erratic cyclists. In other words, the system is designed to solve some of the most difficult problems that face autonomous vehicles.

It’s a “self-driving supercomputer that fits nicely in your trunk,” said Huang. He announced Volvo will be the first to use the new system, with plans to deploy 100 SUVs with the technology in Sweden in 2017.

However, Huang explained that hardware is just one of the elements that is required to allow self-driving cars to reliably sense their surroundings and make the correct driving decisions. He said that much of the most important progress has to do with software, using technology known as neural networks. As opposed to hand-coding rules to define objects that are surrounding a car, Huang said the latest autonomous driving technology is based on showing images to high-powered computers that have been programmed to learn from them. He explained that companies such as Microsoft and Google have demonstrated that their neural networks “beat the best human recognition in the world,” thereby achieving “superhuman” capabilities.

Additionally, Huang said that it will be equally important for automakers to establish their own big-data platforms that gather massive quantities of information about how their vehicles perform. Therefore, while a car could be quite capable as it leaves the factory, it will improve over time based on information from incidents that are encountered by other cars.

Huang believes that the benefits of realizing this vision will be enormous, with the potential to greatly increase road safety, save money, help those who are unable to drive, preserve the environment, and lead to changes in urban planning.

“Parks will replace parking lots,” said Huang.

Although Nvidia is a relative newcomer to auto chips, it is gaining ground quickly, with Ford, Audi and Daimler using its system to develop their self-driving cars. The company said that the new product will be widely available in the fourth quarter of 2016, with some samples shipping in the second period.