Continental invests in building artificial intelligence supercomputer for autonomous driving

Continental has invested in building an AI supercomputer powered by NVIDIA InfiniBand-connected DGX systems to provide computing power and storage capabilities to developers around the world.

Continental's supercomputer consists of more than 50 NVIDIA DGX systems connected with an NVIDIA Mellanox InfiniBand network. According to the company, it is a top system for the automotive industry according to the publicly available Top 500 supercomputer rankings. A hybrid approach is used to expand capacity and storage through cloud solutions when needed.

Neural networks that can help drivers or even self-driving vehicles require thousands of hours of training, including millions of images and data. According to the company, NVIDIA DGX POD not only shortens the time required for this complex process, but also reduces the time to market for new technologies.

“Overall, we estimate that the time required to train a neural network has been reduced from weeks to hours,” says Balázs Lóránd, head of Continental’s AI Competence Center in Budapest, Hungary, who works with Continental on infrastructure development based on AI innovations.

To date, the data used to train these neural networks has come primarily from the Continental test fleet. They currently drive around 15,000 test kilometers per day and collect around 100 TB of data, the equivalent of 50,000 hours of film. The recorded data can already be used to train new systems by replaying them, thereby simulating physical test drives. With supercomputers, the data can now be generated synthetically.

This has several benefits for the development process: First, it potentially makes it unnecessary to record, store and mine the data generated by the fleet, as the necessary training scenarios can be created on the system itself. Second, it increases speed, as a virtual vehicle can drive the same number of test kilometers in a few hours that would take a real car several weeks. Third, the comprehensive generation of data enables the system to handle and respond to changing and unpredictable situations. Overall, this will enable vehicles to safely navigate changing and extreme weather conditions or make reliable predictions about pedestrian activity.