NVIDIA DGX Upgrades Continental's Autonomous Driving Development
The automotive industry is evolving, and development cycles are becoming shorter.
FREMONT, CA: Continental has invested in setting up its supercomputer for Artificial Intelligence (AI), powered by NVIDIA InfiniBand-connected DGX systems, to develop innovative technologies even more efficiently and quickly. It has been operating from a data center in Frankfurt am Main, Germany, since the beginning of 2020 and is offering computing power and storage to developers in locations around the world. AI improves advanced driver assistance systems, makes mobility efficient and safer, and accelerates autonomous driving development.
"The supercomputer is an investment in our future," states Christian Schumacher, head of Program Management Systems in Continental's Advanced Driver Assistance Systems business unit. "The state-of-the-art system reduces the time to train neural networks, as it allows for at least 14 times more experiments to be run at the same time."
Cooperation with NVIDIA secures top quality
"When searching for a partner, we look for two things: quality and speed," said Schumacher. "The project was set up with an ambitious timeline and implemented in less than a year. After intensive testing and scouting, Continental selected NVIDIA, which powers many of the fastest supercomputers around the world."
"NVIDIA DGX systems give innovators like Continental AI supercomputing in a cost-effective, enterprise-ready solution that's easy to deploy," said Manuvir Das, head of Enterprise Computing at NVIDIA. "Using the InfiniBand-connected NVIDIA DGX POD for autonomous vehicle training, Continental is engineering tomorrow's most intelligent vehicles, as well as the IT infrastructure that will be used to design them."
IT masterpiece for AI-based solutions
Continental's supercomputer is built with above 50 NVIDIA DGX systems, connected with the NVIDIA Mellanox InfiniBand network. It is ranked according to the openly available list of TOP500 supercomputers as the best system in the automotive sector. A hybrid approach has been selected to extend capacity and storage from cloud solutions if required. "The supercomputer is a masterpiece of IT infrastructure engineering," says Schumacher. "Every detail has been planned precisely by the team – in order to ensure the full performance and functionality today, with scalability for future extensions."
Advanced driver assistance systems utilize AI to make decisions, assist the driver, and finally operate autonomously. Environmental sensors such as radar and cameras deliver raw data. This raw data is being processed quickly by intelligent systems to develop a comprehensive model of the vehicle's surroundings and devise a strategy for interacting with the environment. Ultimately, the vehicle needs to be controlled to behave like planned. But with systems being more complicated, traditional software development methods and machine learning methods have reached their limit. Deep Learning and simulations have risen to be the fundamental methods in developing AI-based solutions.
Main use cases: Deep Learning, Simulation, and Virtual Data Generation
With Deep Learning, an artificial neural network allows the machine to learn by experience and connect the new information with existing knowledge, mostly imitating the human brain's learning process. But while a child is capable of remembering a car after being shown a few dozen pictures of different car kinds, several thousand hours of training with millions of images and thus enormous amounts of data are necessary to train a neural network, which will, in future, assist a driver or even control a vehicle autonomously. The NVIDIA DGX POD reduces the time needed for this complicated process, but it also reduces the time to market for the new technologies.
"Overall, we are estimating the time needed to fully train a neural network to be reduced from weeks to hours," says Balázs Lóránd, head of Continental's AI Competence Center in Budapest, Hungary, who also works on the development of infrastructure for AI-based innovations together with his groups in Continental. "Our development team has been growing in numbers and experience over the past years. With the supercomputer, we are now able to scale computing power even better according to our needs and leverage the full potential of our developers."
The data used for training the neural networks comes mainly from the Continental test vehicle fleet. They drive around 15,000 test kilometers per day, collecting around 100 terabytes of information – equivalent to 50,000 hours of movies. Already, the recorded data can be utilized to train new systems by replaying and thus simulating physical test drives. With the help of a supercomputer, data can now be generated synthetically. It is a high computing power-consuming use case that allows systems to learn from traveling virtually via a simulated environment.
This can have many advantages for the development process: Firstly, over the long run, it might make the recording, storing, and mining the data generated by the physical fleet unnecessary, as the necessary training scenarios can be created immediately on the system itself. Secondly, it improves speed, as virtual vehicles can travel a similar number of test kilometers in a few hours that would take a real car many weeks. Thirdly, the synthetic generation of data increases the possibility for systems to process and react to the changing and unpredictable situations. Ultimately, this will enable vehicles to navigate safely via changing and extreme weather conditions or make reliable forecasts of pedestrian movements, thus paving the way to higher automation levels.
The potential to scale was one of the main reasons behind the conception of the NVIDIA DGX POD. Through technology, machines can learn faster, better, and comprehensively than through any human-controlled method, with potential performance growing exponentially with every evolutionary step.
The supercomputer is located in a data center in Frankfurt that has been chosen for its proximity to the cloud providers and, more importantly, its AI-ready environment, completing the specific requirements regarding connectivity, cooling systems, and power supply. Certified green energy is being leveraged to power the computer, with GPU clusters being much more energy-efficient than CPU clusters by design.