NVIDIA Unveils a 'Supercomputer' to Speed Development of Self-Driving Cars
Self-driving cars require an entire suite of robust hardware and software to safely navigate on the road. Technologies such as robotics, deep-learning, computer vision and AI are all utilized in the development of autonomous driving vehicles. Afterwards, all of this data needs to be fused together, analyzed and processed, which requires a high level of computing power.
Silicon Valley-based chipmaker NVIDIA just unveiled its new DGX "SuperPOD", which the company says is one of the world's fastest supercomputers (the 22nd fastest to be more precise) that includes AI infrastructure to meet the massive demands of the company's autonomous-vehicle deployment program.
NVIDIA said the system was built in just three weeks using 96 individual NVIDIA DGX-2H supercomputers and Mellanox interconnect technology. This setup delivers a mind-boggling 9.4 petaflops of processing power, which is enough computing power for training the vast number of deep neural networks required for safe self-driving vehicles. It is designed to run in a data center.
The DGX-2H server is basically an enhanced version of NVIDIA's DGX-2 featuring 16 Tesla V100 GPUs running at 450 watts per GPU and higher frequency CPUs built to deliver the highest performance. Each of the 96 DGX-2H servers used in the SuperPOD consume just 12 kW of power and delivers 2.1 PetaFLOPS of computing power.
Training Deep-Learning Models
Being able to process large volumes of data generated by a self-driving car is a challenge for developers. A single data-collection vehicle for example, generates about 1 terabyte of data each hour. Multiply that by years of driving over an entire fleet, and it can quickly add up to petabytes of data.
The huge volume of data collected by a self-driving car is used to train algorithms to drive more like a human (or better) and to find potential failures in the deep neural networks operating in the vehicle, which are then re-trained in a continuous loop. The SuperPOD is able to optimize autonomous driving software and retrain neural networks at a much faster turnaround time than previously possible.
Deep-learning plays a vital role for a self-driving vehicle, enabling the software to get increasingly smarter based on experience.
"AI leadership demands leadership in compute infrastructure," said Clement Farabet, vice president of AI infrastructure at NVIDIA. "Few AI challenges are as demanding as training autonomous vehicles, which requires retraining neural networks tens of thousands of times to meet extreme accuracy needs. There's no substitute for massive processing capability like that of the DGX SuperPOD."
NVIDIA's says that the DGX SuperPOD hardware and software platform takes less than two minutes to train the convolutional neural network ResNet-50. ResNet-50 is most commonly used for image classification.
When this AI model came out in 2015, NVIDIA said it took 25 days to train on the then state-of-the-art system, a single NVIDIA K80 GPU accelerator. However, the DGX SuperPOD can do the same task 18,000 times faster.
A NVIDIA self-driving test vehicle in front of the company's Silicon Valley headquarters.
NVIDIA has been at the forefront in developing the hardware robust enough to run AI for autonomous driving for the part several years, mainly with its DRIVE PX line of processors.
In testimony before a packed hearing of the U.S. Senate Committee on Commerce, Science and Transportation in Washington D.C. in June 2017, Rob Csongor, NVIDIA's vice president of Autonomous Machines said that the use of artificial intelligence in the transportation industry will enable self-driving cars that save tens of thousands of lives, provide mobility to the disabled, improve urban design and save vast amounts of unproductive hours.
NVIDIA's technology is being used by more than 225 automotive companies worldwide, including Audi, Tesla, Toyota, Volvo, Mercedes and many other automakers and startups developing autonomous driving technology.
Building supercomputers like the DGX SuperPOD has helped NVIDIA learn how to design its systems for large-scale AI machines. The supercomputing technology gives developers of autonomous driving technology access to high performance computing to rapidly accelerate their initiatives.
- Electric Truck Startup Rivian Announces a New $2.5 Billion Investment Round Led By T. Rowe Price
- Self-Driving Startup Aurora Debuts its New Long-Range Lidar Called 'FirstLight'
- Lucid Motors Plans to Open 20 Design Studios & Service Locations Throughout the U.S.
- Ford Motor Co is Targeting the Jeep Crowd with its New Bronco Adventure Vehicle Lineup
- Uber Buys Rival Postmates for $2.65 Billion as its Core Ride-Hailing Business Stumbles
- Lidar Pioneer Velodyne to Become a Public Company in a Reverse Merger Deal
- Toyota Launches its Vehicle Throttle Suppression System to Prevent Unintended Acceleration
- Chinese EV Startup BYTON to Suspend Operations for 6 Months, the Company’s Future is Uncertain