NVIDIA is showing a smart vehicle the size of a remote-controlled car called Project Dave this week at GTC 2015. This is a DARPA project which runs NVIDIA DRIVE PX – a smart system made to allow this vehicle to navigate on its own. Deep neural net in action. We first learned about NVIDIA DRIVE PX back at CES 2015 where it was just a mobile supercomputer. In Dave, 3.1 million connections are made, video is processed at 12 frames per second, and 38 million connections are made per second.
Below you’ll see how AlexNet – an imaging engine – is moving quicker than this real-life demonstration of a remote-controlled (self-controlled) car called Dave. This system has 630-million connections, accessed at 184 frames per second, made at 116 billion connections per second. This brain is firing at 116 billion times a second.
This is a system that can learn. This is a system that’s able to sense with FPGA CV ASIC, through DNN, make a plan with the CPU, and act. When it sees an obstacle, it acts. It avoids, it moves, it keeps the driver (or just the vehicle, when there are no passengers), safe.
The NVIDIA DRIVE PX self-driving car computer will be made available in May of 2015 for a cool $10,000 USD. Stay tuned as we learn more through our GTC 2015 tag portal.