Avoiding obstacles, identifying objects and changing lanes for the autonomous cars, they are going to need more power. More power probably from graphics processor units. Nvidia has designed a couple of new GPUs based on the Pascal architecture targeted at supercomputers and servers of the autonomous cars. The Tesla GPUs are been tweaked for deep-learning that aid in classification and correlation of data which are previously targeted only at supercomputers.
A class of algorithmic techniques based on highly connected neural networks is known as deep learning. The data generated from sensors and input devices and transmitted to the cloud they get processed through deep-learning systems for context, insights, and answers. For instance, deep-learning systems developed by Google and Facebook are intended for natural language processing and image recognition respectively. The Baidu’s speech recognition system, Deep-Speech-2 was built around Tesla GPUs claimed Nvidia.
With 12 teraflops of single-precision performance, 24GB of GDDR5 memory, 3,840 CUDA cores and draws 250 watts of power the new Tesla P40 has the powerful enough to be regular GPUs. Delivering 5.5 teraflops of single-precision performance, up to 75 watts of power, 2,560 cores and 8GB of GDDR5 memory the P4 is also not very far behind. But both will have different uses as well.