The new Drive PX2 is an enhanced version of the last year’s Drive CX model. It is a more powerful computing platform which is able to process data from 12 video cameras and other sensors in real time in order to make more accurate driving decisions. The chipset is going to be implemented for self driving cars and it has the size of a lunchbox.
Drive PX2 features 12 different cores representing 8 teraflops of calculation power, 24 deep learning tera operations per second and it is using 16nm architecture. However, the chipset requires 250W for powering up and it is using a liquid cooling system. As it is going to be used in a car therefore, it is not going to be an issue. The automakers will just need to plug the device into their existing cooling solutions.
Jen-Hsun Huang, Nvidia co-founder and CEO said that PX2 is as powerful as 150 13-inch MacBook Pros. For the GPU power, MacBook Pro is using Intel Iris Graphics 6100 chip while the PX2 is using two next-gen Tegra processors and Pascal-based GPU. Therefore, PX2 is able to support up to 8 teraflops and able to recognize up to 2800 images per second using AlexNet neural network-based deep learning algorithm.
Nvidia also has their network based deep learning algorithm named Nvidia Drivenet. It has 37 million neurons and able to recognize up to 120 million objects so far. The good news is that the number will increase over time. However, Nvidia also open to other automakers’ own neural network. To test this chipset, Volvo will be the first partner shipping Drive PX2 in about 100 upcoming test cars. On top of that, Nvidia also partnered with Audi, Daimler, BMW and Ford to test the chipset.