|Nvidia hopes autonomy will be widespread in the transportation industry soon / Photo Credit: JHVEPhoto (via Shutterstock)|
Nvidia encourages carmakers to implement its single unified architecture for everything, ranging from software and datacenter tech to chips for autonomous vehicles (AV), reported Dean Takahashi of Venture Beat, an American technology news website. Nvidia is also making its pre-trained libraries of deep neural networks for AVs open to its partners. The company hopes autonomy will be widespread across today’s $10 trillion transportation industry. But that will require more computing power to handle the rapid growth in the AI models being developed as well as to guarantee the functionality and safety of AVs. Nvidia’s Drive AV is an end-to-end, software-defined platform for smart vehicles.
The platform includes a “development flow, data center infrastructure, an in-vehicle computer, and pretrained AI models that can be adapted by carmakers.”
Last December, Nvidia debuted its massive AI chip called Drive AGX Orin, performing seven times better than its predecessor, Xavier. Danny Shapiro, Nvidia’s senior director of automotive, explained in a press briefing that the company’s graphics chips paved the way “to this kind of self-learning design” that is becoming the heart of AI chips for AVs. Shapiro added, “Everything in the $10 trillion industry will have some degree of autonomy, and that’s what we’re really working to develop.” We should recognize that it is a single, unified architecture to create software-defined vehicles, he stated.
Samples of Nvidia’s Orin will be given out in 2020 for testing. However, the chip will not be shipped in vehicles come 2022. Orin will be capable of 200 trillion operations per second as it is designed to handle several applications and deep neural networks (DNNs) that run simultaneously on AVs. The chip will also abide by systematic safety standards like ISO 26262 ASIL-D. Equipped with 17 billion transistors, Orin is the result of four years of intensive research.
Shapiro explained that customers begin working on a pre-trained model, taking their data and retraining them using Nvidia’s foundation. “And then, with these tools, they can optimize it, retrain it, test it, and create something that is efficient and customized for them,” he said.