Nvidia CEO Jensen Huang speaks during a press conference at The MGM during CES 2018 in Las Vegas on … [+]
It’s no secret that automated vehicles require an enormous amount of compute performance. It seems that closer engineers get to creating vehicles that can reliably get around without a human driver, the more performance they are finding they need. When it comes to delivering that raw computational power, Nvidia is widely seen as the leading vendor of silicon and it’s next-generation chip, called Orin takes that performance to new heights.
Nvidia CEO Jensen Huang announced Orin during one of his marathon keynotes at the company’s GPU Technology Conference in China today. Orin is the successor to the Xavier chip that has been widely used in AV development over the past two years and is coming to production applications in 2020.
The new system-on-a-chip (SoC) is one of the most complex ever produced with 17 billion transistors. As with Xavier and earlier Nvidia SoCs for automated driving, Orin combines an ARM CPU core with graphics processing (GPU) cores and multiple other elements needed for input-output of data at the required speeds on a single silicon die.
Orin is the first Nvidia SoC to use the new ARM Hercules architecture and it is paired with Nvidia’s next-generation GPU technology. With potentially dozens of sensors on and AV and multiple terabytes per hour of raw data to process, I/O performance can easily be a bottleneck in these applications. According to Nvidia, the Orin chip can handle over 200 GB/s of data. By way of comparison, the Tesla Autopilot V2.5 computer that also used older generations of Nvidia chips was limited to about 2 GB/s.
Orin will be available in multiple configurations with different speeds and Nvidia promises the top versions will be able to process about 200 trillion operations per second (TOPS). The Xavier SoC can handle about 30 TOPs and the new Tesla-designed chips for the Autopilot V3 computer can do about 36 TOPS each (Tesla uses two of these).
Nearly as important as raw power is the power consumption for the compute platform powering an AV. Since many of these vehicles will be electric and all of those sensors require power as well, performance efficiency is crucial to minimize the impact on driving range. The Xavier SoC delivers its 30 TOPS at about 30W of power consumption. The two chips used by Tesla consume about 72W. According to Nvidia’s senior director of automotive, Danny Shapiro, the Orin SoC should have about four times the performance per watt of Xavier or somewhere between 60 and 70W for 200 TOPS.
The Orin SoC should be in volume production in about two years and samples for testing should be available sometime before that. As with other Nvidia chips the software developed to run on earlier silicon should port directly over to the Orin platform and customers will have the option to use multiple chips if needed which would easily push the performance past the current Drive AGX Pegasus development platform which uses two Xaviers and two Turing GPUs. That should allow Orin platforms to run everything from advanced level 2 partially automated systems to level 4 or 5 highly automated systems.
