Nvidia's latest AI supercomputer aims to put self-driving robotaxis on the road
Nvidia has announced a new AI supercomputer with enough processing power for fully autonomous (Level 5) driving, which means it's everything a car maker needs to power a driverless vehicle all by itself. The Drive PX Pegasus packs in four AI processors in total, yet only takes up a space the size of a licence plate.
The computer is the successor to the Drive PX 2 we saw last year, now with more than 10 times the performance and capable of running up to 320 trillion operations per second – very helpful for keeping an eye on everything happening on the roads.
What's more, it's enough to run a robotaxi, one that could turn up at your door and whisk you to any destination. The Level 5 capabilities of the computer marks the top rung of the self-driving ladder, from the fully human Level 0 to the fully software-controlled Level 5. At Level 5, even the steering wheel is optional.
Honda is planning to have Level 4 vehicles ready to go by 2025, while some car makers are even more ambitious. Elon Musk reckons Tesla's fleet of electric vehicles could be Level 4-ready within the next couple of years. Level 4 means a car can do all of the driving, but not necessarily in all conditions. Level 5 is then the next step up, where the vehicle drives along on its own, so you can have a nap or watch a bit of Netflix along the way.
"Millions of hours of lost time will be recaptured by drivers as they work, play, eat or sleep on their daily commutes," promises Nvidia. "And countless lives will be saved by vehicles that are never fatigued, impaired or distracted – increasing road safety, reducing congestion and freeing up valuable land currently used for parking lots."
Right now, getting a car up to Level 5 requires a lot of battery power and a lot of on-board electronics, including a mini data center sitting in the trunk. The Drive PX Pegasus promises to cut down on that bulk and energy use, accelerating the development of these fully autonomous cars and making them easier for manufacturers to build.
In other words a computing system that currently fills up the trunk of a car can now be matched by something the size of a standard desktop PC motherboard.
On board there are two of Nvidia's next-generation Xavier system-on-a-chip models and two discrete GPU chips to handle the computer vision and deep learning processing required to evaluate scenes and work out a reaction in a split second.
The benefit of deploying GPUs or Graphics Processing Units – similar to those used to display video games in a PC, only much more powerful – is that their architecture is very well suited to the demands of artificial intelligence, and able to process information from a lot of different inputs and data sources at once.
Nvidia says more than 25 companies are already working on embedding the new supercomputer into their vehicles, and the hardware is going to be more widely available to Nvidia's partners in the second half of 2018.
All of which means when a self-driving taxi cab finally rolls up at your door, it could well be powered by an Nvidia Drive PX Pegasus, or one of its successors.