Nvidia’s New Supercomputer Will Enable Level 5 Autonomous Driving




Nvidia Drive PX Pegasus (Image Credit: Nvidia)

Nvidia has finished development of a new supercomputer, which is capable of driving with no human intervention.

Nvidia is considered one of the world’s leaders in computers and artificial intelligence for autonomous driving. The company’s Drive PX 2 computing platform is capable of 30 trillion operations per second and is utilized in Tesla’s vehicles, which are supposed to eventually drive themselves. Though Nvidia and Tesla both claim that the Drive PX 2 computer is capable of Level 5 autonomy (cars without steering wheels, pedals, or mirrors), the tech supplier has upped its game considerably.

Nvidia just announced the Drive PX Pegasus at a GPU Technology Conference in Munich. It has the capability to process a whopping 320 trillion operations per second. This is more than ten times as powerful as the computer that the company said can already handle full self-driving capability. The chip giant’s director of automotive tech, Danny Shapiro, told reporters:

“After the first generation platform, we realized we needed a lot more performance. This is a true supercomputer for the car.”

The platform — which will go into production during the second half of 2018 — is the size of a license plate and can process huge amounts of data instantaneously through the use of lidar, radar, ultrasonic sensors, and cameras.

The chipmaker says it handles up to one terabyte of data per second, which is comparable to a 100-server data center. It will become the primary system that Nvidia recommends to companies developing self-driving vehicles. Once it becomes available, companies that are currently using Drive PX 2 will be able to upgrade to Pegasus. Shapiro continued:

 “This system is going to save thousands of watts off of what is in these cars today.”

The chipmaker also added that 225 companies are already using the Drive PX customizable platform. Of these, 25 are working on driverless taxis. Some suppliers and startups mentioned are NuTonomy, Zoox, and Optimus Ride. NuTonomy CEO, Karl Iagnemma, shared:

“NuTonomy is building for Level 5 and Pegasus is the kind of platform that will be required to support these types of systems.”

Source: Automotive News Europe, The Verge

Categories: General

Tags: , , ,

Leave a Reply

14 Comments on "Nvidia’s New Supercomputer Will Enable Level 5 Autonomous Driving"

newest oldest most voted

If the PX 2 can already support level 5 driving, of what benefit is the Pegasus? Energy savings? Or is the PX 2 just not up for the task, and this was realised retroactively?

It’s funny that no level 5 system yet exists, but everyone seems to “know” that hardware version X can do it.

Of your options, my guess is the PX isn’t up to the task and they realized retroactively.

The more companies work on autonomy, the more they’ll understand what they need for it to work. But it’s a learning process for everyone.

Funny how some people think.
Thank goodness they are not in the path of innovation.

Take a simple example, a monochromatic video display. It has X total pixels. Say I want decent performance, so it must be able to update all pixels simultaneously 30 times a second. So all I need is hardware (e.g. video display chip) that can do 30X operations per second.

So my hardware is ready. Now I’ll just have to wait for software.

Later on, I can announce hardware that can do 60X/s.

A more funny analogy is “we already have cars that can reach the speed limit, there’s no need for a car that can go faster”

“It’s funny that no level 5 system yet exists, but everyone seems to ‘know’ that hardware version X can do it.”

Yeah, unfortunately there seems to be a lot of that going around.

I note that Waymo, which has a semi-autonomous driving system arguably more advanced than anyone else’s, is not going around telling everyone that their prototypes have all the hardware they are ever gonna need.

All the more reason not to pony up for FSD in current Teslas…there is really no evidence the tech (it’s all about the software) is really ready.

Well, it’s certainly about the software rather than the computer it runs on, contrary to the claims in this article!

But it’s not all about the software. Tesla needs a set of sensors adequate for the task, but unfortunately they are stubbornly resisting using lidar or radar scanning in 360° around the car. Tesla is trying to “cheap out” by using cameras and optical object recognition as the basis for its SLAM* system.

Meanwhile, a new article reports GM has bought a lidar maker. GM, it seems, is moving in the right direction. In this one area it looks like Tesla is falling behind.

*Simultaneous Localization And Mapping. “SLAM is concerned with the problem of building a map of an unknown environment by a mobile robot while at the same time navigating the environment using the map.”

What, did you change your position and are now shorting Tesla you troll?

Obviously that is the only reason for you to say such things. Facts be damned!

“Nvidia has finished development of a new supercomputer, which is capable of driving with no human intervention.”

Gosh, someone is finally making one of those god-like computers that are nearly omniscient and far smarter than people; the kind we see so often in movies and used to see in TV shows back in the seventies!

It’s amazing how often the uninformed confuse computer hardware with software. The computer isn’t what will enable full Level 4 or Level 5 autonomy; it’s the software that runs on the computer that will, or won’t, enable that; software which almost certainly has not yet been developed.

I hope software takes a long time. We shouldn’t ruin cars any further, than we have by designing for what doesn’t exist (and the autonomy-only goal, that people don’t want).

10 years of Level 3 I’d be fine with.

I for one would appreciate lvl5. So would many others who can’t drive for one reason or another.

Yes Level 5 is almost ready. Now we have only one minor issue left, somebody needs to write functional software for it, and we are done! AI is easy-peasy after all.

I’m sure it will be finished in no time before the end of this month. I heard it was 99.9999% complete in December last year!

Badass piece of kit but it does put Tesla in a peculiar situation since they have been selling cars with what they claim is the hardware required for full auto driving. Apparently that is not so, which is also supported by the lack of the actual product from Tesla. In fact it seems that the autonomous system is going backwards every release according to some Youtube videos.

The question is how much power this thing draws. One high end NVidia GeForce GPU draws around 300 watt IIRC and this board seems to have at least 4 chips of a kind that possibly draws even more. Together with other components it could mean several thousand watts. I guess that solves the problem of cabin heating though 🙂

Tesla doesn’t need the fast computer because they’re not dealing with all that pesky LIDAR data. 😉

Google waymo said they are releasing it in 2017, level 5 right here in Arizona.