Nvidia AI and multi-petaOps chips for class 5 automated cars within 4 years

Nvidia Corp chief executive Jensen Huang said on Thursday artificial intelligence would enable fully automated cars within 4 years, but sought to tamp down expectations for a surge in demand for its chips from cryptocurrency miners.

“It will take no more than 4 years to have fully autonomous cars on the road. How long it takes for the vast majority of cars on the road to become that, it really just depends,” Huang told media after a company event in Taipei.

Huang joined peers taming expectations of strong revenue growth from a wave of interest in cryptocurrencies. Advanced Micro Devices Inc expected this week that there will be some leveling off of cryptocurrency demand.

“Revenue for us in crypto is over $100 million a quarter. For us, it’s a small percentage… It’s obviously not a target market,” Huang said.

Nvidia is expected to hold a 66% market share in the global AI market by 2021. In 2018, Nivdia Drive Pegasus PX2 platform will be the first AI “brain” capable of full autonomy.

The Nvidia Drive PX 2 is based on one or two Tegra X2 SoCs where each SoC contains 2 Denver cores, 4 ARM A57 cores and a GPU from the Pascal generation. There are two real world board configurations:
for AutoCruise: 1x Tegra X2
for AutoChauffeur: 2x Tegra X2 + 2 Pascal GPU’s

There is further the proposal from Nvidia for fully autonomous driving by means of combining multiple items of the AutoChauffeur board variant and connecting these boards using e.g. UART, CAN, LIN, FlexRay, USB, 1 Gbit Ethernet or 10 Gbit Ethernet. For any derived custom PCB design the option of linking the Tegra X2 Processors via some PCIe bus bridge is further available, according to board block diagrams that can be found on the web.

All Tesla Motors vehicles manufactured from mid-October 2016 include a Drive PX 2, which will be used for neural net processing to enable Enhanced Autopilot and full self-driving functionality. Other applications are Roborace. Disassembling the Nvidia based control unit from a recent Tesla car exposed the chip markings that hinted expert to a GP106 for the MXM-GPU module and gave them strong hints for the Tegra X2 Parker as the CPU.

Nvidia announced the Xavier AI Car Supercomputer at CES 2017. In 2017 the performance of the Xavier-based system will be 50% greater than Drive PX 2 Autochauffeur system.

In October 2017 Nvidia and partner development companies announced the Drive PX Pegasus system, based upon two Xavier CPU/GPU devices and two post-Volta generation GPUs. The companies stated the third generation Drive PX system would be capable of Level 5 autonomous driving, with a total of 320 TOPS of AI computational power and a 500 Watts TDP.

A few weeks ago NVIDIA revealed the first computer chips for developing fully autonomous vehicles and said it had more than 25 customers working to build a new class of driverless cars, robotaxis and long-haul trucks.

Deutsche Post DHL Group, the world’s largest mail and logistics company, and ZF [ZFF.UL], a top automotive parts supplier, plan to deploy a fleet of autonomous delivery trucks based on the new chips, starting in 2019, NVIDIA said.

The third generation of NVIDIA’s Drive PX automotive line, code-named Pegasus, is a multi-chip platform the size of car license plates with datacenter-class processing power.

Pegasus can handle 320 trillion operations per second, representing roughly a 13-fold increase over the calculating power of the current PX 2 line.

A single NVIDIA Xavier-class processor can be used for level 3 semi-autonomous driving, while a combination of multiple mobile and graphics processors would run level 5 full-scale driverless cars, the company said.

A level 5 vehicle is capable of navigating roads without any driver input and in its purest form would have no steering wheel or brakes. A level 3 car still needs a steering wheel and a driver who can take over if the car encounters a problem, while level 4 promises driverless features in dedicated lanes.

This dramatic improvement is a pre-condition for developing and testing future autonomous cars.

The Pegasus line will be available by the middle of 2018 for automakers to begin developing vehicles and testing software algorithms needed to control future driverless cars, NVIDIA executives told a developers’ conference in Munich on Tuesday.

The deal between Deutsche Post, ZF and NVIDIA will include future Deutsche Post StreetScooter delivery trucks. In Munich, the three partners are showcasing a prototype StreetScooter running NVIDIA Drive PX chips used to control sensors including six cameras, one radar and one lidar, or 3D laser camera.

Initial use cases will be for logistics vehicles on private roads within freight centers or for long-haul trucking in dedicated lanes, Shapiro said: “They are not replacing the drivers, but making the drivers more efficient and safer”.

For its current generation Drive PX2, NVIDIA has said it has 225 customers, including car and truck makers, Tier 1 auto suppliers, high-definition mapping companies, start-ups and research institutions. These customers can make use of PX2-class software when they upgrade to Pegasus chips, NVIDIA said.

Making the new Pegasus platform even more attractive to automakers is its size. Current Level 5 prototype cars require a huge amount of space dedicated to the onboard computers needed to power the system. That’s one of the reasons why minivans and SUVs make such attractive test vehicles. But NVDA says the Drive PX Pegasus delivers the performance of a 100-server data center in a form factor “roughly the size of a licence plate.”

Nvidia says its new system will also “drastically reduce energy consumption and cost.” That is especially important for Tesla and other electric car makers, who need to squeeze as much range as possible out of their batteries.

With a potential for release around the middle of 2018, the Drive PX Pegasus, and next-generation GPUs within it, confirm Nvidia are continuing to push AI and autonomous driving applications above and beyond the current generation consumer graphics cards by at least a few years

10 thoughts on “Nvidia AI and multi-petaOps chips for class 5 automated cars within 4 years”

  1. Good mention of the hardware, but no specifics on the software. With the Mobileye acquisition, my guess would be Intel. Everybody who’s anybody is using it for resims. The electronics in the car is just one tiny sliver of the ecosystem required to develop a self driving vehicle. You have to have a sensor test bed, or better yet a fleet of them. You have to put miles on the road. You have ingest 10s of TB of data from each test vehicle every day into your data lake. You have to analyze the data and identify events. You have to make revisions to your firmware… and then… the Mother of All Road Trips (replay the last 150K miles of data from the test fleet through your testbed with the firmware update.

    Long story short… It ain’t easy.

  2. Just because there will exist computer systems suitable for level 5 self-driving automobiles doesn’t that the software will be there to actually deliver on this. I’m pretty sure that the test vehicles now operating can be configured to run systems with similar performance, albeit they’d be bigger, use more power and cost much more. What this means is that there should already be experimental level 4 and level 5 cars being tested now, which does not seem to be the case. I’m pretty sure we’ll get to level five, but just not this soon.

  3. Self-driving cars is great.
    The only problem for that economy is that self-driving VTOL aircraft is coming not long thereafter.
    Between traveling at 100km/h along a winding road and at 300km/h in a straight line I know which one I will choose.
    I know that Elon has been negative on flying cars. I think he is wrong.

    • You are assuming these autonomous flying vehicles will be affordable to both own and operate. As things stand now neither seems likely.

      • Not sure why they would be prohibitively more expensive than the equivalent automobile.

        Sure, you might need some more fault-tolerant and redundant technology (spare flight computers, redundant propulsion systems etc).

        But they’d also be a lot more efficient in racking up the passenger miles (i.e. revenue over time).

        I think ownership in the current sense of owning a personal car will go largely extinct. So the fact that flying cars might be a bit more expensive to purchase than the rolling equivalent vehicle won’t matter for the end user.

    • Petaflops – usually refers to 64 bit computing – used by Top500 ranking of linpack benchmark
      PetaOps -usually refers to anything from 8 bit to mixed 16/32 bit
      I wish more vendors would use graph 500 or HPGC as benchmark as this provides (IMHO) a better benchmark for real world processing of MOST tasks and, perhaps, closer to how the brain processes information.
      see https://aiimpacts.org/tepsbrainestimate/

Comments are closed.