Tesla Autopilot “five years ahead of time” – Nvidia

Tesla’s Autopilot autonomous driving capability has been both endorsed and doubted, while a high-profile case found a Model S to be operating in Autopilot mode when it ended in a fatal crash. Now, the American electric automaker has another public fan in Nvidia CEO Jen-Hsun Huang, who said that Tesla’s Autopilot is “basically five years ahead” of other autonomous systems.

“Anybody who’s talking about 2021… that’s just a non-starter anymore. And I think that’s probably the most significant bit in the automotive industry. Anybody who is talking about autonomous capabilities in 2020 and 2021 is at the moment re-evaluating in a very significant way,” Huang said.

Tesla Autopilot-01

Autonomous driving is not a “detection problem” but rather, an “AI problem,” Huang added, which can be solved next year with the use of Nvidia’s Drive PX 2 supercomputer, which he claimed in the video above to have “the computational capability of 150 MacBook Pros,” within a package the size of a lunch box.

When asked why Tesla selected Nvidia’s technology for the second generation of Autopilot instead of Mobileye’s, Huang stated three reasons: Nvidia’s GPU platform is better suited for artificial intelligence applications, it is scalable with over-the-air updates for the building of an autonomous fleet, and it is better at energy efficiency.

Looking to sell your car? Sell it with myTukar.

10% discount when you renew your car insurance

Compare prices between different insurer providers and use the promo code 'PAULTAN10' when you make your payment to save the most on your car insurance renewal compared to other competing services.

Car Insurance

Mick Chan

Open roads and closed circuits hold great allure for Mick Chan. Driving heaven to him is exercising a playful chassis on twisty paths; prizes ergonomics and involvement over gadgetry. Spent three years at a motoring newspaper and short stint with a magazine prior to joining this website.



  • 4G63T DSM on Nov 14, 2016 at 10:18 am

    Thats precisely what Musk is doing. Getting this pet project on the roads early to “collect data”.

    There is only so much the sensors can collect and so much time it has from detection to action.

    A good Photographer can anticipate a shot. Likewise a good autonomous driver needs to be able to anticipate events like or better than a human driver. It is called intuition (or in computer sense, AI)

    Like or Dislike: Thumb up 2 Thumb down 2
  • They need to spend 5 years in KL for the AI to learn the habits of Mat Rempits splitting lanes and crazy people suddenly deciding to U-turn across 3 lanes of traffic (because the road design here is confusing)

    Like or Dislike: Thumb up 20 Thumb down 0
    • Ah Kow on Nov 14, 2016 at 11:56 am

      I think 5 years not enough for Malaysia,I think for Malaysia need 10 years or more. Due to the crazy barbaric drivers and rempits.
      And an extra roof sensor too for any unforeseen CRANES falling on the road

      Like or Dislike: Thumb up 16 Thumb down 0
      • Kunta kinte on Nov 16, 2016 at 12:18 pm

        if you think KL traffic is bad?, then they need to spent another 50 years for jakarta traffic, 46 years for vietnam traffic, 35 years for bangkok traffic, 120 years for bombay traffic, need i to mention shanghai?, shenzen?, manila?..

        go visit to other country will open your eyes more.

        Like or Dislike: Thumb up 1 Thumb down 0
  • hayenadeblue on Nov 14, 2016 at 10:28 am

    since when GPU is better at energy efficiency? parallel execution, maybe. we will see the implementation of GPU on satellite IF GPU managed to achieve energy/power efficiency better than FPGA.

    Like or Dislike: Thumb up 0 Thumb down 4
    • Ultra2Disk on Nov 14, 2016 at 10:41 am

      Well Nvidia never compare them self with FPGA. But compare to current GPU exist. Nvidia indeed become much more efficient then ever.

      Like or Dislike: Thumb up 3 Thumb down 0
    • Hmm... Much Hmm... on Nov 14, 2016 at 3:16 pm

      I know that GPU’s architecture is good at SIMD but testing it by controlling a vehicle in the real world seems a bit unwise. It’s like, GPU is for less computing more action, in an event where an instruction is unavailable and the data is vaguely similar to something else, then what would the car do?

      Eg would be; A toddler crawl across the road, the car is going too fast, so there are 2 option, kill the elderly driver or kill the toddler.

      GPU Sees; An object blocking the road, looks to be a stray cat, knock it over and save the elderly driver who’ll die from the heart attack anyway.

      CPU sees; An object is blocking the road, it is a human baby crawling on the road, baby has longer lifespan than this elderly driver, baby could be the person that saves the planet, elderly driver could die by year end, baby could end world hunger, elderly driver is always hungry for McChicken, Hmm… Give baby a chance and kiss gramps good way.

      Perhaps they realized that by the time the CPU manage to figure things out the baby is dead and the car is 10km away from the accident spot with a dead gramps that had an heart attack so they just go with the GPU then.

      Like or Dislike: Thumb up 6 Thumb down 3
    • JustAnotherEngineer on Nov 16, 2016 at 12:52 pm

      Satellite chose FPGA for the future design rewiring option. Power is secondary, as long as it’s not too bad.

      Like or Dislike: Thumb up 0 Thumb down 0
  • drumpf on Nov 14, 2016 at 4:34 pm

    red team goes Ayyyyyy…

    Like or Dislike: Thumb up 1 Thumb down 2

Add a comment