Tesla AI Chip Claimed to be Twenty Times Faster than Nvidia

Elon Musk indicates that a new Tesla AI chip is twenty times faster than Nvidia GPUs for AI. Tesla has a patent for an accelerated mathematical engine for performing complex convolution operations based on matrix multiply operations.

Elon Musk indicates data and AI software are the key to creating super safe self-driving cars. Tesla has 100 times more data on self-driving cars from about 1 million electric cars.

Elon Musk believes that LIDAR with simpler logic can only achieve a limited higher amount of safety.

Tesla’s Pete Bannon said Tesla is aiming to start production of the Autopilot Hardware 3.0 around the end of the first quarter of 2019. The activation of self-driving hardware will be dependent on the regulators for each country or region.

SOURCES- Elon Musk, Ark Invest Podcast, US Patents

Written By Brian Wang

25 thoughts on “Tesla AI Chip Claimed to be Twenty Times Faster than Nvidia”

  1. No its doesn’t work like that. You need the camera but you also need a log of what happened and what decisions were made.

  2. Of course!
    Human brains are also neural networks. They become good at what they are trained for within their individual architectural limitations.

  3. Funny thing is that if traditional automakers simply adds cameras to their current fleet they will have much more data that Tesla.

  4. “However, in a car, something breaks down and humans cause a lot of accidents. Situational awareness is not good/fast enough for the tempo in motorized traffic. Rate of accidents is pretty much unacceptable.”

    So regarding this. I work on a assembly line and I have to say one thing I have noticed is that people that come in and have not been in a physically demanding job where you have to navigate and use your body do a lot worse at the job than those that have been either through work or a lot of sports.

    So part of that may just be that due to the new ways people live and work they just straight up have lost their situational awareness of things.

  5. If I am wrong here, please someone correct me. But Nvidia GPU’s are vector processors that can do convolutions after a fashion. So saying you have patented a processor for doing one type of calculation is not saying a great deal.

  6. I want a screaming (echo-location) system with blinding laser distance detectors in 360 degrees, that is also effectively a padded insane asylum of a car with a thin yet potent layer of orange juice coating the outer surface area such as to start a fire when I do a sick burn-out when I dunk on ya girl.

  7. He should start to finish other ongoing projects first and then start new ones.. he’s like a kid that plays with Legos…

  8. “However, in a car, something breaks down and humans cause a lot of accidents. Situational awareness is not good/fast enough for the tempo in motorized traffic. Rate of accidents is pretty much unacceptable.”

    It’s true that our situational awareness and reflexes are not nearly as efficient/fast as a machine’s. But is that the cause of most traffic accidents? Or is it the lack of a driver’s attention due to distraction, inebriation, sleepiness etc.? I wonder what the accident stats would be if we only considered situations in which the humans were sober, well rested, and paying attention to the road. Also let’s factor out reckless driving and racing.

    I believe Musk was referring to this optimized human condition when he said we do all right with our natural abilities. There’s definitely room for improvement (which autonomous vehicles will eventually provide), but we still have the bar set relatively high. By the way, that’s a fascinating demo of the world through an autonomous car’s eyes!

  9. Humans do amazingly well on foot in crowded places. We never bump into each other unless we are drunk. However, in a car, something breaks down and humans cause a lot of accidents. Situational awareness is not good/fast enough for the tempo in motorized traffic. Rate of accidents is pretty much unacceptable. This indicates that a self driving car need to have a bit more than human sight and hearing.

    I saw an interesting demo the other week from situational awareness software. I think it was open-sourced by Uber. It shows how the software sees the world. https://avs.auto/demo/index.html

    Use the latest Firefox browser for fluid performance. It uses the latest web assembly tech to make a weak little laptop display things like that demo much faster than a professional workstation would. I didn’t believe it until I tried it…

  10. But human vision’s slow and limited in spectrums compared to machine vision. So there’s opportunity for complementarity.

  11. He was referring to the human vision/brain system. We currently use that when we drive and it’s worked (in most cases) for 100+ years.

  12. Elon said we drive with two cameras on a gimbled suspension and we do alright because of the hardware and software between those two cameras. Why should autopilot be different.

  13. Frankly, I want a self driving car with redundant, overlapping sensor systems – so if one system is degraded by the elements, the other is hopefully stiff functional. Give me cameras, LIDAR, GPS – hell, sonar too – echo location works well for bats. Not that any of them would have helped that much driving in yesterday’s blizzard.

Comments are closed.