Tesla Just Bought Software That Can Run Its AutoPilot With 1% SoC Utilization


Tesla is planning to accelerate its plans for full autopilot mode in its vehicles by purchasing computer vision startup DeepScale.

CEO Elon Musk hasn't been quiet about his ambition to turn his customers' vehicles into a massive fleet of Uber-like taxis on-demand. He claims that purchasing a Tesla (NASDAQ:TSLA) could be the very first automobile purchase to actually appreciate in value: a buyer could actually recoup the cost of the vehicle over time by sending it out into the world to help ferry passengers around for cash before it returns in time to give its owner a ride to work.

How to Run Android Apps on Windows 11 (Insiders Only)

Of course, once at the office the self-driving Tesla could then return back to the streets to earn more cash while its owner is plugging away in a cubicle. It sounds like quite the dream.

However, in order to fulfill this promise of "robotaxis" Elon's electric vehicle company still has some challenging technological hurdles to surpass. One of them is the problem of pairing a very resource-limited local computational system with all the inputs a self-driving car would receive such as images, sonar readings and a multitude of other sensor telemetry.

Tesla to put DeepScale's software to work with its in-house hardware

Tesla has made strides there and even has revealed bits about an in-house developed SoC that puts out some seriously impressive numbers. Click here for our coverage of the impressive board.

That's where the DeepScale comes in. The start-up specializes in software that helps computer systems learn and decipher the images that are fed to them, also known as computer vision. Interestingly, neither Tesla or DeepScale announced the acquisition, but some hawk-eyed observers spotted DeepScale's CEO Linked-In page get updated with this post.

AMD Ryzen ‘Rembrandt’ APU With 8 Cores, 16 Threads & RDNA 2 GPU Spotted, Running On Corsair Xenomorph AIO PC With DDR5-4800 Memory

Forrest will go to work for Tesla as a senior machine learning scientist. Iandola earned his PhD in computer science and electrical engineering and his background consists of machine learning in deep neural nets.

Here's the real reason Tesla went after DeepScale: they are extremely focused on doing all this wizardry on a low power budget. DeepScale demoed its software stack on an NVIDIA (NASDAQ:NVDA) Xavier SoC and it consumes merely 2% of the available processing power during live highway automated driving. Keep in mind the Xavier SoC can perform 30 TOPS (terra-ops/second) and Tesla's new chip can perform 72 TOPS. At that rate, DeepScale's software module would be consuming less than 1% of available processing power!

Here is an excerpt DeepScale's blogpost on its modular Carver21 solution:

DeepScale’s full-stack deep learning methodology enables cohesive integration of AI software with various processors and sensors for customizable automated driving features. Full-stack deep learning means that DeepScale has experts working together on every aspect of deep neural network (DNN) training, development, deployment, and even data collection/curation to produce proprietary state-of-the-art AI solutions for our customers:


It will be very interesting to see what Tesla can do with this technology. I will say that Tesla could probably have accessed it before, so we must wonder if this isn't just about giving Tesla comprehensive access but also ensuring no one else gets their hand on this tech. According to DeepScale: "Now, DeepScale brings efficiency and modularity to today’s in-focus automotive technologies of artificial intelligence (AI) and automated driving." So an easy to implement, modular system would be very attractive to any automobile company, and it makes a good case for why Tesla opted to gobble it up.

Tesla or DeepScale did not immediately return a request for comment.


The author has no position in any of the stocks mentioned. WCCF TECH INC has a disclosure and ethics policy.