Is that really so, though? I only watched the hardware section and I wasn't taking notes, but that slide shows the design goals...and the text seems incorrect.
For example, the design goal was 50 TOPS. They said later in the presentation they achieved 72TOPs on the actual chip, so the tweet is already incorrect. The FSD computer has two chips on it (one for redundancy), so in less than 100W they are actually achieving 144TOPs. Now this guy has everything wrong so far but if we take his numbers at face value Mobileye seems competitive as far as performance per watt, but it looks like it might be a chip for next year on a different process node according to this:
https://www.mobileye.com/our-technology/evolution-eyeq-chip/ ...Nvidia, not so much. Yeah, they have a little over double the performance if you're willing to throw 4x as much power and cooling at it, and in an electric car, taming power consumption is pretty important if you want to maximize range.
C'mon, man, SRG01 you're an engineer right? It's disappointing to see you misrepresent stuff like this.