Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Tesla, TSLA & the Investment World: the Perpetual Investors' Roundtable

This site may earn commission on affiliate links.
This is quite fascinating, but is there anyone here actually understand it? :confused:

I'm nowhere near the level of the team working at Tesla. However, I do have a small amount of hardware / design background, and over 20 years of software design / programming / launch / support / etc experience.

In summary: the chip is a very impressive design. The team did a fantastic job in balancing performance / heat production / energy usage / memory management / security. Power vs heat is a trade-off. They pulled it off while also gaining more performance (my guess) than they aimed for.

It's impressive. Add as Elon just said, all they need now is software. This hardware is plenty for what they need (my impression, that matches what they are saying).

of course, in software, there is always the unexpected. I am of the opinion that Tesla is doing the right approach to FSD. I think they're farther away than being ready in 2019, but they are in a league of their own (positive) compared to any competitor.
 
To put this in perspective, I have an Intel 8700k in my gaming rig running at 5 ghz. Its thermal design power (stock) is 95 watts.

The Tesla ASIC is ~72 watts.

Here is is the heatsink I use on my computer, but I believe the Tesla computer is plumbed through the coolant loop.
 

Attachments

  • 844514823_1560631579_o.jpg
    844514823_1560631579_o.jpg
    66.7 KB · Views: 50
.
I'm impressed by Trip - he actually has some knowledge about neural network and asked very specific questions about whether the chip can support other activation functions like tanh or sigmoid functions.
Me, too. Besides requiring technical knowledge, it's a perceptive question about how hardwired this chip is.

Answer, it's definitely hardwired for image recognition but they've reserved some flexibility to tweak their NN over time.
 
  • Love
  • Like
Reactions: kbM3 and neroden