I'm curious to see some Tesla owners track on the compute load on their FSD Computers. In April, Elon
tweeted:
“The Tesla Full Self-Driving Computer now in production is at about 5% compute load for these tasks [i.e. Navigate on Autopilot] or 10% with full fail-over redundancy”
The same day Elon also
tweeted that that the compute load on HW2.5 was “~80%”.
Presumably Tesla wants to use most if not all of the FSD Computer's compute. Before the city driving features go to wide release, presumably Tesla will want to run them passively, i.e. in shadow mode. So, if Tesla owners with HW3 can track the compute load in their cars, we should be able to tell when the city driving features are released in shadow mode.
I suppose it's possible the city driving features will be like Summon and will be able to run on HW2. But my guess is that these features will accompany much larger neural networks that use a lot more compute.
In October 2018, Karpathy
said:
“...my team trains all of the neural networks that analyze the images streaming in from all the cameras for Autopilot. For example, these neural networks identify cars, lane lines, traffic signs, and so on. The team is incredibly excited about the upcoming upgrade for the Autopilot computer which Pete [Bannon] briefly talked about.
This upgrade allows us to not just run the current neural networks faster, but more importantly, it will allow us to deploy much larger, computationally more expensive networks to the fleet. The reason this is important is that, it is a common finding in the industry and that we see this as well, is that as you make the networks bigger by adding more neurons, the accuracy of all their predictions increases with the added capacity.
So, in other words, we are currently at a place where we trained large neural networks that work very well, but we are not able to deploy them to the fleet due to computational constraints. So, all of this will change with the next iteration of the hardware. And it's a massive step improvement in the compute capability. And the team is incredibly excited to get these networks out there.”
It's been over a year and, as far as I know, we haven't seen new neural networks running on HW3 that are too computationally intensive to run on HW2. I would guess that's what's supposed to be coming with the city driving features.