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

AP 2.5 Vs. 3.0 - Upcoming Tesla Arcade Games - Nvidia

This site may earn commission on affiliate links.
Interesting thoughts after visiting Tesla service today..

I was issued a 2014 Model S PXXD loaner today and started poking around it system. Was really shocked when I launched the games as they were god awful slow and choppy.

I like my Model 3 MUCH better! So glad I didn't go the used Model S route.

Anyways, I got to thinking....

After speaking with the service tech about 3.0 I came to this conclusion. It seems that those vehicles equipped with 3.0 hardware (nVidia chipset) will be absolutely faster and smoother at games. (Like not even close)

The tech said 3.0 is buried inside of the one and only main computer of the car. The S/X have two different units that can be upgraded individually.

When asking for a price to upgrade the computer, I was given a general area of about 3k.

The tech was absolutely awesome to deal with. They did not even have to tell me that as its all still speculative. (and frustrating)

This really makes me unhappy because I have EAP with 2.5. I wish we could get clarity on this issue for those of us that went all in early, but were phased over within 12 months. (I’d also like my upgrade to FSD for 2k) - This needed to be a push notification to the car - Not some secret “Have to be on the net 24/7 thing to obtain)

Ugh!
 
This can get a bit confusing because AP 2.x use NVidia for the AP (APE?) computer, but before Model 3 NVidia was also used for the screen (MCU) computer.

Model 3 switched to using an Intel Atom platform with Intel integrated graphics of some kind for the MCU, I believe. I'm not sure if S/X also switched to that platform or continues to use NVidia platform for the MCU.

Additionally for older versus newer S/X, I believe the older ones have older slower NVidia chips for the MCUs, then newer ones on the newer ones.
 
It also controls the safety related features of the car (LDA, ELDA, FCA, BSD, AEB), and the traffic visualization (dancing cars).
A little OT but what is known about the specs of the UI controller? Is the processor known, does it have a GPU, where is the hardware located (somewhere in the dash?) and how about RAM and Flash memory? Seems like adding games like Beach Buggy may be taxing the Flash. Would be nice if Tesla could add more Flash memory if needed.
 
the FSD computer amounts to two asics (with one in use, one for redundancy) that perform a whooooooole lot of matrix math really quickly to locate and label things. that is all. it's not a general purpose computer (though it does include a small one, and a small gpu that can be used as one). it's job is to perceive... layers above make the decisions.

games run on the atom inside of the MCU. if you have an older MCU, it's an nVidia tegra chipset performing the same functions.
 
the FSD computer amounts to two asics (with one in use, one for redundancy) that perform a whooooooole lot of matrix math really quickly to locate and label things. that is all. it's not a general purpose computer (though it does include a small one, and a small gpu that can be used as one). it's job is to perceive... layers above make the decisions.

games run on the atom inside of the MCU. if you have an older MCU, it's an nVidia tegra chipset performing the same functions.
Both SoCs will be in use at the same time, and their outputs compared, to detect problems. It's not active / spare, it's active / active. They each have two big NN modules (not for redundancy this time, but I'm guessing that performance was better achieved with running two NN's in parallel rather than a single bigger NN), 3x 4-core ARM CPUs (this is hardly "small"), and a small but not tiny GPU for various post processing work, among other things.

So, if it had the appropriate outputs/inputs, it could be a general purpose computer, as it has more CPU power in a single SoC than most smartphones - though this would be a silly thing to do since you'd have no use for the NN portion which is sort of the whole point of this chip.

Tesla has also stated they plan to move more and more of the driving logic into the NN where it makes sense to do so, so at least some decision making will be occurring in the NN versus "software 1.0" regular / heuristic code paths, eventually (or at least, that's the plan).