Aaaand this thread reached Electrek
First picture of Tesla’s new NVIDIA onboard supercomputer for Autopilot installed in a car
First picture of Tesla’s new NVIDIA onboard supercomputer for Autopilot installed in a car
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
FIRST PICTURES - Tesla Autopilot 2.0 ECU (Nvidia PX 2)
- Thread starter lunitiks
- Start date
All we know for sure is that it can be "activated" later. No one knows whether the activation is software only as no one has done it yet.
phigment
Member
Surely EAP > FSD is pure software upgrade. And no car has the "Fully Autonomous" version of the PX2. Tesla's hoping that all them computing powerz isn't necessary beacause of their huge fleet of cars (experience). And as a bonus it's cheaper!
But hey, bets have been placed in this thread before so why not go again
EDIT: I'm a little curious about what Nvidia exactly means by "Point-to-point travel" vs "fully autonomous driving". No question that we're expecting fully autonomous point-to-point driving! But, most of the time you sit your butt in the car aren't you planning on going from point A to point B?
So the "fully autonomous" PX2-version must be designed to allow for something more than this regular driving. Perhaps just thoughtless cruising around in the streets, hoping to pick up paying customers?
Anyway, the SAE definitions don't differenciate between "Point-to-point" and "Full auto". They're only concern is to which degree the driver is expected to intervene. L4 is no intervening, but allows for limiting the system to special road conditions or special areas/roads. L5 is also no intervening, but no limits as to where in the world it's supposed to work.
But hey, bets have been placed in this thread before so why not go again
EDIT: I'm a little curious about what Nvidia exactly means by "Point-to-point travel" vs "fully autonomous driving". No question that we're expecting fully autonomous point-to-point driving! But, most of the time you sit your butt in the car aren't you planning on going from point A to point B?
So the "fully autonomous" PX2-version must be designed to allow for something more than this regular driving. Perhaps just thoughtless cruising around in the streets, hoping to pick up paying customers?
Anyway, the SAE definitions don't differenciate between "Point-to-point" and "Full auto". They're only concern is to which degree the driver is expected to intervene. L4 is no intervening, but allows for limiting the system to special road conditions or special areas/roads. L5 is also no intervening, but no limits as to where in the world it's supposed to work.
Last edited:
I would imagine that the point to point version struggles with busy urban environments.
I get the point though - Tesla exceeded expectations with EyeQ3 so maybe they will be able to do it again and squeeze more out of the new hardware than Nvidia expect.
Plus, we still don't know exactly what's in the box...
I get the point though - Tesla exceeded expectations with EyeQ3 so maybe they will be able to do it again and squeeze more out of the new hardware than Nvidia expect.
Plus, we still don't know exactly what's in the box...
JeffK
Well-Known Member
There are other parts to driving such as parking in a parking lot, pulling into a garage, if the car could go charge itself, etc. These are more complicated than acting as a taxi or chauffeur.
That's not to say you couldn't do autonomous driving with the two GPU setup. They are just suggesting that depending on your needs and software efficiency, there is no issue using multiple units in a single car.
Yeah but parking and pulling into a garage is without question coming with the Autochauffeur ("Point-to-point") variant in our Teslas. Making the car go to a charing spot by it self would also be "point to point" in my view ...
Henry169
Member
Ok I don't "know" it but it seems pretty obvious given the physical dimensions. Also read my arguments earlier in the thread about Elon's statement about it's computing power (12 vs 24).
Also I have a hard time believing they would cool a double decker "Fully autonomous" version with only a couple of fans inside the dash.
I might be wrong, but I think I'm right.
Also I have a hard time believing they would cool a double decker "Fully autonomous" version with only a couple of fans inside the dash.
I might be wrong, but I think I'm right.
HumanGenome
Member
I don't think Nvidia's descriptions for each of their boards necessarily equates to that board being a pre-requisite for said capability in reality. Time will tell though.
dhanson865
Well-Known Member
...and this obvious fan noise before AP2 has really started doing any real work?
raises a few potential concerns as the HW is increasingly utilised in the future ...
hey - maybe it will mask the motor whine at least
Shadow mode has those parts running full time before you have access to all the AP features. It won't get any hotter with software updates.
It appears like there's no special duct in place. Judging from the pictures and the statements by @Henry169, it looks like the vents are just circulating air inside the cavity of the dash. It even seems like there's a harness going straight over the fans. So it can't get *that* hot in there
Well it might help with keeping the cabin warm on those chilly days without turning the heat on.
Similar threads
