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

Inside the NVIDIA PX2 board on my HW2 AP2.0 Model S (with Pics!)

This site may earn commission on affiliate links.
Looks like this developer dude has a board with a different connector arrangement than @kdday... You can clearly see a purple connector in the top right corner. (Remember the sticker of @kdday's ecu, it said "No Purple Connector"...)

17125577_288059814957556_1767574227092766720_n.jpg
 
Nonsense !
YOU have no idea what tsla (sw) can do with current hw on AP2 cars ...
Neither do I, but they did highway stuff with 40x less processor power.

I think the difference is that MobilEye never disputed that the AP1 software was capable of what Tesla used it for; they just thought Tesla was being too aggressive with it and weren't properly representing what it was. That's all debatable.

With Nvidia you have them stating what each of their hardware sets is capable of and what it's presumed that we have, is not near what the manufacturer of the hardware says is necessary. They say you need two of their top of the line units. AP2 cars apparently don't even have one.

So I think there's a valid argument to be made that indeed the hardware for full self driving might not currently be present in the cars and they may require an upgrade. On the flip side of that argument, Tesla may have spec'd out their own Drive PX2 from Nvidia or Nvidia might be taking LADAR into account when stating their requirements whereas Tesla doesn't see LADAR as a necessity. All of this muddies the water.
 
  • Informative
Reactions: davidc18
I think the difference is that MobilEye never disputed that the AP1 software was capable of what Tesla used it for; they just thought Tesla was being too aggressive with it and weren't properly representing what it was. That's all debatable.

With Nvidia you have them stating what each of their hardware sets is capable of and what it's presumed that we have, is not near what the manufacturer of the hardware says is necessary. They say you need two of their top of the line units. AP2 cars apparently don't even have one.

So I think there's a valid argument to be made that indeed the hardware for full self driving might not currently be present in the cars and they may require an upgrade. On the flip side of that argument, Tesla may have spec'd out their own Drive PX2 from Nvidia or Nvidia might be taking LADAR into account when stating their requirements whereas Tesla doesn't see LADAR as a necessity. All of this muddies the water.
Nvidia doesn't say you need two... they say you can have multiple depending on your needs. If your models are running half precision floats or even int8 then they can run significantly faster than single precision even on a hybrid half card that Tesla uses for AP2.

I'm not sure what LADAR is, maybe it's related to LiDAR.

MobilEye was embarrassed after the accident and wanted to distance themselves from Tesla. It was their system than didn't detect cross traffic and it was never designed to anyway. Intel is now involved with MobilEye and some auto manufacturers are jumping ship and switching to Nvidia based solutions or combos with Nvidia and the cheap MobilEye EyeQ3. Toyota, Volvo, Audi, Mercedes, Honda, BMW, and Tesla to name a few.
 
I think the difference is that MobilEye never disputed that the AP1 software was capable of what Tesla used it for; they just thought Tesla was being too aggressive with it and weren't properly representing what it was. That's all debatable.

With Nvidia you have them stating what each of their hardware sets is capable of and what it's presumed that we have, is not near what the manufacturer of the hardware says is necessary. They say you need two of their top of the line units. AP2 cars apparently don't even have one.

So I think there's a valid argument to be made that indeed the hardware for full self driving might not currently be present in the cars and they may require an upgrade. On the flip side of that argument, Tesla may have spec'd out their own Drive PX2 from Nvidia or Nvidia might be taking LADAR into account when stating their requirements whereas Tesla doesn't see LADAR as a necessity. All of this muddies the water.
I think the main difference is Mobileye offered an end to end solution including the software and the sensors specified. The impression I get from Nvidia is that they themselves do not have a complete software solution yet and still developing their own. That's why people in general have little confidence what they are saying necessarily applies to Tesla.
 
Nvidia doesn't say you need two... they say you can have multiple depending on your needs. If your models are running half precision floats or even int8 then they can run significantly faster than single precision even on a hybrid half card that Tesla uses for AP2.

Everything I saw in the early literature suggested you needed two.

I'm not sure what LADAR is, maybe it's related to LiDAR.

It's that thing when you're trying to type response to a post and you have two children running around you screaming and type LADAR several times instead of LIDAR.

I think the main difference is Mobileye offered an end to end solution including the software and the sensors specified. The impression I get from Nvidia is that they themselves do not have a complete software solution yet and still developing their own. That's why people in general have little confidence what they are saying necessarily applies to Tesla.

That's a fair point.
 
  • Like
Reactions: davidc18
Everything I saw in the early literature suggested you needed two.
That's never been in the literature... the picture had two though :)

DRIVE PX 2 FOR FULLY AUTONOMOUS DRIVING
Multiple fully configured DRIVE PX 2 systems can be integrated in a single vehicle to enable autonomous driving.
https://web.archive.org/web/20161118233019/http://www.nvidia.com/object/drive-px.html

This means if you need to you can have two or three...or four. If you used single precision you'd need two just to get the performance of one card with half precision.
 
  • Informative
Reactions: davidc18
I think the main difference is Mobileye offered an end to end solution including the software and the sensors specified.

But Tesla didn't use Mobileye's end-to-end solution, they only used part of the EyeQ3 functionality and programmed the rest of AP themselves.

NVidia offers software too, but Tesla chose not to use it either. (And maybe with NVidia's software you would need more than one of their boards.)
 
But Tesla didn't use Mobileye's end-to-end solution, they only used part of the EyeQ3 functionality and programmed the rest of AP themselves.
For the purposes of the argument that doesn't really matter. Mobileye was able to demonstrate commercially working level 2 with the sensors specified (camera built to their specification, supplemented with radar). Automakers can go out and buy this; might not be as capable as Tesla's AP1, but it certainly would be level 2. Thus they have credibility.

NVidia offers software too, but Tesla chose not to use it either. (And maybe with NVidia's software you would need more than one of their boards.)
NVidia offers software tools and libraries to aid in developing semi-autonomous and autonomous cars, but unlike Mobileye, they haven't demonstrated they have all the software developed that automakers can just purchase from them (with specified sensors) to make a full Level 4 or Level 5. Nvidia's own level 4/5 prototypes are very much still under development and not anywhere near commercially ready.

So they themselves do not necessarily know how much processing power is really required or if they can improve the efficiency of the algorithms by the time it's commercially ready.
 
  • Helpful
Reactions: davidc18
I think the debate on whenever you need two or not is a bit pointless.
HW2.5 is clearly two PX2 boards. The biggest question now is if HW2 of later revisions is also two boards or not, and for that somebody needs to do another teardown.

Additionally code implies hw 2.5 is not necessary model 3

My screwdrivers are ready. Point me in the direction of an HW2.5 car. ;)
 
If someone finds a difference in hardware on a 2017 HW 2.5 car, could we also speculate if these owners are more satisfied with autosteer?

From when did Hw2.5 show up? May??
First mentioning of HW2.5 I saw was in 17.17.4 firmware released on May 5th. Dual px2 setup references appeared at the same time.
17.14.23 did not have it (released Apr 6th).
I don't have a copy of 17.16.35 that was released in-beween the two to see what was there.
My car built mid-march 2017 has HW2.0 rev C.
The code is shared between the two so I don't think any meaningful difference is there yet no matter what revision you have.
 
Update that people will likely care about-


TL;DR - Tesla put a note on my file that I had disassembled the glove box and taken pictures of my ECU in May and attempted to deny a warranty claim today for camera calibration. They keep a black list.

Post -

I recently got a large crack in my windshield and after waiting 3 weeks for the replacement glass to come from Tesla, I had their preferred glass shop replace the windshield. They did a good job.

The glass replacement technician warned me that I'd need to visit Tesla to get my forward facing cameras recalibrated. Indeed, when driving today my car was not tracking the lane correctly and was hugging the left lane line and ping-ponging a bit.

I went into Tesla this morning to get my cameras recalibrated. They were good about looking at my car and appeared to check things over. They even found a nail in a tire and offered to replace the tire for $517. (Side note, I had the tire repaired for free elsewhere, but tesla did loan me a wheel for free for the day - props to them). Still, service was good and they appeared to look at the cameras. The technician handed me my keys back and said it'd take about 100 miles for my car to fully recalibrate. Makes sense, I remember going through that when AP2.0 first arrived. However, autopilot was immediately usable and after driving all day, nothing changed and car still hugs to the left and ping-pongs. Cameras seem to be culprit in my mind. Also, the ambient light sensor no longer was working and my lights were always on. I took it back to Tesla later in the day and the moment I pull up I'm told that this will be treated as an out of warranty claim because there "was a note on my file that I had disassembled my ECU" and that investigation further would require me to pay $125/hr. I was respectful and honest and told them I absolutely disassembled the glove box and took pictures of my ECU to see the "super computer" in my car, but I had a hard time believing that doing so 5 months ago (see page 1 of this thread) was related to replacing my windshield yesterday when my lane tracking started behaving erratically.

The technician was kind and said "we'll look at it, and if it's the cameras, Tesla will cover the cost, but if it's the ECU, you'll have to pay". I told him I agreed and they proceeded to look.

They pushed a firmware update to try to fix the ambient light sensor and cleared the previous calibration and told me autopilot would be inoperable for 100 miles and I would not be able to activate it. Turns out that fixed it and the car drives straight as an arrow now.

Just a word of warning, Tesla doesn't like forum posts of this type I suppose.

Also, I take full responsibility for taking pictures of my ECU and am comfortable with those consequences (which should be zero). I just don't like the black balling of a customer. Still, they were kind and professional and I appreciated them . Maybe I'm wrong and deserve it.
 
TL;DR - Tesla put a note on my file that I had disassembled the glove box and taken pictures of my ECU in May and attempted to deny a warranty claim today for camera calibration. They keep a black list.
The problem is this forum not respecting your privacy. Way back in 2015 I said something about some issues with my car, and I got a call from the local service department asking me if they could take a look at it. This was before I even decided whether it was worth my time bothering with.

I didn't post any pictures or any way to trace the post to me. The only way they'd know it was me is if the owners/maintainers of this forum gave my information to them.

[Admin note: This is accusation is absolutely false. Read More.]

Maybe I didn't click some "do not share my information" button, but I don't remember any such button.

To me this is unacceptable. I still use this site, but it's absolutely not a free speech zone.

It's more of a passive-aggressive way to talk to Tesla.

In the spirit of that I'd like to give thanks to Tesla, and my local service department for the most awesome rattles on wheels I've ever owned. :p
 
Last edited by a moderator:
  • Love
Reactions: vrykolas
The problem is this forum not respecting your privacy. Way back in 2015 I said something about some issues with my car, and I got a call from the local service department asking me if they could take a look at it. This was before I even decided whether it was worth my time bothering with.

I didn't post any pictures or any way to trace the post to me. The only way they'd know it was me is if the owners/maintainers of this forum gave my information to them. Maybe I didn't click some "do not share my information" button, but I don't remember any such button.

To me this is unacceptable. I still use this site, but it's absolutely not a free speech zone.

It's more of a passive-aggressive way to talk to Tesla.

In the spirit of that I'd like to give thanks to Tesla, and my local service department for the most awesome rattles on wheels I've ever owned. :p
That's a pretty serious accusation. I'm pretty sure from your location and post history there is probably enough hints to easily piece together (in about 1 minute of search I was able to find what car you have, when you took delivery, and also details about your delivery process).

It does seem a bit big brother-ish, but those of us here long enough do know some Tesla employees do read these forums.
 
Last edited:
  • Like
Reactions: doug
Update that people will likely care about-


TL;DR - Tesla put a note on my file that I had disassembled the glove box and taken pictures of my ECU in May and attempted to deny a warranty claim today for camera calibration. They keep a black list.
...
Also, I take full responsibility for taking pictures of my ECU and am comfortable with those consequences (which should be zero). I just don't like the black balling of a customer. Still, they were kind and professional and I appreciated them . Maybe I'm wrong and deserve it.
I figured something like this might happen from the back of my head. I remember thinking to warn about redacting any identifying information, but I didn't notice any VIN identifying info in the pictures, so dropped the idea. I'm guessing there is enough details posted about the car elsewhere that it's not necessary to pin down which car it is however.

A lot of media also reported your post, so even if Tesla didn't directly monitor this forum (which there is evidence that they do), they probably would have noticed. I have total sympathy. Definitely doesn't feel good that this kind of list exists. I guess this serves as a word of warning to all the others that might want to try to do the same thing.

At least the technician didn't try to pull the wool over you by telling your whole warranty is void. Just to remind everyone: a manufacturer only has the right to void the warranty for a specific part. They can't void the entire warranty just because of actions that would only affect one part of the car.
 
Last edited: