I am sure a retrofittable part would be designed with this in mind..?
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
HW2.5 capabilities
- Thread starter verygreen
- Start date
-
- Tags
- Autonomous Vehicles
I am sure a retrofittable part would be designed with this in mind..?
lunitiks
Cool James & Black Teacher
That's exactly my point nr. 1. Tesla would have to design a retrofittable part for the S and X windshields respectively.
OK. Sure.
I thought you felt they'd have to replace glued parts.
Anyway, I don't see the interior camera retrofit for all happening very likely.
lunitiks
Cool James & Black Teacher
I'm going to post this as many times as necessary. Every HW2 car - no matter what options are purchased - was sold under the claim that it had the hardware on it. They can jack up the cost of post-order upgrades, of course. But if new hardware is needed, it goes to all HW2 cars. Tesla can't tie the hardware upgrade to the software activation. They are separate.
OK - you are going to insist on hardware you are not going to activate because.....?
Do you still think that is true even when Tesla is now saying “The internal name HW 2.5 is an overstatement, and instead it should be called something more like HW 2.1. This hardware set has some added computing and wiring redundancy, which very slightly improves reliability, but it does not have an additional Pascal GPU.”
Yeah I'm really really curious about that. Whether or not it has another GPU will be the most telling statement about what FSD will require. If they just mean it has a GPU but isn't being actively used… that's really different from going into MP for Model 3 without a second GPU altogether.
A couple extra Denver/A57 cores is not terribly exciting to me.
Lol. I don't need a reason other than it was an advertised capability of the car.
They have to give you wheels even if you never want to drive it.Lol. I don't need a reason other than it was an advertised capability of the car.
My five year old can articulate better reasons when asking for things.
No one is busting Bill Gates for saying 640k memory ought to be enough for everybody today.
It just seems like he articulates better reasons because you have no ****ing clue.
No clue about what? How to defend wanting hardware and wasting everyone's time on something that has no value?
If you didn't pay for FSD, there is no tort. Or you could pay for FSD and if that camera isn't needed - there is no tort.
Yes, I still think there's dual gpu setup. Adding just the arm cores makes no sense.
I don't know if the spokeperson was just misinformed or if the dual px2 setup is considered to be one gpu still in their eyes or whatever.
Heh I completely agree with your sentiment. I said the same earlier — I can't possibly believe the motivation for adding those ARM cores alone without adding an additional GPU. Maybe it buys you a little bit of hardware failure redundancy to do some sort of basic pull-over-safely maneuver, but why bother coding that up when you can just shove in a second identical board for failover?
My crazy theory is: They simply mean that the GPU on the second board isn't being used / planned on being used currently.
Does not buy you that.
You see, the cameras are processed by neural net on GPU attached to the primary node (I somehow doubt they'd do dualhomed pci-e setup as they won't need gige to communicate the nodes then).
This means if the primary arm node dies/panics/whatever, the second node is blind, how are you going to pull over without seeing anything, you can just stop in place (i.e. as fast as you can), that's about it, but you don't need another node to just apply brakes superhard, gateway could do that on loss of contact with the ap node.
Sure, understood. I would imagine such an architecture would require some state / "map" (at least to the the bailout location) to be communicated to the second APE, so that no neural net processing would be necessary to execute the maneuver.
With that said, "why bother coding that up" is an understatement for how I feel… It's a dumb solution, has lots of corner cases that it can't cope with, and there's no reason to do it this way if you're revving the board to add a failover/redundancy solution.
So in my mind as someone that's been architecting embedded systems for the better part of a decade, I suspect the motivation is one of:
(1) It really does have a second GPU and that's their backup plan if they cannot achieve their goals with 1 GPU.
(2) They discovered some sort of defect/weakness on the HW2.0 board that could be worked around by having a parallel CPU-only board communicate. It would have to be something outside of the CPU / OS such that the neural net still works on the main board (e.g. CAN output, whatever else).
#2 is way out there too. I'd still guess #1.
lunitiks
Cool James & Black Teacher
Yes it's quite curious that they upped the computing power if it's not for redundency. Why bother? The price of the existing AP2.0 boards must be falling (like the ones in the MCU must be), why not take advantage of that and get more $$$.
I suspect (speculate, rather) that they're worried about AP2.0ECU running at full capacity / too high rate in certain scenarios with only two cameras and some lazy EAP SW "active", and that an additional gpu is required for real EAP or so-called FSD. Like @AnxietyRanger often puts it; we shall see.
I suspect (speculate, rather) that they're worried about AP2.0ECU running at full capacity / too high rate in certain scenarios with only two cameras and some lazy EAP SW "active", and that an additional gpu is required for real EAP or so-called FSD. Like @AnxietyRanger often puts it; we shall see.
lunitiks
Cool James & Black Teacher
Beginning to suspect that hw2.5 was sneaked in around summertime this year.
Couple of reasons:
@verygreen noted a while ago that 2.5 should have somewhat different wiring. Electreks source in the 9-8-2017 article seems to corraborate this:
Anyway I've noticed that the part no. for the main dash/ip harness changed from "XXXXXXX-03-X" to "XXXXXXX-93-X" sometime between april/may and july 2017, both on the S and the X. Interestingly, this is the same period in which the "G"-revision of the ECU showed up in the parts manual.
@skilly's post here kind of supports my theory...
Couple of reasons:
@verygreen noted a while ago that 2.5 should have somewhat different wiring. Electreks source in the 9-8-2017 article seems to corraborate this:
Anyway I've noticed that the part no. for the main dash/ip harness changed from "XXXXXXX-03-X" to "XXXXXXX-93-X" sometime between april/may and july 2017, both on the S and the X. Interestingly, this is the same period in which the "G"-revision of the ECU showed up in the parts manual.
@skilly's post here kind of supports my theory...
Similar threads
- Article
