-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
stopcrazypp
Well-Known Member
Ok, so it's just a similar display to what the instrument cluster is showing, which even a Tegra 3 can handle.
JeffK
Well-Known Member
Tegra 3 can barely handle the internet browser haha
lunitiks
Cool James & Black Teacher
No, I think maybe you have it backwards. The screenshot I posted shows that "Tegra B" (one single CPU) is meant for the "Autopilot UI" tasks. I think "Autopilot UI" means the graphical user interface virtualization on the IC, like the toy cars, fancy speedometres and minority report maps.
"Tegra A + dGPU" seems to be meant for the "Driveworks" stuff. Driveworks is not the UI, instead it's the middleware for all the sensor processing, sensor fusion, computer vision and deep neural network. Granted, it's an SDK, but that PP-slide makes me think that one of the CPUs could/should take care of the dashboard 3D rendering, while the other CPU is supposed to be laser focused on the self driving tasks only -- with the help of dedicated GPU(s).
Or am I reading too much in to that diagram?
stopcrazypp
Well-Known Member
I should clarify. I mean for where it says "Tegra A + dGPU" for driveworks are they talking about 2x dGPU or just 1x dGPU?No, I think maybe you have it backwards. The screenshot I posted shows that "Tegra B" (one single CPU) is meant for the "Autopilot UI" tasks. I think "Autopilot UI" means the graphical user interface virtualization on the IC, like the toy cars, fancy speedometres and minority report maps.
"Tegra A + dGPU" seems to be meant for the "Driveworks" stuff. Driveworks is not the UI, instead it's the middleware for all the sensor processing, sensor fusion, computer vision and deep neural network. Granted, it's an SDK, but that PP-slide makes me think that one of the CPUs could/should take care of the dashboard 3D rendering, while the other CPU is supposed to be laser focused on the self driving tasks only -- with the help of dedicated GPU(s).
Or am I reading too much in to that diagram?
I know that they say Tegra B is for UI. But if it's the latter, then that might imply the Tegra B is paired with a dGPU also, and that would be overkill just for UI tasks.
lunitiks
Cool James & Black Teacher
Aha now I see what you mean.
Well of course I don't know what Nvidia said when that slide came up, perhaps it was just an example / suggestion. My n00b thinking is however that (1) the system is supposed to be scalable, so maybe it allows for 2 Tegras and one dGPU, and/or (2) is it perhaps possible to run both dGPUs against only one of the Tegras?
I dunno, but I completely agree that it seems overkill to run an "entire" Tesla AP2ECU setup for just the virtualization tasks
Well of course I don't know what Nvidia said when that slide came up, perhaps it was just an example / suggestion. My n00b thinking is however that (1) the system is supposed to be scalable, so maybe it allows for 2 Tegras and one dGPU, and/or (2) is it perhaps possible to run both dGPUs against only one of the Tegras?
I dunno, but I completely agree that it seems overkill to run an "entire" Tesla AP2ECU setup for just the virtualization tasks
Last edited:
I don't think so, because the mobileeye chipset is in the upper camera housing I believe, since it's just 1 camera and radar, but I could be very wrong. Also, I seriously doubt they are going to swap in a new chip with that space. More then likely they will swap out the entire board with the new improved one, otherwise they'd have a validation nightmare for each chip / socket connection etc, just not something I see them doing in the service center. Instead they'll ship new tested boxes and swap out the entire assembly, that appears to be the way Tesla rolls.
Duke-U
Member
This board being the EAP only equipped car, and the new board for FSD ordered one, upgraded/swapped for those who pre-ordered?
Correct EAP/FSD is a simple software update, all hardware is the same. Now if it turns out you need a new / improved board to actually make FSD, I would assume those that order FSD would roll into the shop for the new board, and they would update the EAP boards whenever someone chooses to purchase FSD.
JeffK
Well-Known Member
How about a completely different codebase without reliance on the NVIDIA toolkit.... (as they've mentioned it was hardware agnostic)
To be clear, I expect they still leverage the NVIDIA hardware drivers and OpenCL SDK... but not the NVIDIA neural network libraries or their driving models.
JeffK
Well-Known Member
I expect someone to look at the binary or binaries and notice either statically compiled NVIDIA libraries and/or references to dynamically linked NVIDIA libraries. But that doesn't mean they use NVIDIA's driving SDK.
Similar threads
