-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
he may be right. (although other news site say it comes with one)
GTX 1060 is about 3.8 TFLOPS.
Last edited:
The gaming models are called GP106-300/400/410-A1, this one is GP106-505-KC-A1, a different version then.
Found something that looked very similar here:
https://www.chiphell.com/thread-1726772-1-1.html
For example here:
"It has a Tegra chip that combines one six-core CPU code-named Parker and one Pascal-based GPU. It is a watered down version of the original Drive PX 2 announced at CES, which sits in the trunk of a car and is far more powerful with two CPUs and two GPUs."
Nvidia's updated Drive PX 2 computer will drive autonomous cabs
@stopcrazypp
@bjornb
There's no mention of difference between integrated and discrete gpu.
EDIT: on the other hand nvidia mentions one processor...making things confusing.
"It has a Tegra chip that combines one six-core CPU code-named Parker and one Pascal-based GPU. It is a watered down version of the original Drive PX 2 announced at CES, which sits in the trunk of a car and is far more powerful with two CPUs and two GPUs."
Nvidia's updated Drive PX 2 computer will drive autonomous cabs
@stopcrazypp
@bjornb
There's no mention of difference between integrated and discrete gpu.
EDIT: on the other hand nvidia mentions one processor...making things confusing.
JeffK
Well-Known Member
Last edited:
Its part of the Parker SOC (I have never found a picture showing the underside of the Autocruise version though..).can you spot the integrated gpu here?
If you look at the slide above it says 10W power consumption, that excludes a discrete GPU in my opinion.
JeffK
Well-Known Member
Looks like The Tegra X2 with 1.5 flops FP16 is the integrated GPU. Reading way too many articles with contradicting wording had me confused.
Introducing Parker, NVIDIA’s Newest SOC for Autonomous Vehicles | NVIDIA Blog
Introducing Parker, NVIDIA’s Newest SOC for Autonomous Vehicles | NVIDIA Blog
Total speculation here but is it possible that Tesla purposely under-spec'ed this setup to accommodate AP1 parity (and maybe a little beyond that) with the understanding that it would never support full level 5 autonomy and that they'd have to upgrade these cars prior to that rollout?
A few reasons for thinking this:
1. I vaguely recall Nvidia stating that you'd need two of the dual CPU versions for full self driving. Everyone speculated that Tesla would try to make one with, presumably with their own software. But making one of the single CPU versions work seems like a stretch.
2. Because of the Mobil Eye debacle they were forced to find a suitable replacement and integrate it quickly. It would make sense that they'd go for the cheapest setup that would offer both AP1 parity and an easy upgrade path to full autonomy since that was their ultimate goal. The other indirect benefit of doing this is that the better hardware is almost certain to come down in price by the time they actually need it.
A few reasons for thinking this:
1. I vaguely recall Nvidia stating that you'd need two of the dual CPU versions for full self driving. Everyone speculated that Tesla would try to make one with, presumably with their own software. But making one of the single CPU versions work seems like a stretch.
2. Because of the Mobil Eye debacle they were forced to find a suitable replacement and integrate it quickly. It would make sense that they'd go for the cheapest setup that would offer both AP1 parity and an easy upgrade path to full autonomy since that was their ultimate goal. The other indirect benefit of doing this is that the better hardware is almost certain to come down in price by the time they actually need it.
The downside is you'd need two teams to work on it all in parallel. And you are not getting much testing on the nonpublic code,
with two teams they'll reinvent the bicycles all the time and real world lessons from one team might not propagate into the other one.
You've done more than enough and more than anyone else had the stones to do. You don't want to reuse the exposed thermal paste either but you don't have any new stuff on hand. Next time I'll ship some silver paste to you. I know I've got some lying around but its about 4 years old -- I don't think it goes bad though. Thanks for doing this and posting pictures!
stopcrazypp
Well-Known Member
Good on you on finding a picture of the bottom. This confirms it has no discrete GPU (no MXM connector for one, nor one soldered to the bottom).
Last edited:
Reciprocity
Active Member
Nice work man, you have cast iron balls. One question for you and the thread in general. Did you purchase FSD? Is it possible, people that purchased FSD up front could have different hardware? More then likely not, but just trying to eliminate the idea card on @kdday pre purchased options.
Things get alot dicey when u have to make a prediction for every object on the road in real time. Plus they have a back up redundant system.
stopcrazypp
Well-Known Member
Actually it says that right on the Nvidia website:
Autonomous Car Development Platform from NVIDIA DRIVE PX2
Before, most people thought Tesla had Autochauffeur. Now we know they have something between Autocruise and Autochauffeur (more accurately, something equivalent to like half an Autochauffeur board). In neither case does it meet Nvidia's criteria for FSD (which is multiple Autochauffeur boards).
Last edited:
Similar threads
