US$15.000 NVIDIA DRIVE PX 2 supercomputer in every Tesla?

[diamond.g]

Exactly, what this means is that PX2.0 uses the GP102 chip.

Well as long as they are using the fully autonomous version, yeah. Otherwise it would be capped at ~768 CUDA cores which is far short of the 3584CUDA cores GP102 comes with.

That is still a lot of TDP to dissipate, is the processing unit co located where the front facing cameras are?

 

[scaesare]

I suspect Elon was speaking colloquially, and the use of "Titan GPU" was in order to refer to something the audience would be familiar with. The Drive PX 2 is likely in essence the same (or similar enough) technology, but packaged in a form suitable for automotive use, such that the distinction is important for geeks here, but probably not for the general public.

 

[Cosmacelf]

Don't dumb down on us now Elon! Just because some people can't ask you intelligent questions at space conferences doesn't mean the rest of us are stupid!

 

[LargeHamCollider]

Well as long as they are using the fully autonomous version, yeah. Otherwise it would be capped at ~768 CUDA cores which is far short of the 3584CUDA cores GP102 comes with.

That is still a lot of TDP to dissipate, is the processing unit co located where the front facing cameras are?

12 TFLOPS means a full GP102, if the PX2.0 boards were in the rear view mirror I'd be shocked.

 

[Saghost]

12 TFLOPS means a full GP102, if the PX2.0 boards were in the rear view mirror I'd be shocked.

The X has never had the processor in the mirror mount - I'm told they put it somewhere in front of the glove box. Presumably the newer S will go to something similar.

 

[diamond.g]

12 TFLOPS means a full GP102, if the PX2.0 boards were in the rear view mirror I'd be shocked.

Yeah, they are probably in the dash, if they don't use liquid cooling it should be pretty easy to hear the location of the cooling fan for the units after a long enough drive (or if the car is sitting out in the sun).

 

[Cosmacelf]

According to the anantech article about DrivePX, it is always liquid cooled. Either integrated into the car's cooling system or with an external radiator.

 

[diamond.g]

According to the anantech article about DrivePX, it is always liquid cooled. Either integrated into the car's cooling system or with an external radiator.

So would it be in the trunk? To minimize the number of liquid lines coming into the vehicle. I guess that probably doesn't matter.

 

[diamond.g]

hmm, based on the Anandtech article, it may not be a GP102 since the PX2 is multiGPU and the TDP for the whole unit is 250W. the TDP for a single GP102 is 250W. Maybe it is a modified GP104-200. That would put TDP at 300W though.

 

[Xenoilphobe]

Maybe this will be the new heater core for Climate control system.

 

[Tam]

According to the anantech article about DrivePX, it is always liquid cooled. Either integrated into the car's cooling system or with an external radiator.

Autonomous Car Development Platform from NVIDIA DRIVE PX2

There are 3 flavors:

DRIVE PX 2 FOR AUTOCRUISE

DRIVE PX 2 FOR AUTOCHAUFFEUR: Picture shows 2 discrete GPUs: For point-to-point travel and I don't think it is fully Autonomous Driving.

DRIVE PX 2 FOR FULLY AUTONOMOUS DRIVING: Picture seems to suggest multiple boards.

The description seems to say a board with only 1 processor can be passively cooled:

"These range from one passively cooled mobile processor operating at 10 watts, to a multi-chip configuration with two mobile processors and two discrete GPUs delivering 24 trillion deep learning operations per second. Multiple DRIVE PX 2 platforms can be used in parallel to enable fully autonomous driving."

 

[Tam]

I think I understand a little bit better now.

I look up the article that every one was talking about:

NVIDIA Announces DRIVE PX 2 - Pascal Power For Self-Driving Cars

Here's the other side of the Drive PX 2 board:

1 chip board is just a little bigger than a hand but 2 chip board seems much bigger.

So @Frogwheels is correct afterall!

I guess Elon was talking to non-engineer audience when he said "Titan GPU" which made me visualize a Titan X graphic card:

May be he meant in order to run a Drive PX 2 board, he chose the chip from Titan X graphic card called GP 102.

So what are the suggested prices? The board is about $15,000, whole graphic card is about $1,500?

It sounds that $10,000 Tesla Enhanced Autopilot is a real bargain!

 

[snail3]

I wonder how much of PX2's 250W of power consumption going to impact on cars's range?

Nivida will definitely come out with PX3 platform (maybe as early as next year??), maybe single chip can do Full-autonomous.I wonder if PX2's car will be eligible for upgrade

 

[Cosmacelf]

If you are driving 60 miles per hour, the car consumes, say, 350wh/mile, so that's 60x350= 21,000 wh in one hour of driving. The DrivePX 2 will consume 250 wh during that time, so an extra 1.2% power consumption, more or less. So on a full range 250 mile trip, the system would draw an extra 3 miles from the pack...

 

[JeffK]

I wonder how much of PX2's 250W of power consumption going to impact on cars's range?

Nivida will definitely come out with PX3 platform (maybe as early as next year??), maybe single chip can do Full-autonomous.I wonder if PX2's car will be eligible for upgrade

The have info about the future PX3 based on Volta and how many watts online.

Note that the PX2 max usage is 250 watts. The tegra chips themselves draw 10 watts each, the rest of the board can draw more power, it really depends how hard it's working. Even at full usage if ran for 8 hours straight that's 2 kWh or about only 7 miles of range.

Speed is not really relevant.

 

[J1mbo]

The system will probably save more than it uses by driving more conservately than most human drivers.

 

[RFernatt]

Yes! Leve;6! The car decides where you should be going, and takes you to that location. Parents everywhere rejoice.

L6 could go further than parents and teens...

TESLA: "Overriding destination Dunkin' Donuts and re-routing to gym."

 

[AudubonB]

L6 could go further than parents and teens...

TESLA: "Overriding destination Dunkin' Donuts and re-routing to gym."

MMmm... Interesting. The ProfitMaximizingAutomaker might then want to considering offering both the BigBubba as well as the BigBrother versions....

 

[snail3]

The have info about the future PX3 based on Volta and how many watts online.

Note that the PX2 max usage is 250 watts. The tegra chips themselves draw 10 watts each, the rest of the board can draw more power, it really depends how hard it's working. Even at full usage if ran for 8 hours straight that's 2 kWh or about only 7 miles of range.

Speed is not really relevant.

From my understanding of Anandtech:
NVIDIA Announces DRIVE PX 2 - Pascal Power For Self-Driving Cars

250W is just 1 board's power? Full-Autonomous requires at least 2 PX2s so it will be 500W?

But it's a relief that the real world impact isn't large.

 

[henderrj]

All new Teslas are equipped with NVIDIA’s new Drive PX 2 AI platform for self-driving

Wow, that's some processing power!