Volvo Partners with NVIDIA and will Integrate Its Drive AGX Xavier Into Cars In 2020

[dmartin]

NVIDIA has long worked in the field of artificial intelligence applied to self-driving, and the fruits of that work are beginning to appear. Volvo has announced an alliance with NVIDIA whereby the carmaker will use NVIDIA'S Drive AGX Xavier computer in its next generation of vehicles. This system allows level 4 autonomy in controlled areas, although initially they will restrict their capacity to "Level 2+" that will more or less bring it up to the level offered by the current Tesla models.

The first Volvo cars to integrate this system will be available in the early 2020s, and will not only monitor the vehicle's surroundings but will also follow the driver's head and eye movements to detect possible events that the car's sensors have failed to capture. Volvo is not the first to reach such an agreement with NVIDIA: the company has already reached an agreement with Volkswagen earlier this year to take advantage of its Drive IX platform, and other customers such as Uber and Daimler have also made use of NVIDIA solutions in this field.



View at TechPowerUp Main Site

 

[claylomax]

Where's Jensen's leather jacket?

 

[GreiverBlade]

Where's Jensen's leather jacket?

he didn't get enough customer @$$ skin from the RTX pre order to make a new one? err... sorry i couldn't pass on that ...

uh? it's still a thing? i thought that project did a big flop ... i must be out of the loop ...

 

[OutThereSomewhere]

Why not make Electric cars the main priority instead of all the spyware?
Bloody choking from the smog here, and Nvidia's idea of using the driver as a 1st person gamer won't help

 

[CheapMeat]

Is there any info on how they reinforce or make these system more durable? Curious how many layers the PCB has, how the heatsinks are on, etc.

 

[notb]

uh? it's still a thing? i thought that project did a big flop ... i must be out of the loop ...

You though a project called "autonomous cars" did a flop? Sleep much lately?

 

[Vayra86]

You though a project called "autonomous cars" did a flop? Sleep much lately?

No he's spot on:

https://www.inverse.com/article/483...till-a-partner-despite-in-house-autonomy-chip

Drive PX hasn't exactly been making waves.

 

[notb]

No he's spot on:

https://www.inverse.com/article/483...till-a-partner-despite-in-house-autonomy-chip

Drive PX hasn't exactly been making waves.

I don't understand the relevance. This article is about Tesla. The topic is about Volvo....

PX is currently being implemented by 5 car makers: Tesla, Toyota, Mercedes, Volvo and Audi.
Tesla is by far the smallest one of the bunch.
It's no surprise they're thinking about an own AI chip, when they spend so much on AI R&D.

 

[FordGT90Concept]

Why not make Electric cars the main priority instead of all the spyware?

Because batteries suck and the electrical grid can't handle a huge surge of draw.

I don't understand the relevance. This article is about Tesla. The topic is about Volvo....

Drive PX could only analyze 20 frames per second. Tesla developed a solution that could analyze 2000 frames per second. Tesla likely made an ASIC that's lower power and better performing.

Tesla basically just has a bunch of Drive PX systems they aren't using anymore. The same could be true of the other manufacturers. By market cap, Tesla is the largest.

Volvo is a late comer to the AI scene. Volvo may eventually come to the same realization that Tesla did.

 

[notb]

Because batteries suck and the electrical grid can't handle a huge surge of draw.

Maybe in US (because of larger distances and well-known infrastructure problems).
In Europe, electric cars already work perfectly well for commuting, i.e. when charging at night is enough.

Capacity is not a problem anymore. Electric range matched petrol, at least it could. Manufacturers are forced to compromise between range and charging time, so the electric range is usually lower than it could be.
Renault Zoe (electric) and Twingo (petrol) are fairly similar cars aside from the propulsion. Twingo does 500km per tank (in a city), Zoe 400km per charge.

Charging is the big problem. Until we have a solution, electric cars will remain niche.
We need around 100km of range per 1 minute of charging, to make electric cars useful for long journeys. Today it's around 20km/min.
But you know... it's not that hard to imagine a workaround. Few batteries in a car can be charged at the same time. So the actual bottleneck is in the power supply on stations. ;-)

 

[Vayra86]

Maybe in US (because of larger distances and well-known infrastructure problems).
In Europe, electric cars already work perfectly well for commuting, i.e. when charging at night is enough.

Capacity is not a problem anymore. Electric range matched petrol, at least it could. Manufacturers are forced to compromise between range and charging time, so the electric range is usually lower than it could be.
Renault Zoe (electric) and Twingo (petrol) are fairly similar cars aside from the propulsion. Twingo does 500km per tank (in a city), Zoe 400km per charge.

Charging is the big problem. Until we have a solution, electric cars will remain niche.
We need around 100km of range per 1 minute of charging, to make electric cars useful for long journeys. Today it's around 20km/min.
But you know... it's not that hard to imagine a workaround. Few batteries in a car can be charged at the same time. So the actual bottleneck is in the power supply on stations. ;-)

Charging is just the least of electric driving's problems. (BTW I thought this was about autonomous driving but OK )

When electric cars gain traction, the cost of both electricity and the batteries themselves will skyrocket. I think none of us can really predict how much, but it will be tremendous. Its already visible in battery land. No matter what solution you choose (battery lease, swap-in / swap-out systems, etc.), batteries do wear out and need replacing. A petrol tank does not.

 

[FordGT90Concept]

Maybe in US (because of larger distances and well-known infrastructure problems).
In Europe, electric cars already work perfectly well for commuting, i.e. when charging at night is enough.

Capacity is not a problem anymore. Electric range matched petrol, at least it could. Manufacturers are forced to compromise between range and charging time, so the electric range is usually lower than it could be.
Renault Zoe (electric) and Twingo (petrol) are fairly similar cars aside from the propulsion. Twingo does 500km per tank (in a city), Zoe 400km per charge.

Charging is the big problem. Until we have a solution, electric cars will remain niche.
We need around 100km of range per 1 minute of charging, to make electric cars useful for long journeys. Today it's around 20km/min.
But you know... it's not that hard to imagine a workaround. Few batteries in a car can be charged at the same time. So the actual bottleneck is in the power supply on stations. ;-)

USA consumes 9,000,000 barrels of gasoline (45.8 MJ/kg) and 3,000,000 barrels of diesel (45.3 MJ/kg) per day. 42 US gallons (0.1589873 m³) per barrel. Gasoline has a density of 737 kg/m³ and Diesel has a ballpark density of 900 kg/m³. Put that all together and we get :

Gasoline: 9,000,000 b * 0.1589873 m³/b * 737 kg/m³ * 45.8 MJ/kg = 48,298,974,449 MJ / 86,400 = 559,016 Mw
Diesel: 3,000,000 b * 0.1589873 m³/b * 900 kg/m³ * 45.3 MJ/kg = 19,445,736,663 MJ / 86,400 = 225,066 Mw

To change all vehicles (car, truck, train, lawn mowers, etc.) to electric, USA would have to add roughly 784 Gw of power generation capacity. USA had 1,074.3 Gw as of 2016. This chart? Almost double it (need minimum 73% more capacity):

Hydrogen (using aluminum catalyst) fuel cells are the goal, not electric. In the interim, gasoline/diesel make a lot more sense than lithium-ion batteries.

 

[notb]

Charging is just the least of electric driving's problems. (BTW I thought this was about autonomous driving but OK )

What are the bigger problems then?

I prefer AI topic, but somehow it always moves to electric cars. No idea why...
So to keep it on topic, what's the problem with Tesla and PX? And why is it so important? Tesla is such a tiny car maker. Surely, a prestigious one for Nvidia, but not that relevant financially.

When electric cars gain traction, the cost of both electricity and the batteries themselves will skyrocket. I think none of us can really predict how much, but it will be tremendous. Its already visible in battery land. No matter what solution you choose (battery lease, swap-in / swap-out systems, etc.), batteries do wear out and need replacing. A petrol tank does not.

It could be true for batteries. For electricity? Nope. We'll use more electricity, but less petrol. Petrol can be easily processed into electricity (not so easy the other way). Electric-based motorization will be more efficient. So the whole energy cost should become lower.

USA consumes 9,000,000 barrels of gasoline (45.8 MJ/kg) and 3,000,000 barrels of diesel (45.3 MJ/kg) per day. 42 US gallons (0.1589873 m³) per barrel. Gasoline has a density of 737 kg/m³ and Diesel has a ballpark density of 900 kg/m³.

Wrong approach. You started thinking about totals (extensive approach) and got overwhelmed by the huge numbers. Of course we can't instantly build dozens of new power plants and we can't replace all electric lines. But that's hardly necessary.

Here's a simpler, intensive model.

1) We have a combustion car. It runs on petrol.
2) We replace it with a hybrid. It runs on electricity, but you still fill it with petrol. Each car is a tiny power station.
3) We move the power production to the petrol station. So now you fill the car with electricity. This electricity is produced on the petrol station, using a generator.
4) Petrol station owner decides he can buy electricity cheaper than making it itself. He contracts a power supply company, aka a power plant.

As you can see, I haven't mentioned any alternative sources. The whole electricity that cars consume in this model is made of petrol.
However, with each step (2 -> 3 -> 4) the system becomes more effective. So in the end we use less petrol for the same energy consumption.
More importantly, until step 3 there are no dramatic changes to how the supply chain looks. We still have stations, tank trucks, refineries etc.
There are also no big investments needed. The only thing that happens is a petrol station has to buy a generator. Instead, it saves money on distributors, safety measures etc.

 

[Vayra86]

What are the bigger problems then?

I prefer AI topic, but somehow it always moves to electric cars. No idea why...
So to keep it on topic, what's the problem with Tesla and PX? And why is it so important? Tesla is such a tiny car maker. Surely, a prestigious one for Nvidia, but not that relevant financially.

Not saying its a major problem, just responding to @GreiverBlade 's comment about 'that project did a big flop'. Initially, Nvidia was confident it would take over the market and that their PX system was superb and that they were leading, cutting edge. The real world underlined that PX needs a LOT more performance to become useful. Tesla is just early in the race so its the first to notice, but it is telling that a car maker can surpass (and not just a little) performance of such a specific component versus a giant like Nv.

 

[GreiverBlade]

Tesla is such a tiny car maker. Surely, a prestigious one for Nvidia, but not that relevant financially.

except, if you did read, Tesla doesn't need Nvidia at all:
"Drive PX could only analyze 20 frames per second. Tesla developed a solution that could analyze 2000 frames per second. Tesla likely made an ASIC that's lower power and better performing. " and "Tesla basically just has a bunch of Drive PX systems they aren't using anymore."

but also:
"Volvo is a late comer to the AI scene. Volvo may eventually come to the same realization that Tesla did."
which is : Nvidia is not needed, hence : flop

i wasn't talking about self driving vehicle in my initial post PX is a project, SDAV aren't ...

Not saying its a major problem, just responding to @GreiverBlade 's comment about 'that project did a big flop'. Initially, Nvidia was confident it would take over the market and that their PX system was superb and that they were leading, cutting edge. The real world underlined that PX needs a LOT more performance to become useful.

well ... yes indeed PX is a flop in the end.

 

[Vayra86]

Well, things can change. But PX needs work, that much is certain. As do all autonomous drive systems as of today.

 

[GreiverBlade]

well it's kinda pointless since Tesla multiplied their(Nvidia PX's) actual output by 100 ... thus initialy a flop that can get back by getting 100x more frame analysis while consuming less power than the concurrent and cost less (fat chance )

tho not really a fan of SDAV either ... too dangerous ... (situational i agree ... ahahaha : "a baby in the middle of the road, a wall on you right, a cliff on your left, who will the AI choose to kill ... you or the baby?" that kind of extreme is still quite relevant, isn't it? )

 

[notb]

Not saying its a major problem, just responding to @GreiverBlade 's comment about 'that project did a big flop'. Initially, Nvidia was confident it would take over the market and that their PX system was superb and that they were leading, cutting edge. The real world underlined that PX needs a LOT more performance to become useful.

Disagree. Latest PX is said to be fast enough for level 5. The other companies I've mentioned have no problems with implementation. And it's not like Tesla could make something faster or more efficient than Nvidia does. Few IT companies have tried already.

Keep in mind Tesla is mostly a design company. They make some of their batteries and they put the cars together from parts. Almost everything else is outsourced. But they also spend a lot on AI research. So clearly, if there's one other thing they could make (design) for their cars, it's the AI system. It was fairly obvious from the start that the Nvidia partnership could break. It'll also help their margins, because PX is quite expensive.

It's different with other companies. Compared to Tesla, they make a lot more of the car themselves. Hence, they're less likely to invest a lot into AI chips, when there are at least few great providers.

Who knows, maybe at some point Tesla becomes an AI chip provider rather than car maker? There's no reason why they wouldn't sell both, anyway.

well it's kinda pointless since Tesla multiplied their(Nvidia PX's) actual output by 100 ...

No offense, but you know this is impossible, right?
It's just a processor (general or ASIC) underneath. What you said basically means that Tesla could make AI servers 100 times faster than Nvidia can. I'd imagine they could dominate GPU market as well. You still think that's realistic? How are they doing it? They have a secret 1nm semiconductor plant?

I don't want to play a Tesla hater, but the "2000 fps" figure is from Musk's tweet, isn't it? These tweets haven't been the most reliable source of information, as we all know.
The truth about Tesla is that they've already changed the AI partner once - in 2016. Afterwards, Mobileye did say a few words about how Tesla is doing their business.

You're also wrong about Volvo. They've been the leader of AI for years. Their systems are easily one of the best you can find in cars today, which is really hard to miss. The only reason why Tesla could beat Volvo to Level 5 is that... well, it's EU - Sweden in particular. Volvo will make a million tests and countless rubber moose will be killed before they launch the product. It's a bit different in US.

 

[FordGT90Concept]

Wrong approach. You started thinking about totals (extensive approach) and got overwhelmed by the huge numbers. Of course we can't instantly build dozens of new power plants and we can't replace all electric lines. But that's hardly necessary.

Here's a simpler, intensive model.

1) We have a combustion car. It runs on petrol.
2) We replace it with a hybrid. It runs on electricity, but you still fill it with petrol. Each car is a tiny power station.
3) We move the power production to the petrol station. So now you fill the car with electricity. This electricity is produced on the petrol station, using a generator.
4) Petrol station owner decides he can buy electricity cheaper than making it itself. He contracts a power supply company, aka a power plant.

As you can see, I haven't mentioned any alternative sources. The whole electricity that cars consume in this model is made of petrol.
However, with each step (2 -> 3 -> 4) the system becomes more effective. So in the end we use less petrol for the same energy consumption.
More importantly, until step 3 there are no dramatic changes to how the supply chain looks. We still have stations, tank trucks, refineries etc.
There are also no big investments needed. The only thing that happens is a petrol station has to buy a generator. Instead, it saves money on distributors, safety measures etc.

#3 is more expensive than #1. #4 is cheap because the inputs are cheap (natural gas, coal, wind, nuclear). See how small the black region is on the graph? Additionally, power plants are able to capture much more heat and put it to work where reciprocating engines lose most of the potential energy by way of heat.

Hybrids have been a market failure (2% and falling). High cost, little benefit. Cheaper and better driving experience to stick with gasoline/or diesel.

For the record, carbon dioxide is recycled by plants. There is no profit in recycling lithium-ion batteries (only thing of value is trace amounts of cobalt). Diesel/gasoline is more environmentally sustainable.

 

[OutThereSomewhere]

Because batteries suck and the electrical grid can't handle a huge surge of draw.

Solar power, lots of solar power. Our environment demands it.

 

[FordGT90Concept]

Solar power has an average power factor of 0.33% in the USA. For every 1 Gw installed, on average, you'll only get 333 Mw out. Solar is the least efficient power source there is.

 

[notb]

Solar power has an average power factor of 0.33% in the USA. For every 1 Gw installed, on average, you'll only get 333 Mw out. Solar is the least efficient power source there is.

That's electrical efficiency, not financial one. Solar power has very low costs of maintenance. Installation is quite expensive, but we may be able to lower that further.
Hydroelectricity is the best kind: it's very efficient, cheap and more environment-neutral than burning anything (although far behind solar and wind). But not everyone has a river nearby and even if there is one, you need a ton of checks and permissions before you can build a power plant (even a tiny one).

#3 is more expensive than #1.

The model is sequential: I gave 4 steps, not 4 alternatives.

#4 is cheap because the inputs are cheap (natural gas, coal, wind, nuclear). See how small the black region is on the graph?

But the graph only includes power plants - not the whole energy produced!
You said yourself that if we included cars, fuel contribution would explode. Although you made a mistake by just using energy density. Car engines are notoriously inefficient (~40%).

If you replaced all cars with hybrids, you would already have that petrol-made electricity. Then you could just take out those engines and run them in the same room. Bang. You have an oil power plant. Not a very good one, since one huge "engine" would be better than millions of small ones. But... it's already more efficient than if we kept those engines in cars.
We're not adding anything to the system. We're just changing the place where oil burning happens.

Hybrids have been a market failure (2% and falling). High cost, little benefit. Cheaper and better driving experience to stick with gasoline/or diesel.

That's a cultural thing. Hybrids and EVs are more popular on other continents. Europe and Asia are way more congested and densely populated. We also drive smaller cars in general and we value finesse (like: handling, interior quality, noise) over raw horsepower.
Hybrids' market share is in fact falling in many countries, but that's simply because EVs are taking over part of the market. Obviously, a hybrid owner is more likely to try an EV than someone used to petrol.
I don't know the particular US case, though. You guys are just falling back to petrol in general. This is not happening in other large markets.
BTW: you're posting old data. Hybrids and EVs are booming. We already have data for 2018Q2.

For the record, carbon dioxide is recycled by plants.

The question is: how much? Most CO2 processes in oceans, not by land plants!
Earth is a stable ecosystem. We're changing it by adding more CO2 sources, but we don't increase the amount of CO2 "consumers".

Anyway, the thing about EVs is not about how much petrol we consume and how much CO2 is made. As you can see, I proposed a model where not much changes in general (other than improved efficiency). It's about where it happens.
The real problem with cars is that they output CO2 exactly in the same place people use them, i.e. congested cities and so on.
And the higher density of people you have, the higher is the CO2 production. It's a total opposite of what we should want.
The idea behind EVs is that the CO2 production can be moved far from high density population.

Anyway, the topic was about AI and I prefer it over cultural difference and car ownership. Let's end it there.

 

[OutThereSomewhere]

Solar power has an average power factor of 0.33% in the USA. For every 1 Gw installed, on average, you'll only get 333 Mw out. Solar is the least efficient power source there is.

It'll take time but it will evolve.

 

[FordGT90Concept]

That's electrical efficiency, not financial one. Solar power has very low costs of maintenance. Installation is quite expensive, but we may be able to lower that further.

Solar power is the heaviest subsidized of them all and even then, it has the least installations out there. It's financially inefficient too. Solar operates on the basic principle of surface area, surface area has land value and is taxed. Land value tends to increase with age while profitability doesn't. So not only is the land itself high maintenance, the plant itself requires a lot of maintenance too. There's a crapload of motors to keep collectors/reflectors pointed at the sun and they do fail and need replacing. The panels/reflectors also have to be kept clean. If the facility has batteries as a reserve, they need to be replaced every 3-5 years. You are right that maintenance is cheapest on solar PV but because of its low capacity factor (as I pointed out before) it is still relatively expensive (natural gas is cheaper when you take into consideration the subsidies solar gets that natural gas doesn't). O&M = Operations & Maintenance. They break it down into fixed (paying people to fix stuff) and variable (fuel costs):

Hydroelectricity is the best kind: it's very efficient, cheap and more environment-neutral than burning anything (although far behind solar and wind). But not everyone has a river nearby and even if there is one, you need a ton of checks and permissions before you can build a power plant (even a tiny one).

Hydroelectric capacity is being removed, not added. Everywhere that is reasonable to dam already is. There isn't going to be a major expansion in hydroelectric power mix in the grid.

But the graph only includes power plants - not the whole energy produced!
You said yourself that if we included cars, fuel contribution would explode. Although you made a mistake by just using energy density. Car engines are notoriously inefficient (~40%).

Here's a graph for all energy consumed in the USA:

I made no mistake. If all vehicles are changed to electric, demand for petroleum falls and demand for electricity rises. Electricity production growth would mostly be from burning more natural gas. It fundamentally boils down to watts and USA would have to add 73% more electricity capacity to provide electricity to all the vehicles if there were electric. It's displacement.

Batteries are notoriously inefficient too. Power plants that use a steam cycle usually have efficiency greater than 60%.

If you replaced all cars with hybrids, you would already have that petrol-made electricity. Then you could just take out those engines and run them in the same room. Bang. You have an oil power plant. Not a very good one, since one huge "engine" would be better than millions of small ones. But... it's already more efficient than if we kept those engines in cars.

It is not because the vehicle has more mass to move. The net effect is lower power to weight ratio but slightly increased efficiency because of regenerative braking. You'll never recover the installation cost from fuel savings though when comparing an all electric hybrid to a equal-displacement petroleum fueled engine. Most of the fuel savings comes from having a smaller displacement engine.

The question is: how much? Most CO2 processes in oceans, not by land plants!
Earth is a stable ecosystem. We're changing it by adding more CO2 sources, but we don't increase the amount of CO2 "consumers".

I was wrong. A paper published on October 1 suggests planets are hardening themselves against carbon dioxide. Hydrogen fuel cells are still a better goal than all electric (again, because batteries suck).

The idea behind EVs is that the CO2 production can be moved far from high density population.

Which translates to power loss through transmission. Hydrogen fuel cells almost completely eliminate transmission costs because water in -> water out if aluminum nanoparticles work indefinitely as a catalyst.

It'll take time but it will evolve.

Efficiency will improve some but best case scenario (100% efficiency) is ~250 w/m2 which is very, very low. Solar is not the future.