NVIDIA Tegra Architecture

[silent_guy]

Are NVidia really a different source when they are using the same source as everyone else just trying to pass it off as their own? As far as I can see all those Tegra 3, 4 and K1 automotive solutions are not really NVidia technology nor NVidia software. It’s all Mobileye’s EyeQ3 hardware with Mobileye’s software run via MIPS CPU’s not Tegra. It looks to me like NVidia are just using marketing PR to pass off someone’s else automotive solutions as their own. Please correct me if I am wrong. How can they say Tegra automotive when the automotive stuff is not done on the Tegra chip or Tegra software?

There is no way that any of their recent chips (TK1 and later) are already in current designs. And those are the ones that are needed to run the neural nets.
So all Tegras in current cars on the road are not for self-driving tech but for the audio-visual aspects: dashboard, multi-media etc. I don't think they have ever claimed otherwise, so saying that they are passing somebody's technology as their own is not correct.

And doing the audio-visual stuff doesn't make it any less automotive either: it still requires more rigorous quals than regular mobile.

It's only logical that they are focusing their current marketing on the self-driving stuff, because today is when tech solutions are decided for cars that will be on the road 5 years from now.

 

[pharma]

Nvidia wants to leap from video games to self-driving cars

It took Silicon Valley chipmaker Nvidia Corp the better part of a decade to gain acceptance as a global automotive supplier, a lesson for other technology firms hoping to make a similar transition from consumer electronics to car components.

Nvidia had a thriving business in supplying powerful graphics processors for video game consoles and laptop computers when representatives of Volkswagen AG (VOWG_p.DE) asked if it could adapt the same realistic, three-dimensional displays for the dashboards of Audi luxury cars.

That was 10 years ago – several lifetimes in Silicon Valley terms, but just a couple of model changes ago for the big car maker. It took six years for Nvidia to develop the chips to power the 3-D navigation system display that launched on the 2011 Audi A8, the brand’s top-of-the-line sedan, compared with three to four years to develop a new chip for gaming applications.

Nvidia’s success in landing processors on cars including the Audi A8 and the Tesla Model S sedan illustrates the growing willingness among automakers to look beyond their traditional suppliers to obtain a technological edge.

“They offer us computing systems with the strength” to process vast amounts of data required by new driver assistance systems and displays, said Ulrich Hackenberg, Audi’s head of technical development.

“We had to be taught what ‘automotive grade’ meant,” recalls Danny Shapiro, Nvidia’s senior director of automotive operations. In simple terms, that meant no reboots – because a dashboard screen that fails to light up means an angry customer and possibly an expensive warranty repair.

Nvidia has made the leap to infotainment systems in vehicles, but not yet to self-driving cars.

The company had just $23 million in sales in the automotive segment in 2011. Growth has surged since. This year, the company expects automotive revenue of $183 million, and says it has booked more than $2 billion in future automotive business, most of that involving chips for digital displays and infotainment. By 2020, Nvidia expects to have chips on more than 32 million vehicles.

Nvidia is still a niche player with a long way to go to crack the top league of traditional automotive chip suppliers. Much larger companies, including Japan’s Renesas Electronics Corp and U.S. chipmakers Texas Instruments Inc, Intel Corp and Qualcomm Inc, dominate the automotive chip business.

Nvidia's automotive sales represented just 4 percent of the company's $4.7 billion annual revenue in its most recent fiscal year. That in turn is just one-tenth of the $1.86 billion in automotive semiconductors that Texas Instruments sold last year.

The company is pushing to crack the emerging market for driver assistance systems, which include such tasks as self-parking and semi-automated steering and braking. Those systems require huge amounts of computing power in very small packages. Nvidia has developed and begun selling a chip called the Tegra X1 for automotive and gaming uses, which it says can put the power of a supercomputer in a package about the size of a postage stamp.

“Nvidia brings unparalleled graphics capabilities that could prove critical building blocks” for driver assistance systems, said Morgan Stanley analyst Joseph Moore.

http://www.reuters.com/article/2015...20150514?feedType=RSS&feedName=technologyNews

 

[Alexko]

" It took six years for Nvidia to develop the chips to power the 3-D navigation system display that launched on the 2011 Audi A8, the brand’s top-of-the-line sedan, compared with three to four years to develop a new chip for gaming applications."

Since when does NVIDIA develop chips specifically for cars?

 

[Deleted member 13524]

" It took six years for Nvidia to develop the chips to power the 3-D navigation system display that launched on the 2011 Audi A8, the brand’s top-of-the-line sedan, compared with three to four years to develop a new chip for gaming applications."

Since when does NVIDIA develop chips specifically for cars?

Since tablet and phone makers don't want them.

Hey.. t-this was made from the ground-up for cars, I promise!

 

[silent_guy]

I apologize for raining a bit of your satisfaction of you knowing better, but if you look at it from the point of view of the business journalist who's writing these things and his audience, then it's not an unreasonable thing to write at all.

They don't understand the individual steps of that are needed to get from initial idea to car-with-chip for sale at the dealership. And then don't care because it doesn't matter.

What matters for them is that there is a huge lead time and tons of development to get to that final point. The fact that it's silicon that already existed in some other products is a minor detail. In terms of development timeline, the time from idea to silicon may well be smaller than the time from silicon to car at the dealership.

And yes, to call it 'to develop a chip' is a little inaccurate, but, again, that only matters if you want to be pedantic about it, and 'to develop a chip, and use it in another product first, and the spend years to certify it for the harsh conditions of the automobile world' is just too much of a mouthful that would bore the suits to death.

 

[Alexko]

I apologize for raining a bit of your satisfaction of you knowing better, but if you look at it from the point of view of the business journalist who's writing these things and his audience, then it's not an unreasonable thing to write at all.

They don't understand the individual steps of that are needed to get from initial idea to car-with-chip for sale at the dealership. And then don't care because it doesn't matter.

What matters for them is that there is a huge lead time and tons of development to get to that final point. The fact that it's silicon that already existed in some other products is a minor detail. In terms of development timeline, the time from idea to silicon may well be smaller than the time from silicon to car at the dealership.

And yes, to call it 'to develop a chip' is a little inaccurate, but, again, that only matters if you want to be pedantic about it, and 'to develop a chip, and use it in another product first, and the spend years to certify it for the harsh conditions of the automobile world' is just too much of a mouthful that would bore the suits to death.

Or the journalist just asked questions to an NVIDIA representative, and republished everything they were told unquestioningly. Or perhaps they were told something accurate and misunderstood, but made no effort to verify anything.

 

[pharma]

Or the journalist just asked questions to an NVIDIA representative, and republished everything they were told unquestioningly. Or perhaps they were told something accurate and misunderstood, but made no effort to verify anything.

I thought there were quotes in the article from Audi technical representatives as well, but I may have been mistaken. Maybe you're right in that they did not try to verify anything Nvidia mentioned....

 

[Ailuros]

I know it's somewhat OT but I have read a bit around about Mobileye's EyeQ4 solution and that's truly a chip designed for automotive. The VMPs of the Q4 are specifically designed for ADAS and therefore dedicated processing units and not a garden variety GPU. Marketing or not: no wonder Mobileye claims equivalent performance of a 2.5 TFLOPs supercomputer with the Q4 for ADAS.

 

[silent_guy]

Or the journalist just asked questions to an NVIDIA representative, and republished everything they were told unquestioningly. Or perhaps they were told something accurate and misunderstood, but made no effort to verify anything.

And if Nvidia would have told that that it took 6 years to develop from idea to a chip ready for sale in a car, they'd still be right. Pure silicon is worthless. The whole package matters: silicon, software, reference designs, certification reports. And everybody did their job: adjust the message to the target audience of suits instead of bunch of geeks behind a computer who are trying to paint every uttered phrase into an evil conspiracy.

 

[Erinyes]

Some news on Parker. My information is that it is still on TSMC 16nm(FF/FF+?) and they have not shifted to Samsung as rumoured. It is also expected to tape out sometime this month..or has possibly already taped out since my information is slightly dated. If this is correct, we're looking at availability in late Q1'16, at best. My source indicated that it has Denver cores and not Cortex A72.

Depends. In the automotive space it's rare for a silicon vendor to deal with the car vendor directly. There's usually a module supplier/ODM in the middle. So it all depends if they're also gearing up to be their own module supplier as well. That's a massive cost to take on, because the module supplier takes on the burden of certification and fit-for-purpose proofs, longevity testing, pre-physical integration, etc.

Support and software is therefore structured differently because of the middle man aspect. It's not going to be large amounts of revenue for NV to book if they're not the module supplier too, but being the module supplier isn't anything you can do in a short period of time.

It is possible they are trying or that they are already supplying modules. I have seen development kits and Audi dashboard/cabin mockups at Nvidia sometime in 2010 AFAIK. It is also entirely possible that it was just for software development/testing though.

There is no way that any of their recent chips (TK1 and later) are already in current designs. And those are the ones that are needed to run the neural nets.
So all Tegras in current cars on the road are not for self-driving tech but for the audio-visual aspects: dashboard, multi-media etc. I don't think they have ever claimed otherwise, so saying that they are passing somebody's technology as their own is not correct.

And doing the audio-visual stuff doesn't make it any less automotive either: it still requires more rigorous quals than regular mobile.

It's only logical that they are focusing their current marketing on the self-driving stuff, because today is when tech solutions are decided for cars that will be on the road 5 years from now.

Exactly..the chips found on cars today should be Tegra 3 powered at best since AFAIK Tegra 4 lacked the required bus support for automotive.

As you have noted, the design cycles and certification requirements in the automotive industryare long so current gen SoCs like X1 will not be found in production vehicles until a few years from now.

" It took six years for Nvidia to develop the chips to power the 3-D navigation system display that launched on the 2011 Audi A8, the brand’s top-of-the-line sedan, compared with three to four years to develop a new chip for gaming applications."

Since when does NVIDIA develop chips specifically for cars?

AFAIK that was a Tegra 2 chip..which was not specifically developed for cars. Given that it taped out sometime in 2009..six years sounds about right (silent_guy's post explains this well)

 

[Samwell]

Some news on Parker. My information is that it is still on TSMC 16nm(FF/FF+?) and they have not shifted to Samsung as rumoured. It is also expected to tape out sometime this month..or has possibly already taped out since my information is slightly dated. If this is correct, we're looking at availability in late Q1'16, at best. My source indicated that it has Denver cores and not Cortex A72.

If your info is correct and you mostly have good info, then we have a new question. What does Nvidia want to manufactur at Samsung if it's not Parker? They mentioned Samsung as manufacturer in their SEC-filing for the first time this year so i doubt it's only because of test wafers.

 

[Ailuros]

If your info is correct and you mostly have good info, then we have a new question. What does Nvidia want to manufactur at Samsung if it's not Parker? They mentioned Samsung as manufacturer in their SEC-filing for the first time this year so i doubt it's only because of test wafers.

Even further: why the heck does it take so long? I personally expected Parker earlier, but shifted it in my mind to somewhere 2016 because of the supposed switch to 14FF.....(<----me confused as hell....)

 

[pharma]

Nvidia: We’ve learnt a lot from the automotive industry
May 16, 2015

Cars nowadays require a lot of general-purpose processing horsepower and visual processing capabilities, something that Nvidia can deliver thanks to its expertise in GPUs and mobile application processors. However, it is impossible to just flip a mobile SoC into a car, something that has taken years for Nvidia to learn.

Nvidia cannot just install its Tegra into an Audi, it has to design the chip for maximum reliability and create software that works perfectly 100 per cent of time. In a bid to be inside cars, Nvidia needed to rethink its development cycles (cars are updated once in a few years, new smartphones are released every year) and design chips in accordance with new requirements.

In a bid to catch up with rivals, Nvidia wants to power self-driving systems inside future cars with extremely advanced chips. The company has excellent graphics processing architectures, which can be used for highly-parallel workloads like image recognition, as well as powerful general-purpose processors for a variety of serial tasks. In the future, Nvidia will probably have to integrate ARM Cortex-R-series cores for real-time computing tasks into its Tegra chips or Drive platforms. There are a lot of other technologies that Nvidia will need to license or build in house for its future solutions for the automotive industry. But the company seems to be ready for a long journey, which is only starting.

KitGuru Says: By discussing its automotive initiatives without revealing any new details Nvidia wants to make it clear for investors that it is absolutely serious about its car chip business. From technology point of view it means that Tegra SoCs will be tailored for cars, not for mobile applications.

http://www.kitguru.net/components/a...ve-learnt-a-lot-from-the-automotive-industry/

 

[Exophase]

Nvidia is massively behind its competitors in the market of chips for automobiles. Nvidia’s automotive sales represented just 4 per cent of the company’s $4.7 billion annual revenue in its most recent fiscal year, according to Reuters. By contrast, sales of automotive semiconductors totalled $1.86 billion at Texas Instruments last year.

So, $188m revenue with 7.5m Tegra chips in cars (http://venturebeat.com/2015/02/11/n...ers-and-the-cloud-but-not-mobile-tegra-chips/), or an average of $25 per chip. I was expecting a much higher margin market than this, with each chip coming with a relatively lucrative full platform sale, including both hardware and software.

 

[AlNom]

So what's a reasonable target TDP/SoC for a car? Tablet?

 

[wco81]

Where are these 7.5 million cars? Specific models and the packages which use Tegras?

 

[Samwell]

So, $188m revenue with 7.5m Tegra chips in cars (http://venturebeat.com/2015/02/11/n...ers-and-the-cloud-but-not-mobile-tegra-chips/), or an average of $25 per chip. I was expecting a much higher margin market than this, with each chip coming with a relatively lucrative full platform sale, including both hardware and software.

It's 4 million for last year. 7.5million is the total number of cars ever shipped with Tegra. So we are at 45$ for Tegra3 SoCs. That seems a good margin for me.

@wco81
Half of Audis cars use Tegra, also Volkswagen is using it. So the biggest number is coming from one company. Additionally BMW and from this Year Honda in the Civic.

 

[Rys]

If it's really 4M for last year, that feels healthy as a business. VAG as a group only sell ~10M cars a year.

@AlNets: while it's tempting to think that automotive modules might allow for larger design powers because they have the room for active cooling, and in some cases that's true, the chip designer still needs to make a design suitable for passive installations. While there are some chip designs that rely on overvolting to bin into quite high-end, actively cooled SKUs, design powers tend to be not too far away from tablets in the main.

 

[ntclick]

It's probably tricky to get any sort of bleeding edge chips into cars purely on safety grounds. Tegra 3 is fairly modern on car maker timescales...

 

[Ailuros]

It's probably tricky to get any sort of bleeding edge chips into cars purely on safety grounds. Tegra 3 is fairly modern on car maker timescales...

Compared to R Car H2, OMAP5, iMX6 and the likes? Probably a matter of perspective. Tegra4 would be a completely different chapter but T3? The only thing that's probably saving them is their robust sw advantage for a T3 in some cases.