[Ars] Tegra 4i benchmarks are in!

[xpea]

And why do you think the kind of visual computing the manufacturers are putting in a car aren't commodities? What does Tegra brings to the table that Intel, Mediatek, Qualcomm and others won't be able to match in terms of hardware capabilities if they decide to exploit those same markets? A car is a much less constrained environment than, let's say, a tablet, and car software isn't above what an android tablet or windows convertible are going to offer you, and volumes are fine as long as you do not have Globalfoundries as your main foundry partner.... what are your constraints on this one? So yes, it is commoditizable.

It's crazy to see that they are still so many people that don't understand. Nvidia is a mainly a software company (yeah, at nvidia they are much more people working on software than hardware). This is their strength.
For Tegra and automotive, their advantage is Visionworks and more specifically ADAS:
http://www.nvidia.com/object/visionworks-cv-toolkit.html
good luck for mediatek to bring an equivalent to the table

 

[mrmt]

Oh it will need an embedded SoC to drive the display and decode, yes, but what do you think the margins are on an A53 core and a h.264 decoder?

If you are thinking Ford Fiesta, then yes, you are right. But think Audi, BMW, Mercedez, Cadilac... Do you really think manufacturers will try to save $10 on the SoC?

 

[teejee]

If you are thinking Ford Fiesta, then yes, you are right. But think Audi, BMW, Mercedez, Cadilac... Do you really think manufacturers will try to save $10 on the SoC?

I work in car industry, saving 10$ on one single component is a huge amount even for Mercedes.
The industry is under extreme price/profit pressure since long time ago.
And they buy stuff like this according to requirements, not according to what engineers thinks is cool.
Nvidia will have exactly the same issue they've had with smartphones: How to convince a purchase department to pay more than they have to?

 

[sontin]

What Nvidia is telling you is that automotive margins will be higher than their margins, which is a true statement. What they didn't tell you is that their *current* margins on consumer Tegra is far below what they used to get with Tegra 2 and even sometimes Tegra 3 (when Tegra margins were ABOVE corporate average, as stated in a Q&A)
In other words, margins on their consumer business are so low that an embedded business suddenly became more profitable than sell bleeding edge devices on the consumer market. This is the elephant in Nvidia's room.

Just no.
nVidia's coperate margins are around 53%. And more than 50% of the revenue comes from their Geforce-Business.
Tegra's Margins are under the coperate average with the automobile on the upper end.

And why do you think the kind of visual computing the manufacturers are putting in a car aren't commodities? What does Tegra brings to the table that Intel, Mediatek, Qualcomm and others won't be able to match in terms of hardware capabilities if they decide to exploit those same markets? A car is a much less constrained environment than, let's say, a tablet, and car software isn't above what an android tablet or windows convertible are going to offer you, and volumes are fine as long as you do not have Globalfoundries as your main foundry partner.... what are your constraints on this one? So yes, it is commoditizable.

Neither of them is able to match nVidia's experience in visual graphics. If you think this would be so easy why is Intel not going after another $6 billion market with more than 50% margins?

 

[blackened23]

It's crazy to see that they are still so many people that don't understand. Nvidia is a mainly a software company (yeah, at nvidia they are much more people working on software than hardware). This is their strength.
For Tegra and automotive, their advantage is Visionworks and more specifically ADAS:
http://www.nvidia.com/object/visionworks-cv-toolkit.html
good luck for mediatek to bring an equivalent to the table

Agreed completely. A lot of people fail to understand this, NV has software engineers that are devoted to creating software to drive the experience of their hardware and SOCs. It's a big differentiation for nvidia and that is precisely why NV completely dominates the PC mobile discrete GPU market and the professional markets by more than 80% for both. Software. Software development is THAT important. This is something they're pairing with the K1 for the automotive market (and premium high end automobiles as well). That is something that i'm afraid mediatek just doesn't have, and AMD doesn't quite understand, and also why AMD has been an utter failure in the professional market.(despite the hardware being good for the most part.)

That doesn't mean the K1 will be a massive hit, but I wouldn't hedge by bets against nvidia. In any case, NV isn't making the K1 a mass commodity consumer device to sell millions. It's going to be a smaller niche product for super high end devices. You can't compete with Qualcomm's modem IP, it's just not possible, and you can't compete with the bargain bin junk garbage that mediatek has in the Chinese market. So NV is going for premium devices with super high performance SOCs. I think it's the smart move personally, you just cannot compete with those other two companies on a commodity level. Not possible.

 

[mrmt]

I work in car industry, saving 10$ on one single component is a huge amount even for Mercedes.
The industry is under extreme price/profit pressure since long time ago.
And they buy stuff like this according to requirements, not according to what engineers thinks is cool.
Nvidia will have exactly the same issue they've had with smartphones: How to convince a purchase department to pay more than they have to?

But if you *have* to pay $10 for a given capability, is it easier to do in a luxury car or in a budget car?

 

[mrmt]

Just no.
nVidia's coperate margins are around 53%. And more than 50% of the revenue comes from their Geforce-Business.
Tegra's Margins are under the coperate average with the automobile on the upper end.

And what you are saying here? That the Tegra business went from above to below corporate average, and that today's consumer margins are below what they get with embedded for cars?

What part did you disagree, the part where I said that Nvidia failed on the consumer market and had to turn to embedded?

 

[mrmt]

It's crazy to see that they are still so many people that don't understand. Nvidia is a mainly a software company (yeah, at nvidia they are much more people working on software than hardware). This is their strength.
For Tegra and automotive, their advantage is Visionworks and more specifically ADAS:
http://www.nvidia.com/object/visionworks-cv-toolkit.html
good luck for mediatek to bring an equivalent to the table

Mediatek can bring small cheaper SoCs to the market for low requirements operations, you know, the kind of thing that would Nvidia critical mass for their business. OTOH, Intel will also pursue the software route:

http://www.intel.com/content/www/us/en/intelligent-systems/automotive/automotive-overview.html

As for software supplanting hardware as the main source of revenue, beware of the IBM and SUN examples. Maybe Nvidia will end up dumping Tegra and adopting ARM and Intel solutions in General for their software package? hmm.....

 

[tviceman]

That doesn't mean the K1 will be a massive hit, but I wouldn't hedge by bets against nvidia. In any case, NV isn't making the K1 a mass commodity consumer device to sell millions. It's going to be a smaller niche product for super high end devices. You can't compete with Qualcomm's modem IP, it's just not possible, and you can't compete with the bargain bin junk garbage that mediatek has in the Chinese market. So NV is going for premium devices with super high performance SOCs. I think it's the smart move personally, you just cannot compete with those other two companies on a commodity level. Not possible.

K1 should (and needs to be) more successful than Tegra 4. Xiaomi's Mipad and being the reference design for the (supposedly strict) Google TV specs should help out quite a bit. I also think that if they would have put a 1080p screen and 2gb of ram on the Tegra Note and priced it similarly to Google's Nexus 7 2013 version (to make up for the added costs), it would have sold better.

 

[ams23]

What Nvidia is telling you is that automotive margins will be higher than their margins, which is a true statement. What they didn't tell you is that their *current* margins on consumer Tegra is far below what they used to get with Tegra 2 and even sometimes Tegra 3 (when Tegra margins were ABOVE corporate average, as stated in a Q&A). In other words, margins on their consumer business are so low that an embedded business suddenly became more profitable than sell bleeding edge devices on the consumer market. This is the elephant in Nvidia's room.

That is false. Tegra consumer business gross margins have been a drag on overall corporate gross margins since Kepler first came out (ie. for years now).

And why do you think the kind of visual computing the manufacturers are putting in a car aren't commodities? What does Tegra brings to the table that Intel, Mediatek, Qualcomm and others won't be able to match in terms of hardware capabilities if they decide to exploit those same markets? A car is a much less constrained environment than, let's say, a tablet, and car software isn't above what an android tablet or windows convertible are going to offer you, and volumes are fine as long as you do not have Globalfoundries as your main foundry partner.... what are your constraints on this one? So yes, it is commoditizable.

Think logically about what you are saying. Very high parallel graphics throughput and parallel programming in a very low power envelope are not easily commoditizeable due to the hardware and software expertise required. There is a reason why Tegra automotive has so many design wins lined up in the automotive space while still maintaining gross margins that are close to corporate average.

 

[mrmt]

That is false. Tegra consumer business gross margins have been a drag on overall corporate gross margins since Kepler first came out (ie. for years now).

Nope. They became a drag since Qualcomm flooded the market with Krait processors and LTE modems, something Nvidia had no answer.

Think logically about what you are saying. Very high parallel graphics throughput and parallel programming in a very low power envelope are not easily commoditizeable due to the hardware and software expertise required. There is a reason why Tegra automotive has so many design wins lined up in the automotive space while still maintaining gross margins that are close to corporate average.

What are the applications of it? Mobile? A car is not as remotely as power/thermally constrained as a tablet or a phone. Intel, Qualcomm or even Mediatek can provide the hardware to cars. Hardware will have small gross margins, whether software can make up for it that's another story.

But I'll cross this gross margin bridge when I get there. Until I see embedded chips shipping in volume and Nvidia gross margin intact I'll believe Nvidia's claims that the margins on this business are fine.

 

[ams23]

Both Apple and Google are pushing thin client car systems powered by the driver's phone- given the mismatch between car and phone replacement rates, this makes a lot more sense if you don't want your in car system to look like outdated junk in 5 years. I just don't see that big a market for Tegra there.

Cars use dozens of microprocessors for a variety of different needs. Google and Apple are simply providing an interface for their smartphone users to use when in car. The cars will still need very advanced microprocessors built in to the car. Remember that Tegra automotive is a growing and accelerating business.

 

[ams23]

Nope. They became a drag since Qualcomm flooded the market with Krait processors and LTE modems, something Nvidia had no answer.

No, the gross margins for Tegra are not dictated by Qualcomm. The sales volume in the T4 generation was affected though.

What are the applications of it? Mobile? A car is not as remotely as power/thermally constrained as a tablet or a phone. Intel, Qualcomm or even Mediatek can provide the hardware to cars. Hardware will have small gross margins, whether software can make up for it that's another story.

But I'll cross this gross margin bridge when I get there. Until I see embedded chips shipping in volume and Nvidia gross margin intact I'll believe Nvidia's claims that the margins on this business are fine.

Hardware and software test requirements for automotive systems are much more stringent than for consumer hardware. It is not as simple as being the lowest cost or highest volume SoC supplier.

 

[mrmt]

No, the gross margins for Tegra are not dictated by Qualcomm. The sales volume in the T4 generation was affected though.

How so? If Qualcomm Krait was the benchmark for performance and it was in volumes big enough to put pressure on the market. Everything on the mobile market had to be priced according to Qualcomm price, Tegra included. With Krait Tegra 3 went from being a top notch solution for a subpar power hungry SoC. It's very naive of your part to think that Nvidia could charge the same for Tegra 3 after Krait was launched.

Hardware and software test requirements for automotive systems are much more stringent than for consumer hardware. It is not as simple as being the lowest cost or highest volume SoC supplier.

Intel develops mission critical servers and Qualcomm provides hardware for the aerospace industry. These two applications have extremely tight quality parameters, more than the automotive industry I suspect.

 

[ams23]

I work in car industry, saving 10$ on one single component is a huge amount even for Mercedes.
The industry is under extreme price/profit pressure since long time ago.
And they buy stuff like this according to requirements, not according to what engineers thinks is cool.
Nvidia will have exactly the same issue they've had with smartphones: How to convince a purchase department to pay more than they have to?

NVIDIA already has $2+ billion of multi-year contracts signed with various European and Japanese automakers, so convincing them to use Tegra is not an issue. It is not easy to find a very low power (<5w), low cost, highly programmable (CUDA) and modular ultra mobile SoC with > 300 GFLOPS graphics throughput. If anything, an SoC such as TK1 represents a huge cost savings for ADAS compared to what automakers would have needed to use in the past.

 

[ams23]

How so? If Qualcomm Krait was the benchmark for performance and it was in volumes big enough to put pressure on the market. Everything on the mobile market had to be priced according to Qualcomm price, Tegra included. With Krait Tegra 3 went from being a top notch solution for a subpar power hungry SoC. It's very naive of your part to think that Nvidia could charge the same for Tegra 3 after Krait was launched.

You aren't making sense here. NVIDIA never started with higher than normal gross margins with T3 consumer business. Both T3 and T4 were cost-effective ~ 80 mm^2 SoC's built on mature fab processes with no built-in modem cost, whereas Qualcomm S800 SoC's are ~ 50% larger, include the cost of a built-in modem, and are fabbed on a more advanced fab process.

Intel develops mission critical servers and Qualcomm provides hardware for the aerospace industry. These two applications have extremely tight quality parameters, more than the automotive industry I suspect.

And so what? Intel and Qualcomm do have a footrint in the automotive space. Your earlier suggestion that something like ADAS is an easily commoditizeable space that a company like Mediatek can dominate is completely false.

 

[mrmt]

You aren't making sense here. NVIDIA never started with higher than normal gross margins with T3 consumer business. Both T3 and T4 were cost-effective ~ 80 mm^2 SoC's built on mature fab processes with no built-in modem cost, whereas Qualcomm S800 SoC's are ~ 50% larger, include the cost of a built-in modem, and are fabbed on a more advanced fab process.

Krait impacts on the price Nvidia can fetch on T3.

And so what? Intel and Qualcomm do have a footrint in the automotive space. Your earlier suggestion that something like ADAS is an easily commoditizeable space that a company like Mediatek can dominate is completely false.

ADAS isn't really commoditized, but Tegra, the SoC, is.

 

[ams23]

Tegra 4i is an SoC that was ultimately aimed at an easily commoditizeable space. TK1 and future Tegra SoC's are not aimed at spaces that are easily commoditizeable, nor do they need to be because the CPU/GPU perf. and feature set and software support will be exceptionally good compared to most other ultra mobile SoC's on the market.

 

[rgallant]

Rationalization

is what I goggled after reading
Tegra 4i is an SoC that was ultimately aimed at an easily commoditizeable space

 

[CrystalBay]

It doesn't matter to me that much, if the total package is of desirability.

 

[xpea]

Intel develops mission critical servers and Qualcomm provides hardware for the aerospace industry. These two applications have extremely tight quality parameters, more than the automotive industry I suspect.

you still don't understand.
We are talking about full turnkey solution (hardware + software).
The processors sold by Intel and Qualcomm are just hardware deal with custom specs (especially related to radiation shielding). The software is developed by NASA.
On the automotive market aimed by Nvidia, the green team provides K1 in a form of an upgradable VCM module with a full stack of software ready to use, like enhanced night vision, pedestrian detection, collision detection system and speed limit sign recognition just to name a few. This "ready to go" software saves automakers months of development and thus ton of money. That why Nvidia can charge more per SoC.

 

[sontin]

And what you are saying here? That the Tegra business went from above to below corporate average, and that today's consumer margins are below what they get with embedded for cars?

You have really no clue, huh?
Since the introduction of Fermi in 2010 nVidia's margins have been rising. At the same time their Tegra business went massiv up. Tegra had never margins "above" coperate margins. And nVidia hit a record gross margins result last year while Tegra hit a bottom in revenue...

What part did you disagree, the part where I said that Nvidia failed on the consumer market and had to turn to embedded?

How exactly "failed" nVidia in the consumer market when yet they had a record revenue and gross margins year?

You should get your facts right, really.

 

[NTMBK]

How exactly "failed" nVidia in the consumer market when yet they had a record revenue and gross margins year?

You should get your facts right, really.

They failed in the consumer SoC market, which is what we are talking about... Tegra 4 and 4i were utter catastrophes, and K1 is not shaping up any better. What happened to all those 1h14 devices that JHH promised?

 

[sontin]

He is clearly talking about their graphics consumer products. Or why would they turn with their "embedded" product into the "embedded" market?!

However, nVidia's Tegra Business is up since FQ3 2014...

 

[soccerballtux]

Years needed to get their own CPU design group producing is not surprising given just implementing ARMs bog standard designs can take a year+. Still no 64 bit standard ARM out yet only Apple with their billion+ dollar custom design. That Nvidia's 64 bit ARM design is mature enough to apparently get a Google Nexus design win is impressive.

Even so, Tegra 4i was late and even if it had been on time was seemingly targeting mid-range phones which now relegates it to the "budget" segment.

bog?