NVIDIA Tegra Architecture

[Ailuros]

Try using some common sense for a minute here.

My common sense is allergic to any sort of marketing bullshit.

NVIDIA is only recently starting to engage with companies who may or may not be interested in licensing Kepler GPU technology.

Yes they "just" recently entered the SFF market and just recently seemed to have failed in it.

If companies moving towards vertical integration do decide to license NVIDIA GPU technology, it is unlikely that we will know anything about it until real products based on these technologies come to fruition. The important takeaway from NVIDIA's press release is that, due to vertically integrated business models for large players in the ultra mobile space, Tegra processors designed and built by NVIDIA cannot service the entire marketplace alone, irrespective of how good or how fast they may some day become.

In other words the same old same old marketing stuff as in your former post just in a different package; are there any sentences or reasoning of your own available?

It was quite a few years ago where I was convinced by NV followers that not licesing GPU IP for SFF SoCs (like they already did for SONY's PS3) was in favor of their business model. It made and still makes sense to me; so instead of harping on the same vague substance free stuff I'd like to read a few reasonable points to convince me that licensing GPU IP suddenly makes sense. Or better try to convince me that they're not as desperate as they look like at the moment.

 

[ams]

Not if they lose business by enabling their competitors to make better SoCs that in turn steal Tegra sales.

The majority of companies that rely on Tegra and other similar ultra mobile SoC's today will still need to rely on Tegra and other similar ultra mobile SoC's in the future. Whether or not large companies such as Samsung or Apple license NVIDIA GPU technology doesn't change that one bit.

 

[ams]

so instead of harping on the same vague substance free stuff I'd like to read a few reasonable points to convince me that licensing GPU IP suddenly makes sense.

It is very obvious why NVIDIA is now willing to license their GPU technology to some large companies. The reason is that some large companies are moving increasingly towards more and more vertical integration, and licensing technology is the only practical way to try and gain business from these companies moving forward.

 

[Exophase]

The majority of companies that rely on Tegra and other similar ultra mobile SoC's today will still need to rely on Tegra and other similar ultra mobile SoC's in the future. Whether or not large companies such as Samsung or Apple license NVIDIA GPU technology doesn't change that one bit.

Interesting choice of words there. Almost nobody relies on Tegra, at least as an initial design decision. There is no unique feature it brings to the table that its competitors can't do. The choice is made to use Tegra based on how it competes in price, performance, features, etc vs those competitors. Samsung could easily be one of those competitors if they start selling their SoCs more openly, which IMO they will have to do to continue justifying their fab space post-Apple (especially if Samsung mobile themselves are losing interest). A Kepler-licensed Samsung SoC offered to the same customers Tegra is could absolutely change things one bit.

And if they help Apple sell more tablets that will be inevitably at the expense of Tegra tablets, even if it doesn't change who buys the SoCs.

But I don't know what this licensing announcement really means. You would think that they're offering to license it to anybody, but in reality they may only be extending the offer to parties that are not perceived as operating in the same markets (which would also make it pretty pointless). Or they could be charging so much for the license that it still works in their favor.

 

[Ailuros]

It is very obvious why NVIDIA is now willing to license their GPU technology to some large companies. The reason is that some large companies are moving increasingly towards more and more vertical integration, and licensing technology is the only practical way to try and gain business from these companies moving forward.

Did I press on "replay" by mistake? It is very obvious that it's their last resort right now; IP licensing has ALWAYS been ideal in those markets it's just that the upper management at NV just now woke up and smelled the coffee. Better late than never. Apart from that I've still not a clue how it's going to fit into their business model.

 

[ams]

Interesting choice of words there. Almost nobody relies on Tegra, at least as an initial design decision. There is no unique feature it brings to the table that its competitors can't do. The choice is made to use Tegra based on how it competes in price, performance, features, etc vs those competitors. Samsung could easily be one of those competitors if they start selling their SoCs more openly, which IMO they will have to do to continue justifying their fab space post-Apple (especially if Samsung mobile themselves are losing interest). A Kepler-licensed Samsung SoC offered to the same customers Tegra is could absolutely change things one bit.

Companies such as Samsung and Apple that sell so many devices directly to consumers increasingly strive to create custom SoC's for their own products, not for other people's products. So I don't think it is plausible that they would all of a sudden start to design, build, and sell these same application processors to other companies. There is enough demand for fab space from fabless semiconductor companies around the world that Samsung can fabricate application processors for these companies without being forced to supply everything on their own.

 

[Ailuros]

Interesting choice of words there. Almost nobody relies on Tegra, at least as an initial design decision. There is no unique feature it brings to the table that its competitors can't do. The choice is made to use Tegra based on how it competes in price, performance, features, etc vs those competitors. Samsung could easily be one of those competitors if they start selling their SoCs more openly, which IMO they will have to do to continue justifying their fab space post-Apple (especially if Samsung mobile themselves are losing interest). A Kepler-licensed Samsung SoC offered to the same customers Tegra is could absolutely change things one bit.

One of the reasons semiconductor manufacturers prefer to license IP in the SFF mobile market is the flexibility (GPU) IP can give you. One can chose between a 1 cluster and an 8 cluster core for instance. Considering how much capacities Samsung might have soon, it ist an interesting case indeed.

And if they help Apple sell more tablets that will be inevitably at the expense of Tegra tablets, even if it doesn't change who buys the SoCs.

NV couldn't that easily afford 140-165mm2 sized SoCs like Apple. See flexibility above.

But I don't know what this licensing announcement really means. You would think that they're offering to license it to anybody, but in reality they may only be extending the offer to parties that are not perceived as operating in the same markets (which would also make it pretty pointless). Or they could be charging so much for the license that it still works in their favor.

I can see quite a few CUDA related licensing opportunities albeit not really relevant to Tegra in the given case.

 

[ams]

Apart from that I've still not a clue how it's going to fit into their business model.

Their other businesses will continue as usual. The only difference is that now NVIDIA can engage with large vertically integrated players in the ultra mobile space, whereas before they had little to no chance of doing so due to the lack of flexibility in meeting the needs of these potential clients.

 

[Ailuros]

I just vertically integrated myself in a horizontal universe (it's worse than a flashing banner of an obnoxious advertisement on websites *bling*vertically integrated*bling*)

 

[jimbo75]

It's Nvidia admitting Tegra has failed as a competitive mobile chip. With this move they can use all their early wafers on discrete GPU and still have a decent revenue stream from Tegra automotive sales that don't need to be at the cutting edge. If somebody else will pay for their graphics IP that's good for them, but they wouldn't be making this move if they believed Tegra had a future in mobile.

 

[Exophase]

Companies such as Samsung and Apple that sell so many devices directly to consumers increasingly strive to create custom SoC's for their own products, not for other people's products. So I don't think it is plausible that they would all of a sudden start to design, build, and sell these same application processors to other companies. There is enough demand for fab space from fabless semiconductor companies around the world that Samsung can fabricate application processors for these companies without being forced to supply everything on their own.

Samsung has and will continue to sell more Exynos SoCs outside of their own products. Look at the Exynos 5250s showing up in Chinese tablets. And you can find plenty of examples if you look back further to their ARM9 and ARM11 SoCs back before they spent most of their fab capacity on Apple (off the top of my head, GP32 and Freerunner used Samsung SoCs). iPhones also started out using close to off-the-shelf Samsung SoCs, not ones made through a special partnership. Only gradually did Apple gain influence that resulted in getting custom SoCs (the GPU in the 3GS is a big starting point). There's nothing all of a sudden about Samsung selling SoCs. Doesn't apply to Apple, but I didn't say it did.

Now on the other hand, can you name even one partner Samsung has fabbed for outside of Apple? It's odd that you say they make the SoCs only for themselves but the fabs are open for everyone. I don't think the demand is really there unless they succeed in courting away TSMC's big cutting edge clients. I'm sure they could sell to someone, but it's going to be awfully hard to fill that Apple sized hole. Convincing Samsung Mobile to use Exynos in their lower tier phones that were traditionally sporting ST-E or TI (now both defunct), weaker Qualcomm, or even nVidia SoCs sounds like an easier proposition.

 

[ams]

Samsung has and will continue to sell more Exynos SoCs outside of their own products.

I didn't realize that Exynos are spreading in Chinese tablets! But do you really think it is likely that we will see proliferation of the Exynos 5 "Octa" SoC (used in some of the latest and greatest SGS4 international variants) in any non-Samsung product anytime soon? I seriously doubt that.

 

[Exophase]

It doesn't have to just be Exynos 5410 or a similarly top-tier product, because that's not the only thing that'd be competitive with any of nVidia's product line. An update to Exynos 4 could be positioned against Tegra 4i, for instance.

Anyway, since Samsung Mobile freely uses whatever SoCs they feel like, enabling Exynos over Tegra should absolutely be a concern even if Samsung were the only ones using Exynos. It's not like they haven't used Tegra in the past. In fact, I would go so far as to say that SoC makers operating primarily outside of the low cost Chinese market should be trying to win Samsung Mobile more than pretty much any other phone or tablet manufacturer.

 

[ams]

If a small or medium sized company is in a position where they have to buy the SoC (as opposed to designing the SoC on their own), and they really want to have NVIDIA graphics/compute technology inside, why would they want to use an Exynos SoC (with licensed CPU/GPU technology) rather than a Tegra SoC if the Tegra SoC is able to serve the markets that they are interested in? In the markets that Tegra does not serve (such as super low cost tablets and super low cost smartphones), an Exynos SoC (with licensed CPU/GPU technology) would make much more sense than the Tegra SoC. I do understand what you are saying, but in practice, I don't think that there would be a huge opportunity cost of reduced Tegra business when licensing NVIDIA GPU technology to vertically integrated players. Tegra will always compete much more closely with Snapdragon, Atom, etc.

 

[liolio]

There is something I don't get, possibly I did not get the time line right (calendar vs fiscal year...).

So Nvidia is to release its tegra4i just ahead of the release of Cortex A12? I expect Chinese manufacturers to jump on this core as soon as possible, I wonder if Nvidia could just have postpone the launch and go with A12 vs pooring money in customizing a A9.

 

[Helmore]

NVIDIA didn't do any customizing to the Cortex-A9 core they're using. It's a revision that was done by ARM. Another thing is that, yes I think you're a little confused about timelines. ARM expects the first Cortex-A12 SoCs to show up around a year after Tegra 4i at the earliest.

 

[A1xLLcqAgt0qc2RyMz0y]

NVIDIA didn't do any customizing to the Cortex-A9 core they're using. It's a revision that was done by ARM.

Actually it's the other way around. Nvidia suggested changes which ARM implemented.

The company said it even made changes to the Cortex-A9 architecture itself to squeeze out even more performance from the chip, and submitted them to ARM, thus the resulting Revision 4, or R4, designation.

"ARM and Nvidia worked closely to further optimize the Cortex-A9 processor to drive performance and efficiency in areas such as streaming and responsiveness," said Tom Cronk, executive vice president and general manager, processor division, ARM. "This is an example of the collaboration and innovation that enables ARM technology-based solutions to be market drivers through multiple generations of SoC solutions."

http://www.tomshardware.com/news/Tegra-Super-Phone-Project-Grey-i500-LTE-GeForce,21124.html

 

[Homeles]

Well I'll be damned. Nvidia's actually doing custom ARM designs outside of Denver.

 

[kalelovil]

It isn't surprising, ARM has a history of such collaboration with their licencors. NEC Electronics for example helped them develop the ARM11 MPCore and Cortex A9.

 

[Homeles]

I was making a remark about how Nvidia has been using copy-pasted ARM designs, which has lead to lackluster products.