NVIDIA Tegra Architecture

http://www.anandtech.com/show/7080/nvidia-isc-2013-cuda-55-more

 

I did not see this one coming...

http://seekingalpha.com/currents/post/1092382?source=email_rt_mc_readmore

 

http://www.anandtech.com/show/7083/nvidia-to-license-kepler-and-future-gpu-ip-to-3rd-parties

They are being aggressive with this too by making future architectures available for licensing as soon as they tape out. It's still an open question how well Kepler scales down to tablet and smartphone power profiles to determine if there will be license interest in the mobile space compared to existing players who have licensable architectures designed specifically for mobile.

 

NVIDIA's intention to license their GPU technology to other vertically integrated companies has been known for quite a few weeks now, so I'm not sure why it is being widely reported just now.

Kepler.M mobile processors are designed specifically for mobile.

 

I thought their about to do so splendid in the foreseeable future; why would they all of the sudden want to license GPU IP where margins are rather crapalicious for their usual business model? Selling at the typical GPU IP royalty average of 30 cents gives you with 100Mio units a mere revenue of $30Mio.

Time for the usual GPU IP vendors to look over their shoulders no doubt, however I'd need a reasonable explanation how it'll suddenly fit into their usual business model and how the heck do they exactly plan to reach their claimed $1b revenue/break even point exactly.

You don't have the RTL by chance for it do you?

 

you'd have thought they would have kept this under the radar, and only announced publicly once they had a decent licensee to announce with it.

 

Based on no insider information, I'd be willing to bet quite a lot of money than their primary targe is Samsung by far (and possibly Apple longer-term for Macs). On the other hand, I would be extremely surprised if they seriously considered the possibility of ever licensing anything to Qualcomm (or Broadcom/Mediatek/etc.)

I wonder if this means I should start back on my project to do extremely low-level analysis of Kepler/GK110...

 

I doubt Qualcomm would be interested given that the Adreno 330 appears to have overtaken Tegra 4 - http://arstechnica.com/gadgets/2013...on-800-benchmarked-sports-extremely-fast-gpu/

Can't see Apple going for it either, they've been hiring a lot of their own guys for a reason. So yeah Samsung would appear to be the most likely candidate however if that was on the cards you'd think they'd have had an annoucement to make regarding it.

 

Tegra 4 does not have a Kepler GPU in it, so its characteristics are irrelevant to the discussion. The right comparison is between Tegra 5 and whatever else is on the market at the time.

 

https://devtalk.nvidia.com/default/...-319-and-linux-arm-support-for-discrete-gpus/

 

Apple went with AMD for the Mac Pro design they showed and it was interpreted as rejecting CUDA for OpenCL.

What kind of price competition is there between Qualcomm and Nvidia? That more than benchmarks is probably what drives deals.

 

It seems to be a good time to target Samsung for it; question then also is what Samsung is willing to pay and not after all.

 

NVIDIA is expanding their business opportunities in the ultra mobile space. Large companies that design their own SoC's are increasingly moving towards a more vertically integrated business model where it makes the most sense to license processor technology. Small and medium sized companies that do not design their own SoC's will still need to rely on companies such as NVIDIA to design and build the entire SoC. So NVIDIA's licensed technology is suitable for the former case, while NVIDIA's Tegra processors are suitable for the latter case.

The GPU R&D budget for NVIDIA stays essentially the same whether or not their GPU technology is licensed to other companies because there is still a need to design and build Tegra/Geforce/Quadro/Tesla processors. So NVIDIA's corporate margins would actually increase if they are able to gain new business by licensing their GPU technology to large vertically integrated companies.

 

The Adreno 330 GPU in the S800 SoC appears to be much stronger in GLBench 2.5/2.7 compared to the Adreno 320 GPU in the S600 SoC, but for some reason it is "only" about 20% faster than S600 in 3dmark Ice Storm.

 

Any viable and reasonable examples outside the above NV marketing drivel that you seem to parrot quite well? In case you haven't noticed you're not answering you're just playing back NV's marketing CD.

Sounds like you have a fairly good understanding how IP licensing in general works and what it exactly takes nowaydas. Now apart from that do you have anything substantial to answer my questions? You may vertically integrate your answers into something that actually makes sense.

 

Oh that's easy: GLB2.7 is more forward looking or stressful than GLB2.5 or Ice Storm and therefore irrelevant

 

Well, First i though there would be a separate thread for Nvidia licensing their GPU IP for such as big news, and yet no one seems to care ( yet ) on this forum.

I too thought that Samsung was the primary target, afterall they tried to move away from IMG to ARM ( possibly because they dont like using the same GPU tech as Apple ) but ARM didn't work out since Mali sucks.

But I thought Apple would be a much better fit, since they develop the drivers themselves they could save resources by moving the whole Mac and iOS GPU under one roof. Although putting everything in one basket is very un-Apple in recent years. Especially when Tim Cook is now the CEO.

 

It's my understanding (for which I'd gladly like to stand corrected) that Samsung tried to develope it's own GPU for their SoCs in order to not license any GPU IP and failed. Samsung licensing NV GPU IP would change the SFF mobile landscape quite radically to be honest and would be quite a pleasant surprise; albeit if they'd license now or soon it would mean at least 18-24 months for integration.

As for Apple an IP license for Macs would be equally interesting; albeit I don't think it's much of a secret that fruitco is a damn tough bargain

 

Try using some common sense for a minute here. NVIDIA is only recently starting to engage with companies who may or may not be interested in licensing Kepler GPU technology. 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.

 

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