Tegra revenue down 48% Y/Y

[erunion]

here's my thoughts on tegra:
Nvidia needs tegra to survive as its core businness is a dead end. Their core business is coprocessors for Intel motherboards. That business will go the same way Nvidia's Intel chipset business went, to zero.

The problem that Nvidia had with Tegra4(in addition to what problems they already had) is they targeted the wrong market. They got so caught in Windows RT and the chance to compete directly with Intel, they failed to maintain their position with android vendors. And it seems failed to make the product OEMs demanded.

The shield has the tegra 4 under a fan. That says a lot about it. NVIDIA was building the most powerful SoC possible with what they had available to them. Not neccesarily, the best product for tablets.(and definitely not the best product for phones)

Windows RT flopped and the surface along with it. So NVIDIA lost its android sockets, like the nexus 7, but with little to show for it.

Now the K1 is announced but I haven't seen any way it changes course versus tegra4. Its just a performance bump to the tegra4 and so will have the same fate.

 

[sontin]

Right...

BTW: A GTX750TI with Maxwell and a power consumption of 60W will offer 3x the graphics performance of Kaveri - for the same price.

Thing about it: The power of a GTX480 in a 15" notebook.
If any person believe that nVidia's core "business" is "a dead end" they should look at Maxwell closely. It will define a new future. A future where improvements come from architecture and not from shrinks. And for that you need R&D.

 

[erunion]

Right...

BTW: A GTX750TI with Maxwell and a power consumption of 60W will offer 3x the graphics performance of Kaveri - for the same price.

Thing about it: The power of a GTX480 in a 15" notebook.
If any person believe that nVidia's core "business" is "a dead end" they should look at Maxwell closely. It will define a new future. A future where improvements come from architecture and not from shrinks. And for that you need R&D.

You miss the point. Its not about what what product Nvidia brings to market. They are subject to x86(and thus Intel) because their product compliments an x86 product. If Intel takes away the PCI ports, Nvidia is instantly without a market for their product. (Not saying that will happen, its just a hypothetical to illustrate the point).

Nvidia once had a profitable chipset business, until Intel locked them out. It can, and will eventually, happen to dGPUs.

Intel holds all the cards, and between iGPU and Xeon Phi they have clearly targeted Nvidia. That why nvidia needs ARM to free their GPUs from x86 dependency.

 

[USER8000]

The vast majority of PCs use IGPs,and only hardware enthusiasts on tech forums,think otherwise. The increasing IGP performance of Intel and AMD IGPs,especially the Intel ones mean the low end market for discrete GPUs is under attack.

Why increase costs for an OEM desktop or laptop,by having a more expensive motherboard and cooling arrangement by having a discrete GPU in them?? Walk into most computer retailers worldwide and the vast majority of desktops and laptops uses IGPs. Its a race to the bottom.

Even Nvidia knows this and that is why the K1 has a large IGP section. The problem is Nvidia is relying primarily on the desktop discrete GPU and professional markets for its profits.

However,Xeon Phi is attacking the Nvidia large chip GPUs in the compute market,which makes up a significant amount of Nvidia profits and Intel has the money to burn,to get marketshare. Those high end desktop graphics cards like the GTX780 only exist,due to R and D costs for the large chips being subsidised by the professional(compute) market. If not Nvidia would limit itself to smaller and less powerful chips which are cheaper to make.

 

[NTMBK]

Right...

BTW: A GTX750TI with Maxwell and a power consumption of 60W will offer 3x the graphics performance of Kaveri - for the same price.

Thing about it: The power of a GTX480 in a 15" notebook.
If any person believe that nVidia's core "business" is "a dead end" they should look at Maxwell closely. It will define a new future. A future where improvements come from architecture and not from shrinks. And for that you need R&D.

Uh huh, sure. If you don't mind I'll wait for an objective review before announcing that a 60W 28nm card matches a GTX480. Marketing slides aren't exactly the most accurate.

Maxwell should be a more interesting companion for Tegra, though. The unified memory features will make for a more efficient SoC.

 

[ams23]

Intel holds all the cards

Well no they don't. Intel licenses NVIDIA's GPU technology (not the other way around). Intel and NVIDIA will also be collaborating more closely in the future (because their products tend to compliment each other more often than not). And in any ecosystem that is not tied to x86, NVIDIA will have advanced ARM processors available to pair with NVIDIA GPU's.

 

[ams23]

Intel holds all the cards

Well no they don't. Intel licenses NVIDIA's GPU technology (not the other way around). Intel and NVIDIA will also be collaborating more closely in the future (because their products tend to compliment each other more often than not). And in any ecosystem that is not tied to x86, NVIDIA will have advanced ARM processors available to pair with NVIDIA GPU's.

Now the K1 is announced but I haven't seen any way it changes course versus tegra4. Its just a performance bump to the tegra4 and so will have the same fate.

Right, except for the fact that TK1 is actually on time (it has been sampling for more than six months now), and has not just superior performance vs T4 but also has superior power efficiency (>3x GPU perf. per watt, >1.5x CPU perf. per watt) and superior feature set too (OpenGL 4.4, DX11.2, CUDA 6, unified shader architecture, geometry and compute shaders, tesselation, next gen camera and ISP architecture, 32-bit and 64-bit pin compatible variants, etc).

 

[Arachnotronic]

They got so caught in Windows RT and the chance to compete directly with Intel, they failed to maintain their position with android vendors. And it seems failed to make the product OEMs demanded.

Agreed. They won't make that mistake again. In all of NV's investor presenations, JHH is always talking up how great and fantastic Android is.

I can't believe anybody actually thought Windows RT would succeed.

 

[krumme]

Agreed. They won't make that mistake again. In all of NV's investor presenations, JHH is always talking up how great and fantastic Android is.

I can't believe anybody actually thought Windows RT would succeed.

Well it wasnt many either. Lol.

 

[ams23]

I can't believe anybody actually thought Windows RT would succeed.

The fundamental idea of having Windows on ARM for consumer products makes sense, but the implementation was not particularly smooth. It would have made more sense to unify Windows Phone and Windows RT platforms and offer a really efficient and lean Windows [Consumer] OS for consumer products while offering the full blown Windows [Professional] OS for professional products and power users.

 

[NTMBK]

Actually it is a great thing. The Denver variant will be well-positioned for a 64-bit Android platform. The fact that it is pin compatible with the R3 A15 variant is great too, because that compatibility should speed up implementation in real world devices.

64 bit ARM devices can also run 32 bit software. There should be no need for the A15 model.

 

[NTMBK]

Well no they don't. Intel licenses NVIDIA's GPU technology (not the other way around). Intel and NVIDIA will also be collaborating more closely in the future (because their products tend to compliment each other more often than not).

Sure, the same way that Nvidia chipsets used to 'complement' Intel processors- until Intel made them obsolete. And the same thing is happening in graphics. Intel don't want to co-operate with Nvidia, they want them dead.

and superior feature set too (OpenGL 4.4, DX11.2, CUDA 6, unified shader architecture, geometry and compute shaders, tesselation, next gen camera and ISP architecture, 32-bit and 64-bit pin compatible variants, etc).

The only market for Tegra is Android, which doesn't use OpenGL, DirectX or CUDA. And I've never seen any convincing camera advantage for Tegra tablets or phones.

 

[sontin]

64 bit ARM devices can also run 32 bit software. There should be no need for the A15 model.

And there was no need for all the 32bit processors in the past. D:
And what CPU should nVidia have used instead of A15? A9?! :hmm:

The only market for Tegra is Android, which doesn't use OpenGL, DirectX or CUDA. And I've never seen any convincing camera advantage for Tegra tablets or phones.

Which makes sense, because there is no OpenGL >4.x and Cuda compartible hardware on the market.

Seriously, either you're trolling or you have no clue...

 

[Homeles]

Well no they don't. Intel licenses NVIDIA's GPU technology (not the other way around). Intel and NVIDIA will also be collaborating more closely in the future (because their products tend to compliment each other more often than not). And in any ecosystem that is not tied to x86, NVIDIA will have advanced ARM processors available to pair with NVIDIA GPU's.

Intel licensing Nvidia's GPU tech is for legal protection. It's so they don't get sued again. They don't actually have Nvidia designs in their products.

And there was no need for all the 32bit processors in the past. D:
And what CPU should nVidia have used instead of A15? A9?! :hmm:

Qualcomm's gotten by just fine with A9. Apple's got a 64-bit A57-like implementation. An A15 isn't the only choice. I don't think it's a bad one today, however, given that the (IMO) broken architecture should be revised enough to be fixed now.

on a side note, I don't understand all this bashing on Tegra. What people want ? to have only a limited choice of 3~4 SoC brands ? I thought that choice is good for competition and consumers. This market is tough, many huge companies already left, so we must be glad to see Nvidia trying to find their place...

Try telling software developers that having more designs to support is a good thing.

Choice is great, but when the choice is between a company that consistently delivers subpar products, and a couple of companies that don't, it's fairly obvious that there's no need for the subpar company.

 

[ams23]

64 bit ARM devices can also run 32 bit software. There should be no need for the A15 model.

Now you are making zero sense. You are suggesting that NVIDIA should have delayed TK1 time to market by > 6 months to bring only a 64-bit Denver variant to market, even though there is no 64-bit Android OS to take full advantage of it yet?

 

[ams23]

Intel don't want to co-operate with Nvidia, they want them dead.

LOL, so now you speak for Intel? Like it or not, their increased collaboration has already been confirmed by NVIDIA. Obviously they will still compete in some areas, but AMD is a much more direct competitor to Intel in the x86 space, and Qualcomm is a much more direct competitor to Intel in the ultra mobile [Android] space.

The only market for Tegra is Android, which doesn't use OpenGL, DirectX or CUDA. And I've never seen any convincing camera advantage for Tegra tablets or phones.

Don't be silly, of course Android will make use of OpenGL. CUDA will be used in select applications as well (such as Tegra automotive). And Windows on ARM will make use of DirectX.

 

[ams23]

Intel licensing Nvidia's GPU tech is for legal protection. It's so they don't get sued again. They don't actually have Nvidia designs in their products.

While Intel is not using NVIDIA GPU's per se, NVIDIA's graphics IP and graphics patent portfolio is fundamental to Intel's in-house GPU designs.

Choice is great, but when the choice is between a company that consistently delivers subpar products, and a couple of companies that don't, it's fairly obvious that there's no need for the subpar company.

NVIDIA is arguably the worldwide leader in GPU computing, and probably has more fundamental graphics-specific patents than any other company in the world. Hardly a "subpar" company, and TK1 is hardly a "subpar" product. If anything, TK1 appears to have the most advanced ultra mobile GPU (Kepler.M) and most advanced ARM CPU (Denver 64-bit variant) on the market for 2014.

 

[Homeles]

NTMBK didn't say that Intel doesn't collaborate with Nvidia. He said they don't want to. And they don't. They're going to cut Nvidia out from the mainstream laptop market. They're barely hanging in there as it is.

While Intel is not using NVIDIA GPU's per se, NVIDIA's graphics IP and graphics patent portfolio is fundamental to Intel's in-house GPU designs.

It's a legal thing. Like I said.

And Windows on ARM will make use of DirectX.

Windows on ARM? Yeah, that won't be lasting very long.

 

[NTMBK]

LOL, so now you speak for Intel? Like it or not, their increased collaboration has already been confirmed by NVIDIA. Obviously they will still compete in some areas, but AMD is a much more direct competitor to Intel than NVIDIA is in the x86 space (and Qualcomm will be a much more direct competitor to Intel than NVIDIA is in the ultra mobile space).

Of course I'm not speaking for Intel; their actions speak for themselves. They have eliminated NVidia's chipset business, they are targeting their Tesla business with the Xeon Phi (with limited success so far, granted), and their rapidly improving integrated GPUs are directly targeting NVidia's bread and butter low end GPUs. Do you think it's just a coincidence that Intel directly compares Iris Pro with NVidia's Geforce parts in its marketing? They are quite clearly not collaborating. There is limited patent cross licensing, but that is as far as it goes.

Don't be silly, of course Android will make use of OpenGL. CUDA will be used in select applications as well (such as Tegra automotive). And Windows on ARM will make use of DirectX.

Android uses OpenGL ES, and there has been no indication that that will change, even if some ARM SoCs support full OpenGL. Google have previously deliberately disabled OpenCL on their devices, even on SoCs that were fully capable of supporting it. CUDA is a ridiculously niche application; why on earth would in car entertainment systems need unportable GPGPU code? CUDA on ARM is all about the HPC market- Denver cores running stripped down Linux distros, driving massive Maxwell GPUs without an Intel CPU in sight, that is their goal for CUDA on ARM. As for Windows on ARM? Windows RT is dead in the water. Windows Phone does have more of a future though, I will grant that.

 

[jpiniero]

If anything, Intel is looking to kill off discrete graphics on laptops. They are moving the default laptop processor to the one chip ULT models, and those don't support discrete graphics. I wouldn't put it past Intel to try to force GT3e+ on what's left of socketed laptop models past Haswell Refresh.

64 bit ARM devices can also run 32 bit software. There should be no need for the A15 model.

Denver probably isn't available at 28 nm.

 

[NTMBK]

Now you are making zero sense. You are suggesting that NVIDIA should have delayed TK1 time to market by > 6 months to bring only a 64-bit Denver variant to market, even though there is no 64-bit Android OS to take full advantage of it yet?

Sorry, I didn't explain that argument very well.

The division of resources just seems very odd. NVidia have already struggled to get Tegra 4 and 4i out in any sort of timely fashion, but now they are bringing out an A15 and Denver K1 within a (supposedly) narrow window of each other. Everyone knows that Denver will be the "good one", so why go for an A15 K1 when the Denver is coming 6 months later? It seems like they are Osborning their A15 part, by promising the "real" K1 so soon after.

Then again, this is assuming any sort of consumer knowledge of what is in their tablets, which is probably a silly idea. I suppose that most people really don't care that much and will buy whatever they buy, and manufacturers will use whatever SoC they can get their hands on.

But it still smells of a stopgap measure, a "Denver isn't going to get here in time but Tegra 4 can't hold the line" part. It's not like it's a disaster or anything, I'm just disappointed that Denver isn't ready yet. *shrug*

 

[ams23]

they are targeting their Tesla business with the Xeon Phi

They are not "targeting" Tesla per se, they are targeting HPC and Exascale computing using a different architectural approach to the Tesla GPU's.

and their rapidly improving integrated GPUs are directly targeting NVidia's bread and butter low end GPUs.

LOL, NVIDIA's "bread and butter" is most certainly not low end GPU's!

Do you think it's just a coincidence that Intel directly compares Iris Pro with NVidia's Geforce parts in its marketing?

Iris Pro is an alternative to a low-to-midrange NVIDIA mobile GPU. Iris Pro is obviously not being compared to higher end mobile or desktop NVIDIA GPU's. And NVIDIA GPU's keep evolving and improving too.

Android uses OpenGL ES, and there has been no indication that that will change, even if some ARM SoCs support full OpenGL.

OpenGL ES is a subset of OpenGL. Adding support in Android for full OpenGL is inevitable as Android evolves, and by definition, Android will also support OpenGL ES as a subset.

CUDA on ARM is all about the HPC market

It has already been confirmed that automakers such as Audi will be tapping into the compute potential of SoC's such as TK1 (and that includes use of CUDA).

 

[NTMBK]

Which makes sense, because there is no OpenGL >4.x and Cuda compartible hardware on the market.

Seriously, either you're trolling or you have no clue...

There is OpenCL capable Android hardware on the market, but Google have deliberately prevented the use of OpenCL. They clearly have their set of APIs that they want developers to stick to- it makes sense, as they want to prevent fragmentation. And besides, what developer would write CUDA code for Android when only a tiny fraction of consumer devices will ever even support CUDA? And even if full OpenGL were available, would the next Angry Birds use it if only a fraction of top end devices were capable of supporting it? OpenGL ES is the guaranteed lowest common denominator, and that's not going to change. Just look at how long we were stuck with DirectX 9 games on PC.

 

[ams23]

They're going to cut Nvidia out from the mainstream laptop market. They're barely hanging in there as it is.

NVIDIA has never been a big player in the "mainstream" laptop market. They can and will add value to premium laptops and gaming laptops.

Windows on ARM? Yeah, that won't be lasting very long.

It is almost inevitable that Windows Phone and Windows RT will merge in the somewhat near future to create one Windows on ARM consumer OS. Almost every single major SoC hardware vendor other than Intel is moving towards using ARM CPU's in the consumer space (including AMD).

 

[NTMBK]

They are not "targeting" Tesla per se, they are targeting HPC and Exascale computing using a different architectural approach to the Tesla GPU's.

Who else do you think is in the HPC market with big, hot coprocessors which are excellent at highly parallel floating point maths? I'll give you one hint- they like the colour green a lot. Of course the Phi uses a different architecture, but in terms of potential customers you better believe that they're going after the same people.

LOL, NVIDIA's "bread and butter" is most certainly not low end GPU's!

NVidia's profits come from selling GPUs, and the bulk of the volume of GPUs are sold in the low end. Of course they offer much higher performance products with higher margins, but they have much lower volume. Hence "bread and butter". In this analogy, Quadro would be the fine paté spread on top of the bread.

Iris Pro is an alternative to a low-to-midrange NVIDIA mobile GPU. Iris Pro is obviously not being compared to higher end mobile or desktop NVIDIA GPU's. And NVIDIA GPU's keep evolving and improving too.

The majority of PC sales these days are laptops and tablets. And yes, NVidia is improving too, no doubt about that; but Intel's iGPUs are improving faster. NVidia is out of all but the highest end Macbooks and iMacs, and Apple tend to be an indicator of how the PC market as a whole will go.

OpenGL ES is a subset of OpenGL. Adding support in Android for full OpenGL is inevitable as Android evolves, and by definition, Android will also support OpenGL ES as a subset.

No, it's not inevitable. OpenCL has already been blocked deliberately by Google.

It has already been confirmed that automakers such as Audi will be tapping into the compute potential of SoC's such as TK1 (and that includes use of CUDA).

That does sound interesting actually, do you have a link? I wonder what on earth they're up to which needs CUDA.