NVIDIA Tegra Architecture

OL1vfcd.jpg


http://www.youtube.com/watch?v=Pfp_ZFs7DIA

Wished Nvidia would release the Tegra Note 7 with Tegra K1 sooner.

Looks good. I don't know what Google has in store for the 2014 Nexus 7, but since the main drawback of the first Tegra Note was its 720p display, this could substantially improve NVIDIA's position. At the right price this could be a great gaming tablet.


This, however, is the sort of marketing fluff I think this thread could do without.
 
http://www.nextpowerup.com/news/6732/microsoft-surface-3-to-pack-nvidia-tegra-k1-soc.html

Rumors has it that MS has chosen Tegra K1 for the next Surface tablet, another design win for Nvidia if true.

That news isn't too surprising. What is surprising is that the TK1 variant with Denver CPU will be available in 2H 2014 (with a pin-compatible layout to boot!). So you can see why Tegra K1 is such an incredible milestone for NVIDIA. In one generation (!) of ultra mobile product in 2014, they will have support for: unified shader architecture, DX11.2, OpenGL 4.4, CUDA 6, OpenGL ES 3.0, geometry shaders, tesselation, 64-bit Windows on ARM, 64-bit Android, fully custom low power CPU.

Due to the timeframe that Surface 3 will be released (ie. 2H 2014), the inclusion of a Denver 64-bit CPU could make for a very interesting product and perhaps an OS overhaul too.

The only thing I'm not too optimistic about is seeing TK1 in a smartphone in the USA (since carriers with CDMA technology are forced to use Qualcomm's tech). If a TK1-powered smartphone becomes available in the USA on a non-CDMA carrier (such as AT&T), I will probably buy it.
 
Last edited by a moderator:
"Incredible milestones" should come with a fairly sizeable market share increase. Let me see the first high volume design win instead of a fancy huge digital photo frame & then it'll blow up some skirts.

Ah yes the Surface3. Why Microsoft thinks that even the current one would be a sales success with an as high price is beyond me.
 
Can NV deliver these K1 at prices that will take market share away from Snapdragon is the question.

Will the design and architecture allow it to be cost-competitive with Qualcomm?
 
"Incredible milestones" should come with a fairly sizeable market share increase. Let me see the first high volume design win instead of a fancy huge digital photo frame & then it'll blow up some skirts.

We'll see what happens. Given that certain things are out of their control (such as Qualcomm's monopoly on CDMA technology, and the lack of broader adoption/acceptance of a third Windows OS), there is only so much headway that NVIDIA can make. That said, the status quo is always subject to change too: http://www.extremetech.com/mobile/1...cdma-from-phones-in-2014-in-favor-of-pure-lte

Ah yes the Surface3. Why Microsoft thinks that even the current one would be a sales success with an as high price is beyond me.

I would guess that Microsoft has modest expectations for Surface 2 given what happened with Surface. Their goal is to have a very competitive product to the ipad, but they still have a lot of work to do with their application processor, OS, and industrial design in order to be competitive. The ipad Air is a much more attractive consumer tablet right now IMHO.
 
That news isn't too surprising. What is surprising is that the TK1 variant with Denver CPU will be available in 2H 2014 (with a pin-compatible layout to boot!). So you can see why Tegra K1 is such an incredible milestone for NVIDIA. In one generation (!) of ultra mobile product in 2014, they will have support for: unified shader architecture, DX11.2, OpenGL 4.4, CUDA 6, OpenGL ES 3.0, geometry shaders, tesselation, 64-bit Windows on ARM, 64-bit Android, fully custom low power CPU.

Due to the timeframe that Surface 3 will be released (ie. 2H 2014), the inclusion of a Denver 64-bit CPU could make for a very interesting product and perhaps an OS overhaul too.

The only thing I'm not too optimistic about is seeing TK1 in a smartphone in the USA (since carriers with CDMA technology are forced to use Qualcomm's tech). If a TK1-powered smartphone becomes available in the USA on a non-CDMA carrier (such as AT&T), I will probably buy it.

The NextPowerUp article linked above only uses MSFTnerd's tweets as a source for the Denver information, but a quick look at said tweets shows that he never actually said Denver would be used in the Surface 3:

6732-8d9767b8_562_490.jpg


He's just saying Windows won't support "instant go" until Threshold, and that the Surface 3 won't be released until Threshold either. This doesn't mean the Surface 3 will use TK1-Denver.

As for hitting incredible milestones in one generation, let's see what NVIDIA actually launches and when before getting too carried away.
 
Will the design and architecture allow it to be cost-competitive with Qualcomm?

The cost boils down largely to SoC die size and/or process node. Given how large the die sizes have become in high end Qualcomm SoC's, I cannot imagine that TK1 would not be cost-competitive.
 
The NextPowerUp article linked above only uses MSFTnerd's tweets as a source for the Denver information, but a quick look at said tweets shows that he never actually said Denver would be used in the Surface 3:

Well if you read between the lines, that is exactly what he says. He says that 64-bit ARM [Denver] CPU will not be supported until Threshold, and Surface 3 will launch at (or after) that time.

The Denver TK1 variant is pin-compatible with the A15 TK1 variant, and will clearly be available in 2H 2014 (it is probably already sampling right now), so it would be pretty bizarre not to go with Denver in the time frame of Surface 3 (late 2014). And considering that ipad Air already has a very impressive 64-bit application processor, Microsoft needs to move to 64-bit Windows on ARM as soon as possible for any ARM-powered Surface.

As for hitting incredible milestones in one generation, let's see what NVIDIA actually launches and when before getting too carried away.

TK1 with R3 Cortex A15 has been sampling for at least seven months now, so I'm not too concerned about that. Hope to see some products in Q2 2014.
 
Last edited by a moderator:
The cost boils down largely to SoC die size and/or process node. Given how large the die sizes have become in high end Qualcomm SoC's, I cannot imagine that TK1 would not be cost-competitive.

Qualcomm's SoCs feature integrated LTE. With Tegra you need a separate chip with all the corresponding additional cost, power and complexity. Obviously if you don't need LTE, e.g. in a tablet, that's a different story, but then NVIDIA competes with QCOM's LTE-less SoCs, as well as MediaTek's, Rockchip's, etc., and those are pretty cost-effective.

Well if you read between the lines, that is exactly what he says. He says that 64-bit ARM [Denver] CPU will not be supported until Threshold, and Surface 3 will launch at (or after) that time.

The Denver TK1 variant is pin-compatible with the A15 TK1 variant, and will clearly be available in 2H 2014 (it is probably already sampling right now), so it would be pretty bizarre not to go with Denver in the time frame of Surface 3 (late 2014). And considering that ipad Air already has a very impressive 64-bit application processor, Microsoft needs to move to 64-bit Windows on ARM as soon as possible for any ARM-powered Surface.

Lots of 2014 SoCs will feature 64-bit compatibility, not just TK1-Denver. The latter is only scheduled for "H2 2014" and, well, you know how these things work: it could just mean that NVIDIA will start shipping it to OEMs in December so that they can introduce devices in 2015.

That's what I meant when I urged prudence about the "one generation" thing. As far as I can tell, TK1-A15 (which appears to be on time) is still Tegra 5 and TK1-Denver is still Tegra 6, nothing but their names has really changed.
 
Qualcomm's SoCs feature integrated LTE. With Tegra you need a separate chip with all the corresponding additional cost, power and complexity. Obviously if you don't need LTE, e.g. in a tablet, that's a different story, but then NVIDIA competes with QCOM's LTE-less SoCs, as well as MediaTek's, Rockchip's, etc., and those are pretty cost-effective.

Not all high end Qualcomm SoC's have integrated LTE (for instance, S600 and S4 Pro did not have integrated LTE). And in general, an LTE-equipped tablet is offered at a much higher price premium than a Wifi-only tablet.

Lots of 2014 SoCs will feature 64-bit compatibility, not just TK1-Denver.

Not for high end SoC's on Android and Windows on ARM. Qualcomm's first 64-bit SoC's will be more in the value segment.

That's what I meant when I urged prudence about the "one generation" thing. As far as I can tell, TK1-A15 (which appears to be on time) is still Tegra 5 and TK1-Denver is still Tegra 6, nothing but their names has really changed.

Tegra 6 will have a Maxwell GPU, not a Kepler GPU, and will use a different fabrication process node with 3D FinFET transistors.
 
Tegra 6 will have a Maxwell GPU, not a Kepler GPU, and will use a different fabrication process node with 3D FinFET transistors.

Whereby 16 FinFET TSMC is oversimplyfied just 20SoC + FinFET; as for Parker yes it'll have a GPU with mostly more arithmetic throughput.
 
Not all high end Qualcomm SoC's have integrated LTE (for instance, S600 and S4 Pro did not have integrated LTE). And in general, an LTE-equipped tablet is offered at a much higher price premium than a Wifi-only tablet.

Some S600 and S4Pros feature integrated LTEs, some do not.
http://www.qualcomm.com/snapdragon/processors/s4/specs
http://www.qualcomm.com/snapdragon/processors/600

I'd wager that the LTE-less variants are quite competitive with Tegra in die size, especially TK1. Obviously they'd be slower than TK1 too, but let's see what QCOMM releases in 2014.

Not for high end SoC's on Android and Windows on ARM. Qualcomm's first 64-bit SoC's will be more in the value segment.

Qualcomm's first SoC, yes, but what about the second one? And Mediatek's? Samsung's? Just because NVIDIA announces something first doesn't mean they'll actually get it to market first.

Tegra 6 will have a Maxwell GPU, not a Kepler GPU, and will use a different fabrication process node with 3D FinFET transistors.

You're right, it seems Denver was originally supposed to be paired with Maxwell, not Denver:

tegra-roadmap-640x425.jpg


So TK1-Denver is different from Parker as we know (or knew) it. The question is: what happened? I can think of two possibilities:

a) Denver was early (or not as late as NVIDIA thought) so it was pulled in and added to the Logan variant we now know as TK1-Denver. Nothing has changed for Parker and it will show up in mid-2015 with Denver and Maxwell.
b) Maxwell was late, so TK1-Denver replaces Parker and will stick to a Kepler GPU. The next Tegra will move to a Maxwell GPU about a year later, i.e. probably in late 2015~early 2016.
 
I don't think Maxwell is late, it's already taped out on 28nm and 20nm process for the smaller Maxwell GPUs, TSMC's 16nm FinFET process might be late though and since Parker is gonna use 16nm FinFET, it depends on TSMC's ability to deliver, which everyone knows TSMC is no Intel.
 
I don't think Maxwell is late, it's already taped out on 28nm and 20nm process for the smaller Maxwell GPUs, TSMC's 16nm FinFET process might be late though and since Parker is gonna use 16nm FinFET, it depends on TSMC's ability to deliver, which everyone knows TSMC is no Intel.

Point taken about TSMC's FinFET process. Even if it's not late, NVIDIA might just be hedging with TK1-D, which sounds like a reasonable thing to do.

But while Maxwell has almost certainly taped-out in 28nm, I haven't read anything about 20nm anywhere at this point.
 
But while Maxwell has almost certainly taped-out in 28nm, I haven't read anything about 20nm anywhere at this point.
Well as many believe, I think first Maxwell dGPUs are 28nm and second gen will be 20nm, but I also believe TK1-D is 20mn as the first sample came back few days before CES. In fact TK1-D Product timing and size are suitable for 20nm. And maybe NV taped out 2 versions of TK1-D, one on 28nm, one on 20nm and will chose the best solution... who knows...
 
I don't think Maxwell is late, it's already taped out on 28nm and 20nm process for the smaller Maxwell GPUs

I don't doubt that any Maxwell irrelevant of size has taped out yet under 28nm, but do you have a link or anything half way credible to support that there's also been a 20SoC GPU chip tape out already?

, TSMC's 16nm FinFET process might be late though and since Parker is gonna use 16nm FinFET, it depends on TSMC's ability to deliver, which everyone knows TSMC is no Intel.

In TSMC's defense they're serving a whole damn LOT of different partners while Intel to 99% itself. That said chances are high that TSMC will actually start 16 FinFET production in H1 15' but it'll come to when chips in high quantities will be affordable to manufacture which is usually about half a year later than the latter.
 
Back
Top