NVIDIA Tegra Architecture

[Erinyes]

Since you know that their track record isn't exactly admirable, why are you asking in the first place?

Haha ok but at least they're better than Charlie/Fudzilla/SA

Well they are for 20SoC as it seems and while I've no idea if they'll also use Samsung to dual source rumors so far say no but you can't trust them either. Who said they'd manufacturing on 28nm/TSMC?

Even if they do go to TSMC for 20SoC, I think they will have to dual source. I did a rudimentary analysis on number of wafers required by Apple based on their shipments and I do not think TSMC has the wafer capacity for Apple to single source. I haven't posted the figures as we're sort of OT here but if anyone is interested I can post the figures in an appropriate thread/new thread.

There have been rumours of Apple and TSMC since 2011. Just google "Apple TSMC 28nm" and you'll find lots of news/"articles".

And I don't think that Apple intends to use TSMC ever as single source; in fact I rather think that 20SoC for A8 is just a singled out case. But I can however believe that TSMC would still want a piece of their future manufacturing case and hence my question where the good news exactly are. If it turns out to be true it's only good news for TSMC and Apple.

Yes..I do not think they can ever be a single source. Regarding future manufacturing, TSMC would have to increase their capex by a substantial amount if they want to have sufficient manufacturing capacity for Apple. I do not see them doing this, nor do I see them risking alienating all their other customers. Besides, with Samsung and GF having excess capacity and even rumours of an Intel manufacturing deal, I'm sure Apple can get better availability and pricing elsewhere.

No idea but I consider it absurd to have both A57 cores and Denver SoCs. For K1 they probably created a second variant with Denver to jump asap on the 64bit train.

It depends on which direction they plan to go with Denver. If they want to aim for even higher performance than the current Denver core in K1, it may not be suitable for phones and most tablets. In that case it would make a lot more sense to have an A57 variant to cater to the mass market and a separate Denver variant to cater to servers/HPC/etc. However, if they stick with a similar design to the current core then I agree with you.

2 SMMs sounds reasonable.

Yep..would be a logical step up from Tegra K1 though I do not see clocks going up substantially if it is still on 20nm. The other major improvement should be 4k H.265 decode (a certain Maxwell chip will get this).

According to NVIDIA's CEO, Erista is "right around the corner". I take that to mean that first silicon for Erista will be back from the fab. within ~ 3 months (with mass production starting ~ 9 months after that).

As for choice of CPU cores, I would think that NVIDIA needs to stick with it's guns and use fully custom Denver cores moving forward, rather than a mix and match approach. It would be strange to spend the last 5-7 years working on a fully custom Denver CPU, use it in Logan v2, and then abandon it in Erista v1. Each new generation of Tegra should always be more advanced and more sophisticated than the prior generation in all areas.

That sounds reasonable and fits the 20nm TSMC schedule as well. Lets see.

Regarding CPU cores, please see my reply to Ailuros above.

I agree it doesn't make sense, but a less absurd option is to have A53 cores (and only them).
A 3rd party SoC vendor would license the nvidia GPU tech and build the lower end SoC, not nvidia.
Cortex A12 is another option (32 bit)

I dont see a 3rd part SoC vendor going for NV GPU tech for a low end A53 part. ARM would likely be able to offer much better pricing with a Mali GPU. However I would not be surprised to see Nvidia using A53 cores in the successor to Tegra 4i, that is the market it makes the most sense. IMHO I do not see A12/A17 really doing that well in the market compared to A53. A12/A17 seem a bit late and there will be a marketing problem compared to A53 as well.

 

[Wynix]

According to a Chinese news site, Microsoft will be unveiling an 8" Surface Mini powered by Tegra K1 on April 2.

MS needs to shift all surface products to x86 with windows 8.1; This half-arsed shit just ain't going to cut it.

 

[ams]

This wouldn't be the first (maybe not even the second?) time nVidia has tried licensing out GPU IP cores like this and seemingly come up with nothing real to show for it.

http://www.nvidia.com/object/IO_11430.html

Linking to a 10-year old article doesn't do you any favors, nor does it do this discussion any justice. Anyway, NVIDIA only started talking about IP licensing for their modern GPU cores ~ 9 months ago! They are targeting vertically integrated customers only. This represents a very small number of players (ie. they are not going to license their technology to every Tom, Dick, and Harry), and this is not something that happens overnight. NVIDIA probably has the largest number of fundamental graphics patents in the world, and with their modern Kepler/Maxwell architecture make some of the world's most power efficient GPU's too (even Kepler.M in TK1 is reportedly 1.5x more power efficient than the best ultra mobile GPU's available today on the same fab. process node). In my opinion, it would be foolish for a vertically integrated player not to consider licensing NVIDIA's GPU tech. if given the opportunity to do so.

 

[ams]

MS needs to shift all surface products to x86 with windows 8.1; This half-arsed shit just ain't going to cut it.

The non-Pro Surface tablets are consumer tablets without DVD/CD drives. The vast majority of consumers don't care about legacy x86 support. If restricted simply to x86 SoC's, the innovations from ARM SoC makers would be missed out entirely. The smarter thing to do, in my opinion, would be for Microsoft to merge Windows Phone and Windows RT OS's to create one very lean and efficient Windows OS aimed for the consumer segment.

Surface needs improvements to industrial design and improvements to internal hardware (along with slightly more aggressive pricing) in order to keep pace with other premium tablets. It already has some killer free apps included (MS Word/PowerPoint/Excel/Outlook), and even with the Windows OS fragmentation, the Windows app store has well over 100,000 apps available for download to any Windows tablet. Clearly Microsoft still has a lot of work left to do, but improvements in each of these areas is certainly well within their reach.

 

[Lazy8s]

Just because nVidia waits several years in between their attempts at the mobile GPU IP core licensing game doesn't mean anything has actually changed for them for the better.

The list of potential customers is only narrowing now. And nVidia's benchmarks of decidedly non-smartphone Tegra K1 configurations versus last year's Apple SoC which runs in a relatively compact phone conveniently avoid a meaningful, thermally limited, comparison that accounts for form factor.

If you're betting that a smartphone K1 will hold off A8 in performance, you haven't been paying attention.

 

[ams]

Just because nVidia waits several years in between their attempts at the mobile GPU IP core licensing game doesn't mean anything has actually changed for them for the better.

NVIDIA has not talked about licensing any of their modern day GPU cores until around 9 months ago. You can spin it how you want, but that is the reality of the situation. And the Maxwell architecture is the first architecture that they have designed to be "mobile first", so that is their first architecture (in my opinion) that makes sense to license to vertically integrated players in the ultra mobile space.

And nVidia's benchmarks of decidedly non-smartphone Tegra K1 configurations versus last year's Apple SoC running in a relatively compact phone conveniently avoid a meaningful thermally limited, form-factor comparison.

Looks like you are the one who hasn't been paying attention NVIDIA compared the TK1 GPU to the A7 GPU and the S800 GPU at the same level of performance. With performance equalized, the latter two GPU's consume ~ 1.5x more power than the TK1 GPU: http://cdn.pcper.com/files/imagecache/article_max_width/review/2014-01-05/slides22.jpg . So the A7 GPU and S800 GPU will actually be more power hungry than the TK1 GPU for any given graphics workload, not the other way around.

If you're betting that a smartphone K1 will hold off A8 in performance, you haven't been paying attention.

Don't put words in my mouth. I never said that the A8 GPU would not be competitive. But when TK1 comes out, I think that the GPU performance will be head and shoulders above any other ultra mobile GPU on the market. And obviously there is more to ultra mobile than just smartphones.

 

[Blazkowicz]

The non-Pro Surface tablets are consumer tablets without DVD/CD drives. The vast majority of consumers don't care about legacy x86 support. If restricted simply to x86 SoC's, the innovations from ARM SoC makers would be missed out entirely.

But if you take two Windows 8 tablets, one that runs Windows 8 and the other one that runs Windows 8 RT, they do the same things. They both run two interfaces, Metro on WinRT and explorer.exe on win32. Yet the ARM tablet can only run office, notepad and paint (if paint is available at all) whereas the x86 tablet can run say Winrar and Paint.NET.

It seems completely arbitrary, in fact the product obviously is designed for Microsoft's strategic interests, such as getting people to use the Microsoft store (whatever it's called). But if the consumer understands anything about the situation.. Why is he asked to consider this tablet as "not a real computer", this other one as "definitely a real computer" and any complex shade of this stuff?

Your position is to say "You don't want to run winrar" and the rationale seems to be "because you're a consumer". And I don't understand it, since real computer runs arbitrary software. It's like some pen could be used only for drawing and another pen can only be used for accounting, and the ink stops flowing when machinery and logics inside the pen's body detects you're using the pen for the wrong application.
There's maybe room for a new OS that works on ARM, has tablet-like apps and has a desktop like Windows RT ARM does with a file manager etc. (I think Android comes with no file manager out of the box) and allows you to run desktop apps if you really want to.

 

[ams]

My point is that most consumers who buy tablets have little to no interest in legacy x86 software support, even if it exists in a Windows 8 tablet. And the extra baggage from a full Windows OS is not necessarily a great thing for a basic consumer tablet.

 

[itaru]

https://mobile.twitter.com/pixelio/status/449297033535700992

 

[Xmas]

My point is that most consumers who buy tablets have little to no interest in legacy x86 software support, even if it exists in a Windows 8 tablet.

Maybe so, but without legacy support Windows loses its biggest advantage over Android and iOS. Most consumers who buy tablets have little interest in Windows RT software support as well.

 

[ams]

Maybe so, but without legacy support Windows loses its biggest advantage over Android and iOS. Most consumers who buy tablets have little interest in Windows RT software support as well.

Legacy support has resulted in what is arguably a bloated OS for the consumer tablet space. As for Windows RT, it doesn't make sense to have three OS variants (between Windows Phone, Windows RT, and Windows 8), but two OS variants makes some sense (to support the consumer and professional space for ARM and x86 processors respectively), and any Windows store app should be made to be compatible for these OS variants.

 

[Lazy8s]

Performance characteristics tend to change quite a bit from prototype to volume production units that have to account for variability of bins, completed driver builds, and thermal/power profiles which have to accommodate actual compact form factors. Comparing K1 when they did to A7 and Snapdragon was premature.

The GfxBench database already indicates that Apple may be shipping A7 GPU drivers with significantly improved performance by the time K1 launches, potentially making nVidia's initial comparison even less representative of a comparable end product comparison.

 

[RecessionCone]

Performance characteristics tend to change quite a bit from prototype to volume production units that have to account for variability of bins, completed driver builds, and thermal/power profiles which have to accommodate actual compact form factors. Comparing K1 when they did to A7 and Snapdragon was premature.

The GfxBench database already indicates that Apple may be shipping A7 GPU drivers with significantly improved performance by the time K1 launches, potentially making nVidia's initial comparison even less representative of a comparable end product comparison.

I don't read too much into that particular comparison either, but K1 silicon and drivers were also pretty raw back then...

 

[A1xLLcqAgt0qc2RyMz0y]

RIP Tegra 4i

The Tegra 4i is the first and LAST Nvidia integrated LTE modem SOC.

What's Next for NVIDIA's Modems?

http://www.fool.com/investing/general/2014/04/06/whats-next-for-nvidias-modems.aspx

and perhaps not surprisingly -- is that the company no longer plans to try to go head-to-head against the remaining players in the integrated apps processor and cellular baseband space.

RIP Tegra 4i
While there had been some encouraging early signs for the company's Tegra 4i (integrated apps processor and baseband), it looks as though the part ultimately failed to gain enough traction to continue developing successors. The problem here, and one that many had feared would ultimately be the case, is that this market is all about bringing in the cheapest, most integrated chip at the lowest possible cost.

Nvidia's external modem will continue to be available.

David Wong: Chris ... you've exited mainstream smartphones, so where does that leave your baseband capabilities?

Chris Evenden: So we still need the baseband ... if only for automotive. I mean, connectivity is going to become increasingly important in all of those markets. So, we'll keep developing a standalone baseband, voice and data; we're not going to develop an integrated modem/applications processor. That product, you've got one target market in mind, which is mainstream smartphones.
​

NVIDIA does plan to continue developing its own standalone baseband chips (useful for high-end phones, tablets, and -- the star of Tegra's show -- automotive).

 

[Alexko]

"So we still need the baseband ... if only for automotive."

Sounds like they might be giving up on tablets too.

 

[Deleted member 13524]

Wow, T4i must have tanked really hard if nVidia is giving up on integrated modems right after their first try.

For major SoC suppliers, who do we have with integrated baseband processor? Mediatek, Qualcomm Broadcom and Samsung (still lacking LTE, though)?

 

[wco81]

Are people really going to get a mobile data plan for their cars?

Apparently Teslas come with electronics which have a mobile radio?

 

[silent_guy]

Better: in the future, will you be able to buy a high end car that doesn't have a radio?

 

[Deleted member 13524]

Are people really going to get a mobile data plan for their cars?

Apparently Teslas come with electronics which have a mobile radio?

I think the logic thing to do is for these high-end cars to come with lifetime data plans like the Kindle.

To be honest, no one will be using their car to make huge downloads, it'll be mostly navigation data and some text-based info like weather reports and news.
If one pays >$150 000 for a car, a simple (and limited) lifetime data plan shouldn't make a significant dent in the profits.

 

[ams]

An SoC with baseband modem integrated on die is ideal for cost-sensitive products that require connectivity. Since NVIDIA is focusing on high end ultra mobile products, it doesn't make sense for them to integrate a baseband modem. NVIDIA will have quicker time to market and higher application processor perf. for a given SoC die size by foregoing baseband modem integration.

Obviously Qualcomm has a very strong baseband modem product roadmap. Note that most people change their phone every 1-2 years and will have newer baseband tech with each upgrade cycle, so having a software-defined modem is not a huge benefit there. But in the automotive space, most people don't change their car for at least 5 years, so having a software-defined modem may make some sense there.