Nvidia Tegra

So nVidia bought both the rights to make a Cortex A15-based SoC and the rights to make their own custom application processor core based on the ARM's instruction core (like Qualcomm did with Snapdragon).

I guess we should expect Tegra 4 to use A15 (2012) and Tegra 5 to use an in-house cpu (2013)?
 
I'm not so sure Denver will go into the tegra line.

stolen from Ars Technicas Liveblog said:
we're buiilding a full custom processor in partnership with ARM, from a processor based on ARM. This team has come together to build the world's first ARM processor targeted at HPC.
 
Well that's pretty big news! I'm sure we'll also see this custom processor core in the notebook arena (see: Windows 8 thread - so maybe even desktops too) and to a lesser extent in tablets. But presumably that's coupled with a more desktop-like GPU than handheld-like unless the two converge completely. Anyway the only way I can see this being targeted at smartphones too is if they can implement it in an heterogeneous configuration with something Cortex-A5 or Cortex-A9-level for lighter workloads. I don't think that's the most likely possibility, but I'll be sure to poke NV at MWC ;)

In a related note, the Motorola Olympus/Atrix 4G having a 960x540 screen makes me a lot more tempted to consider it as a replacement for my iPhone 4 if we get an European variant with Honeycomb. And there's the Atheros acquisition by Qualcomm news today too - pretty busy day!

EDIT: Oh, and my expectation is Tegra4 will still be 4xA9 on 28nm. Tegra5 will be 4xA15 on 20nm (which would presumably tape-out in early 2013 versus late 2008 for Tegra2 and mid 2010.for Tegra3). Just a guess, but I don't think it makes sense to go from 4xA9 to 2xA15, and 4xA15 is a bit much (area/power-wise) on 28nm...
 
So Anand mentions that Tegra 2 doesn't have unified shaders and instead has 4 vertex shaders and 4 pixel shaders. Isn't that a weird ratio considering that PC GPUs had more PS than VS culminating in the 8VS : 48PS ratio for the X1900? Or are mobile games vertex shader heavy compared to PC games?
 
So Anand mentions that Tegra 2 doesn't have unified shaders and instead has 4 vertex shaders and 4 pixel shaders. Isn't that a weird ratio considering that PC GPUs had more PS than VS culminating in the 8VS : 48PS ratio for the X1900? Or are mobile games vertex shader heavy compared to PC games?

Perhaps it has to do with resolution? Pixel shader load scales with the number of pixels being shaded, while vertex shader load depends on the geometry. Since mobile devices typically have smaller resolutions, it would make sense for them to have relatively fewer pixel shaders.
 
Since the embargo is now officially over.

Tegra 2 = awesome. I have been using a prototype phone running on Tegra 2 for testing purposes and it is amazing. Sadly there are some software hiccups on Android 2.2, but the phone company assure me that it will all be fixed by release in March.

I think it is good that Nvidia dropped support for Symbian^4 and WP7 and put all of their development time into Android. It has worked out very well for them since Android is now becoming the dominant smartphone OS, and Tegra 2, if it performs as well as this on all devices, will become the dominant high-end platform for Android devices.

Talking to other companies, they are waiting to see how Tegra is received by consumers, i.e. whether they notice the difference and whether the premium over a Snapdragon 1.2GHz SoC is worth it. I think they will, because having used Samsung hardware and other devices based on Snapdragon I can safely say it is a massive improvement.

Interesting times ahead for Tegra 2, and Tegra 3 if Nvidia can pull off the biggest design win of their history (not related to any iDevice).
 
Interesting times ahead, indeed... especially when Apple throws a custom CPU architecture into the mix based off the ARM ISA.
 
I was really enjoying the Anandtech article until I got to this page:
http://www.anandtech.com/show/4098/...-more-architectural-details-and-design-wins/5

Tegra-optimized games? TegraZone?! "Custom marketplace for devices with Tegra"?!

It seems nVidia will try to spread the TWIMTBP infection to the mobile market.
It seems it's now only a matter of time before we see games that lock graphics enhancements to every device that doesn't have a nVidia GPU -> even when they're capable of doing it as fast, if not faster, than the Tegra's GPU.


Damn, we really didn't need this dirty game in the mobile market..
 
Only 8 FLOPs/clock from the PS (Vec4) ALU? Meh.....So it's basically just 1 Vec4 PS and 1 Vec4 VS ALU.
Yeah, that's very weak if true - we're talking the same ridiculously conservative ratio as the SGX535 here. See this is exactly the kind of reason why I didn't want to publish anything about any GPU architecture before I go to MWC11 in February ;)

NathansFortune said:
I think it is good that Nvidia dropped support for Symbian^4 and WP7 and put all of their development time into Android.
It is good indeed that they are focusing on Android, but I'm not aware of MS even giving the option of running WP7 to anyone but Qualcomm and I'm pretty sure NVIDIA never planned to support Symbian^4 (or at least never got it past an early planning stage) - so I'm curious what you're basing that on... :)

Talking to other companies, they are waiting to see how Tegra is received by consumers, i.e. whether they notice the difference and whether the premium over a Snapdragon 1.2GHz SoC is worth it. I think they will, because having used Samsung hardware and other devices based on Snapdragon I can safely say it is a massive improvement.
Comparing it to a QSD8x50A or a MSM7x30/MSM8x55 (the latter two are the exact same chip at different clocks) obviously doesn't make sense. As you seem to imply but maybe don't realise, the real competition is the MSM8260, and that's *dual-core* 1.2GHz with surprisingly good 3D performance and strong codec support. I certainly do not expect HTC or Sony Ericsson to bother with Tegra2 phones because of the MSM8260. And I *guess* (pure speculation) that Samsung will use OMAP4 in the ultra-high-end, its own processors in the high-end, Qualcomm in the mid-range (i.e. MSM7x30), and Broadcom in the low-end. And obviously Nokia will stick to TI/ST-E (plus maybe Renesas eventually) in the high-end. Not sure about Palm/HP but I think they're sticking to Qualcomm and TI in phones at least. So besides more devices from Motorola and LG, I think we'll mostly see a number of Tier2 phones (which can add up to a fair bit of volume anyhow) and a lot of tablets rather than widespread Tier1 phone adoption.

In the end, Tegra2 is the right chip at the right time both for some key phone manufacturers and for tablets in general and despite some delays NVIDIA's execution has still been better than the competition. There's nothing magical about it technically and I expect OMAP4 to be faster in nearly every category including 3D (but at noticeably higher cost). Much more important by far is Tegra3's execution now that NV has managed to get some good momentum with Tegra2 - if they can get tablets and phones out in 4Q11 and 1Q12 respectively with good software maturity, I think we're going to see some much more widespread adoption by OEMs (and consequently a lot more stress at competitors).
 
Tegra-optimized games? TegraZone?! "Custom marketplace for devices with Tegra"?!

It seems nVidia will try to spread the TWIMTBP infection to the mobile market.
It seems it's now only a matter of time before we see games that lock graphics enhancements to every device that doesn't have a nVidia GPU -> even when they're capable of doing it as fast, if not faster, than the Tegra's GPU.

Damn, we really didn't need this dirty game in the mobile market..
Two key points:
  • It's not a custom marketplace like the ones some operators are making. It's a separate interface for the standard Android marketplace - if the game isn't available there (e.g. you're in Europe and it's only available in the USA if that's even possible) I think you won't see it there either. From that perspective it's a just a great way for users to know what games with high quality graphics will work great on their device.
  • On the other hand, it's a also a foundation for NVIDIA to develop exclusive ties with developers ala TWIMTBP and help them implement features at the condition that they don't enable them on the competition. This is indeed very unfortunate - there is legitimate room for NV to do some good here (e.g. encourage devs to support 5xCSAA on Tegra2 since that's a proprietary feature they might not bother with otherwise), but there's even more room for them to do dubious things and not be called on it.
We'll see how it develops.
 
http://www.businesswire.com/news/ho...y-Analytics-Tegra-Products-Generate-1-Billion

-NVIDIA is well-positioned to take advantage of the emergence of high-performance smartphones and tablet devices in 2011, as described in “Apps Processor Profile: NVIDIA can Earn One Billion Dollars from Tegra in 2014,” from the Strategy Analytics Handset Component Technologies service. Strategy Analytics predicts that in 2011 alone NVIDIA’s Tegra 2 product could generate up to $300 million in revenue for the company.

Stuart Robinson, Director of the Strategy Analytics Handset Component Technologies service, says, “NVIDIA skipped ARM’s Cortex-A8 generation completely, choosing instead to work on the dual-core ARM Cortex-A9 generation for Tegra 2, which gave them an advantage against competitors. NVIDIA’s direct migration from ARM11 to Cortex-A9 could provide a three- to six-month advantage, particularly in the tablet segment.”
 
It is good indeed that they are focusing on Android, but I'm not aware of MS even giving the option of running WP7 to anyone but Qualcomm and I'm pretty sure NVIDIA never planned to support Symbian^4 (or at least never got it past an early planning stage) - so I'm curious what you're basing that on... :)

Originally Tegra was aimed at WM6.5, WM7, Symbian^3/4 and Android. Nvidia dropped all except Android.

Comparing it to a QSD8x50A or a MSM7x30/MSM8x55 (the latter two are the exact same chip at different clocks) obviously doesn't make sense. As you seem to imply but maybe don't realise, the real competition is the MSM8260, and that's *dual-core* 1.2GHz with surprisingly good 3D performance and strong codec support. I certainly do not expect HTC or Sony Ericsson to bother with Tegra2 phones because of the MSM8260. And I *guess* (pure speculation) that Samsung will use OMAP4 in the ultra-high-end, its own processors in the high-end, Qualcomm in the mid-range (i.e. MSM7x30), and Broadcom in the low-end. And obviously Nokia will stick to TI/ST-E (plus maybe Renesas eventually) in the high-end. Not sure about Palm/HP but I think they're sticking to Qualcomm and TI in phones at least. So besides more devices from Motorola and LG, I think we'll mostly see a number of Tier2 phones (which can add up to a fair bit of volume anyhow) and a lot of tablets rather than widespread Tier1 phone adoption.

I don't think it is fair to compare a real product that is going to be used in real devices (tablets and phones) to something that has no design wins and no prototype products forthcoming. None of Qualcomm's Tier 1 partners have committed to their dual core designs and not even HTC have a dual core phone announced using MSM8260. No the latest phones using Qualcomm SoCs are still using single core at 1/1.2GHz, the Sony Ericsson Arc, the new HTC phones and others have all been announced with single core SoCs. So it is a very fair comparison, if Qualcomm can't get their dual core SoCs out on time to compete that is their problem. The LG Optimus 2 is very real and is out in EU territories starting next month, and another 5-7 Tegra 2 products should launch in H1 with double that coming in H2. Other dual core SoCs are very much in the design phase.

In the end, Tegra2 is the right chip at the right time both for some key phone manufacturers and for tablets in general and despite some delays NVIDIA's execution has still been better than the competition. There's nothing magical about it technically and I expect OMAP4 to be faster in nearly every category including 3D (but at noticeably higher cost). Much more important by far is Tegra3's execution now that NV has managed to get some good momentum with Tegra2 - if they can get tablets and phones out in 4Q11 and 1Q12 respectively with good software maturity, I think we're going to see some much more widespread adoption by OEMs (and consequently a lot more stress at competitors).

Exactly this. Nvidia have executed the strategy almost perfectly. With Tegra 3 set to be announced sometime this year with the first designs being shown at CES2012 and products in H2 2012 OMAP4 will be competing with Tegra 3 not Tegra 2.

The phone/tablet manufacturers who have looked at Tegra 2 include:

LG
Samsung
Apple
Moto
Sony Ericsson
Google
Microsoft

Of these LG, Samsung and Moto will have products based on Tegra 2, while SE will probably use Tegra 3 in their Tablets.
 
I don't think it is fair to compare a real product that is going to be used in real devices (tablets and phones) to something that has no design wins and no prototype products forthcoming.

Indeed

With Tegra 3 set to be announced sometime this year with the first designs being shown at CES2012 and products in H2 2012 OMAP4 will be competing with Tegra 3 not Tegra 2.

Which goes completely against what you just said.

Omap4 has dev kit, has design wins, and is likely to launch in H1 (probably with playbook and possible other RIM devices), so it'll definitely compete with Tegra2, Whether it will also be competing with Tegra3 later depends on Nvidia executing correctly, but at the moment, there *IS* no tegra3. TI generally does its Omap announcements at MWC so we'll see in Feb what its OmapX plans are, then.
 
Yes, nVidia was the first one to show-off dual-core A9s in Q4'10, but it came at the cost of being completely absent of the whole market until that time (I don't think the sales numbers of Kin are relevant to the worldwide smartphone numbers).

Even if Tegra2 is in fact the development platform for Honeycomb, how long will it take until it gets outperformed?
The 1st-gen Snapdragon was the development platform for Éclair and Froyo, but during a great part of its existence it was outperformed by the Hummingbird (graphics-wise, it was even outperformed by the old 100MHz SGX530 in OMAP34xx)

So how much does it really matter, to be the first in the dual-core game and getting the dev-platform crown? Is Google going to do exclusive optimizations for the Tegra2 parts? It didn't seem to do with Snapdragon, as its clock-for-clock performance advantage against A8 didn't seem to show, except for a few synthetic benchmarks.
 
As long as Tegra 2 is the only dual-core chip available in the smartphone/tablet firmament, devices containing the chip will be the fastest and have the most brag-worthy specifications. Certainly a very good USP which should help sales quite a lot.

I hadn't realised (not really paid attention, to tell the truth!) that the dual-core Snapdragon derivatives weren't arriving soon so when do we expect to see the first dual-core competitor for Tegra 2? Will NV have time to get enough of a really good foothold in the market before other high-end chips are available in devices that match or exceed the Tegra 2 capabilities?
 
Originally Tegra was aimed at WM6.5, WM7, Symbian^3/4 and Android. Nvidia dropped all except Android.

IIRC original Tegra was only aimed at Windows CE 6.x and principaly WM7 aka Photon (not WM6.x at all) and Zune HD & Pink (Kin). Motorola was already one of the first OEM partrer back then working on several WM7/Photon based devices powered by Tegra scheduled to ship in Summer/Fall 2009 but then MS scrapped the whole thing and nVidia decided to go 100% Android (even thought they continued to do some CE on the side just in case). WP7 was never an option btw so nVidia didn't really have any other choice (don't even think that they even looked really hard at Symbian).
 
Which goes completely against what you just said.

Omap4 has dev kit, has design wins, and is likely to launch in H1 (probably with playbook and possible other RIM devices), so it'll definitely compete with Tegra2, Whether it will also be competing with Tegra3 later depends on Nvidia executing correctly, but at the moment, there *IS* no tegra3. TI generally does its Omap announcements at MWC so we'll see in Feb what its OmapX plans are, then.

H1 for real products is very, well, optimistic. There will be some design wins announced for OMAP and like you said the BB PlayBook will be the first real device to use it, but whether BB can actually get that out of the door in a respectable state is yet to be seen. From what I know it is behind schedule and the original Q2 date is likely to slip to Q3 [Edit: Almost confirmed, BB launching the PlayBook on Sprint in the 'Summer' rather than Q2] which does OMAP into competition with Tegra 3. If TI can only get a single OMAP4 device out by Q3 which is when Tegra 3 details will start to filter out then they are competing with it.

Also, the upgrade path from Tegra 2 to Tegra 3 is said to be quite easy, products can be adapted quite late in the design phase to include Tegra 3 because it has a similar power signature and the architecture is still going to be based on A9. My guess is that Tegra 2.5 will be dual 1.5GHz A9s and Tegra 3 will be quad 1.2GHz A9s with Tegra 2.5 launching in Q2 to slot into existing T2 designs and Tegra 3 will launch late in 2011 with 10-12 design wins and 1 or 2 products shown at CES 2012. All speculation ofc, but it is my job to guess this stuff.
 
Last edited by a moderator:
Back
Top