NVIDIA Tegra Architecture

This One X+ looks very powerful. Does it have USB host? This thing is a computer, so it would be useful.

 

Yes, not too shabby at all.

One interesting thing to note is that, even in battery saver mode, the HTC One X+ will still reportedly make use of up to four CPU cores (albeit at reduced peak operating frequencies). On the other hand, the LG Optimus G (with quad-core Krait Snapdragon S4 CPU) in battery saver mode will reportedly shut down two of the four CPU cores, and in this mode will make use of up to two CPU cores.

So the fifth "companion" CPU core on Tegra 3 should give enhanced battery life during light use using one CPU core, and the new battery saver mode should give enhanced battery life during heavy use using up to four CPU cores.

 

It should, beta Jelly Bean roms for the One X have it enabled, though the testers have only shown it connected to flash drives and hard drives.

 

It's fun. I'm feeling more light hearted about smartphones. I was thinking of connecting keyboard, ethernet and sound card right away.
I've decided to be interested when 4G is there (it's hopefully efficient, long range, cheap..) and when it's on cheap enough phones, it will also be widespread on phones and with decent coverages.

It might be good if nvidia was to make a crippled Tegra, two disabled cores and less GPU. Else maybe Tegra 3 can become a cheaper offer when Tegra 4 is out.

 

Some Tegra 4 news

http://www.fudzilla.com/home/item/29257-tegra-4-to-come-at-ces-2013

 

Isn't twice the performance of Tegra 3 still going to put it under pretty much every other SoC coming out in the next few months?

Samsung Exynos 5 series and Apple A6/A6X for example.

 

GPU yes, CPU no. If it has quad Cortex-A15s before the competition does it'll be able to boast that.

 

I don't expect Fudo to be able to connect the dots and see where there are several oxymorons in the stuff he's claiming. As a start I'd be VERY surprised if that ancient roadmap showing Wayne being 10x faster than T20 and only 2x over T30 is either severely outdated or someone just made a simple mistake and placed Wayne just a notch to low on that roadmap scale and the difference is actually significantly bigger within the typical nonsensical marketing measurements which no participant in the market is actually immun to.

Wayne is supposed to sport a quad A15 CPU with a significantly enhanced GPU and it's only going to be twice as fast than Tegra3 while Tegra3 was supposed to be 3x times as fast as Tegra2. No idea if it's just me but there's a gigantic loophole in that reasoning. Even worse how can it be only twice as fast when Fudo itself claims that it signficantly outpaces competing SoCs like the S4?

Assume the GPU as rumored consists of 64SPs clocked at 500MHz that alone would suggest a much higher performance increase than quoted for the entire SoC. We'll eventually find out what is going on and none of the above is necessarily true, but I'm sorry when someone starts to write a newsblurb like that I'd expect him/her to be able to detect such huge loopholes in common sense.

If Wayne turns out what I'd expect it to be, I'd expect a rough up to 4x times performance increase compared to T30 at least. Wayne is supposed to be able to scale from superphones up to clamshells according to a way more recent slide.

 

I think Fudo is just parroting what that Nvidia rep stated a few months ago. I doubt he actually knows what Tegra 4 is, the date however might be correct

 

Looks like Tegra will be of extreme importance to NVIDIA in the coming years. Tegra will be the building block for Tesla (and presumably Geforce and Quadro too). See the Exascale computing section towards the latter half of the SC12 video clip here:

http://www.ustream.tv/recorded/26957114

With Tegra 4 (aka Wayne), Tegra will be scaled up to clamshell-type devices, but the end goal appears to be much more grand than that for future generations of Tegra. Perhaps not too surprising, but certainly much different than the top down approach that they have used in previous years.

 

It doesn't come as a surprise at all, given all the so far announcements for SoCs in general from NVIDIA especially those including Project Denver. The way I understand things the custom designed Denver GPU will serve (probably amongst other things) to scale up to bigger SoCs for notebook/PC SoCs. The possibility that NV might view all those projects under the "Tegra" marketing umbrella, isn't something I'd object to. After all Jensen projected some time ago that some time in the future the Tegra department will account for roughly 50% of the company's total revenue and IMHO it cannot come from small form factor SoCs alone.

Besides Denver there's also Boulder in the works as it seems: http://www.xbitlabs.com/news/cpu/di..._ARM_Based_Boulder_Microprocessor_Report.html

http://www.xbitlabs.com/news/cpu/di...ARMv8_64_Bit_Processor_with_Secret_Sauce.html

 

Did anyone notice that the fifth battery saver CPU core on Tegra 3 is still under development for Windows RT?

http://www.nvidia.com/object/tegra-3-processor.html

Surely that must mean that devices such as Microsoft Surface, Asus Vivo Tab RT, and Lenovo IdeaPad Yoga 11 are not making use of the fifth CPU core?

 

I wasn't aware that the companion core had any opacity to software at all (that's how nVidia seemed to be presenting it). I mean, it even says that in the bullet point, doesn't it? Of course, even if this is the case, it'd never actually kick in if more than one core was powered on. If Windows RT isn't powering down cores that's a more fundamental and dire problem.

 

It's the case, but a software update will enable it.
The thing is, with a linux kernel, you can modify the scheduler and include it yourself, as a CPU maker you could have kernel support for your product while it isn't even out of the R&D lab.
With Windows, only Microsoft can write the new kernel code and they only push it when they have tested it themselves a lot, they can't afford to release buggy code and put out another point release two weeks later etc.

This is an area where free and open source software allows support for features to come faster, similar thing with the Bulldozer scheduler patch for Windows 7 which took relatively quite some time.

(on the other hand, the opposite is true for 3D graphics drivers. Intel commits to open source for its linux graphics drivers but nvidia pushes strong proprietary drivers and doesn't collaborate with any open source one)

 

Another Tegra 4 update (not yet verified information)

http://www.chiphell.com/thread-616838-1-1.html

6X Tegra 3.. Curious about how they have implemented their Shadow/Ninja core in a Cortex A15 setup. It cant be Big.Little no?

 

That's for GPU performance:


The problem is that if it's 6x Tegra 3, then it can't be 20x Tegra 2, because Tegra 3 isn't 3.33x faster than Tegra 2.

This probably means that Tegra 4 is just some 3-4x faster than Tegra 3 in GPU performance: just barely enough to level with the Snapdragon's Adreno 320 in smartphones, but it will have a hard time dealing with the Mali T604 in Exynos 5 for tablets.

However, with TI and ST-Ericsson apparently out of the game, Exynos being almost exclusive to Samsung and the demand for S4 Pro being too great for higher-priced smartphones, one can guess that Tegra 4 will be the SoC of choice for mid/high end tablets during 2013.

 

These are obviously "up to" figures.

At this stage and exclusively based on GLBenchmark2.5 results the Google Nexus 10 (T604) is by 2.77x times faster than the best Tegra3 GPU score and the Google Nexus 4 (Adreno320) by 2.55x time faster than T30. It's also interesting to note that the highest T30 is just a slight notch above the Adreno305 which I assume is clocked at around 400MHz just as the 320 (being a single cluster variant of the four cluster 320).

I'd expect the Wayne ULP GF to be clocked at =/>500MHz, for which depending if those "72 cores" are actual ALU lanes or SIMDs + SFUs? (which has become an annoying trend as of lately for everyone out there) should be =/>68 or =/>72 GFLOPs.

Not that 3-4x times faster in real time isn't unrealistic for Wayne compared to Kal-El, however at 3x times they'd still be somewhat faster than the former two above and at 4x the difference is rather significant.

TI announcements state that they'll honor their OMAP5 design wins; no idea if there are any tablets amongst those. No idea about ST-E.

Think compiler/drivers and windows tablets for NVIDIA and things can become quickly hairy for all competing solutions, Mali T604 and Adreno320 included.

 

Using the exact same layout as the Tegra 3 diagram in a way that could easily be photoshopped and saying "6x graphics" with "72 cores" (i.e. 6 times Tegra 3's 12 cores) which doesn't fit Kepler at all makes this very suspicious. It could be right, but I don't buy it so far.

 

Not that it matters at this stage since no official announcement means anything, but for the GeForce block that slide has 240 borgo sphinxter squares