Tegra 3 refresh coming, 'Tegra 3+', In House LTE modem included

[Bateluer]

Gee, its a good thing Nvidia will be able to make their own LTE modems for their Tegra chips before the end of the year, isn't it? Now, a working 28nm process from TSMC, and Tegra 3+ would be more than competitive until Tegra 4.

Then we can start this entire conversation all over again, about how Tegra 4 is weak and poor compared to chipsets 6 months after it on the road map.

[alent1234]

its still a long way from building a usable SoC and modem

my mom was looking at a smartphone last year and tegra 2 was off the list because it didn't work with Skype at the time because nvidia didn't include some feature into the CPU

lets see what kind of stunt they pull with tegra3+. they still have to write decent software for it

[sontin]

Originally Posted by LOL_Wut_Axel

Wrong on multiple accounts. (1) Each core on Snapdragon S4 is significantly faster than the one's on Tegra 3; (2) Four Tegra 3 cores working in parallel are slower than two S4 cores in parallel, as has been shown numerous times in benchmarks. (3) The benchmarks themselves are able to execute on all cores, which is why Tegra 3 is faster than previous CPUs with dual-core Cortex A9s.

The problem is that even if the benchmarks peg all the cores, most apps won't. That, and the S4's architecture is much, much faster, which is why you have a humongous gap in single-threaded and still a big gap in multi-threaded. Tegra 3 is slower than Snapdragon S4, whether all four of its cores are being utilized or not. That's why IO was talking earlier about there being a lot of room in ARM to improve performance--because it's so behind x86 in that metric.

And yet Krait is beaten by another A9 QuadCore design in a Single/Dual-Thread application:

http://www.anandtech.com/show/5810/s...rmance-preview

[Ararat]

Originally Posted by sontin

And yet Krait is beaten by another A9 QuadCore design in a Single/Dual-Thread application:

http://www.anandtech.com/show/5810/s...rmance-preview

The difference there is in the software. Samsung's browser on the gs3 is clearly far better optimized. The article you got that chart from clearly stated that.

[sontin]

Originally Posted by Ararat

The difference there is in the software. Samsung's browser on the gs3 is clearly far better optimized. The article you got that chart from clearly stated that.

I know. And i said this about Krait, too.
But there are people who believed that the performance in Linpack/Sunspider/Vellamo would be the result of the cpu architecture.

Krait performs like A9. It will loose badly to A15 and will have no chance against the 28nm/32nm QuadCore A9 SoCs.

[dagamer34]

Originally Posted by sontin

And yet Krait is beaten by another A9 QuadCore design in a Single/Dual-Thread application:

http://www.anandtech.com/show/5810/s...rmance-preview

I honestly wouldn't put any stock into that chart, as it actually disproves what you want it to prove. You want it to say "Krait sucks, Quad-core is better!" The problem is that the HTC One X should technically beat the Galaxy S III, since it's clocked at 1.5Ghz instead of 1.4Ghz, and yet it looses dramatically. I would not be surprised if Samsung's optimizations of the S4 make it far more competitive that this chart would have you believe.

Which brings me to my point, it is disingenuous to compare the SoC in one device with another SoC in different device, and then state "SoC X is clearly better than SoC Y, look at this chart!" Software HUGELY influences the performance of these devices, which is why the iPhone 4S is still competitive with all of the 1.2-1.5Ghz devices in that chart, despite being clocked at a paltry 800Mhz.

Or to make the example far more obvious, why is the 600Mhz Cortex A8 CPU in the iPhone 3GS (from 2009) even remotely competitive with the dual-core 1.2-1.5Ghz Snapdragon S3 in the HTC Sensation 4G (from 2011)? Software.

[poofyhairguy]

Originally Posted by dagamer34

I would not be surprised if Samsung's optimizations of the S4 make it far more competitive that this chart would have you believe.

I would be surprised if Samsung optimized for the S4, since they did not optimize for Qualcomm SGS2s.

[dagamer34]

Originally Posted by poofyhairguy

I would be surprised if Samsung optimized for the S4, since they did not optimize for Qualcomm SGS2s.

See this image from Engadget: http://www.blogcdn.com/www.engadget....chmarkspng.png

HTC Rezound Review: http://www.engadget.com/2012/02/15/a...y-note-review/

A few benchmarks are within the margin of error, but the Skyrocket gets almost 1000 more points in Quadrant and 17 more MFLOPS in Linpack even though they use the same exact CPU. That's all software optimization right there. (I ignore the Nenmark scores because the Rezound has a higher screen resolution).

[poofyhairguy]

Originally Posted by dagamer34

A few benchmarks are within the margin of error, but the Skyrocket gets almost 1000 more points in Quadrant and 17 more MFLOPS in Linpack even though they use the same exact CPU. That's all software optimization right there. (I ignore the Nenmark scores because the Rezound has a higher screen resolution).

Quadrant is also very screen-resolution dependent and Linpack is a pile of crap on Android.

Optimization would happen in the browser, and they didn't do that for the Qualcomm SGS2 like they did with Exynos units. Look at the Sunspider scores.

[Bateluer]

You guys will poo poo any benchmark that doesn't show your SoC in favorable light. Quadrant is bad, unless it shows your cpu beating the competition. Linpack is bad because it shows your hated cpu on top. Sunspider is all browser, and any scores are a waste, until your cpu is on top.

This thread was supposed to be about Nvidia refreshing the Tegra 3 to over come its shortcomings. On 28nm and with an lte modem, its a perfectly competitive chip.

[poofyhairguy]

Originally Posted by Bateluer

You guys will poo poo any benchmark that doesn't show your SoC in favorable light.

That is more a reflection of the poor mobile benchmarks of our era.

This thread was supposed to be about Nvidia refreshing the Tegra 3 to over come its shortcomings. On 28nm and with an lte modem, its a perfectly competitive chip.

I'll agree with that.

[stormkroe]

does anyone know the plan on t3+ gpu-wise?
Also,

Originally Posted by sontin

Krait performs like A9. It will loose badly to A15 and will have no chance against the 28nm/32nm QuadCore A9 SoCs.

roflwut?

[fstime]

Too all you people posting benchmarks: You can sit there and tell me with a straight face that the phone experience (real world, not pointless benchmarks) going from a One X to a SGSIII is as dramatic as your little bar graph?

ROFL, Apples phones have been butter smooth for the past 2 years and Androids not to far after that (as long as bloat is removed).

Less of your kind and pushing for excessive 8-core phones and and we might have had decent battery life on Android sooner.

[ITHURTSWHENIP]

I think Wayne (Tegra 4) is going to be a completely different beast than we have seen before from Nvidia. If you think about it, why would they spend a lot of R&D money on a powerful SOC when it wont get any real design wins anyway because of the radio issue? With Icera in the package they will get a lot more design wins for android phones, add in W8OA tablets and Nvidia can spend more money on their SOC without fear of losing on the deal