As CES 2013 approaches closer, we're beginning to see more and more information leak out. Some of the information is willingly provided, and other information is leaked out without the approval of the company. While it is generally difficult to say which of the two it is in the situation of the Tegra 4, we cannot ever fully confirm these rumors as true until the company itself makes the announcement.
If Nvidia is in fact making their Tegra 4 announcement at their CES 2013 press conference on January 6th, that leaves them with a little over a month and a half until Mobile World Congress 2013. If the schedule stands, the company could announce the chip at CES 2013 with one or two partners, and then have a multiple partner product launch at the Mobile World Congress in Barcelona. The very latest leak comes to us via Chiphell, a popular Chinese rumor website.
Looking at what appears to be a leaked Nvidia slide, we can see that the Tegra 4 chip (Wayne) will employ a similar design to Tegra 3 - it will have a quad core processor with an added low-power chip for keeping the device alive while the screen is off. This SoC will be manufactured on TSMC's 28nm HPL process, which should give it significant power savings over 40nm Tegra 3. Additionally, it will support DDR3L, LPDDR2 and LPDDR3 in dual channel mode, which should improve memory bandwidth over previous generations.
The most important part of Tegra 4 probably is the implementation of the new GPU architecture. This GPU is supposed to support resolutions like 2560x1600 and 1080P at 120Hz for Real 3D gaming (similar to their desktop graphics cards). Also hidden behind the ChipHell logo is support for 4K resolution (but not video) over a single HDMI cable.
Furthermore, the slide claims that the high-end version of Tegra 4 will bring 72 cores in the GeForce GPU, without disclosing if the cores in question support CUDA or not. Remember that two years ago, Nvidia mentioned that their T40-class hardware will be fully compliant with the OpenCL and CUDA, and based on the Kepler architecture, which wasn't even announced at the time. The company claims a 20x GPU performance improvement over Tegra 2 and a 6x GPU performance improvement over Tegra 3. This GPU also enables full 1440P video, VP8 acceleration and H.264 High Profile. It also claims to support video encoding and decoding of up to 2560x1440, which means that this thing probably won't be able to encode or decode 4K video.
The other two interesting aspects of this chip are the fact that it is supposed to support Super Speed Imaging of sensors up to 350 Megapixels. Obviously, there are no camera sensors for phones that are anywhere near that resolution yet, so it's a hyperbolic statement at the very least. Having a lot of headroom, however, could enable OEMs to put in Nokia PureView-like cameras that oversample which result in better low-light and video performance. The other interesting thing is USB 3.0 support, which we believe to be a big deal. Right now, none of the current shipping smartphones or tablets have USB 3.0 which means transferring your movies and large amounts of music will take forever. USB 3.0 is needed for smartphones and tablets if they are really going to replace our desktops and laptops.
Overall, we're very interested to see what Tegra 4 will really have to offer and whether or not they went for a Kepler based solution with Tegra 4. If true, it would easily explain the GPU performance increase considering the generation jump from GeForce 6 NV40 series technology (launched in 2004, also present inside the PlayStation 3 console) to GeForce GTX 600 series technology with Kepler from 2012. Essentially making a 9 year leap in GPU generations, totaling 9 different GeForce generations of Nvidia GPUs. Yes, it took Nvidia a long time to make the leap, but if they really do it they could find themselves far head of the competition.