NVIDIA Tegra: The Mobile Charge
Its not X86, but who cares?
Today NVIDIA is announcing its Tegra line of products which are aimed at the mobile market and are designed as essentially a “system on a chip”. This is not exactly NVIDIA’s first foray into the world of mobile devices, but it certainly is their first that contains the graphics core, southbridge, and a full Arm 11 CPU. We are not quite at the point of saying watch out Intel and AMD, but from what I have seen so far the Tegra lineup is pretty impressive considering what it is able to do and how it is able to do it.
NVIDIA licensed the Arm 11 CPU for the Tegra line, which is a low power and small die size product. This may be the “core” of the Tegra line, but it may not necessarily be the most important part. Looking over Arm’s site about the Arm 11 architecture, it was originally designed as a 130 nm part with performance up to 2.1 MIPs and scoring around 740 Dhrystone. I doubt that NVIDIA will be using a 130 nm process node to create this part, so it is quite likely that performance of the integrated Arm 11 core could be higher than the above mentioned figures.
In our line of thinking we often consider the CPU the core of the platform, but that is beginning to change. In the Tegra line the CPU acts as the conductor of the platform, but most of the performance features that NVIDIA is pushing are actually accelerated by the graphics portion of the platform. Consider how video decode and encode has gone from the CPU to the GPU in the past several years. This is merely an extension of that overall trend, but trimmed down to the very basics. The graphics portion is actually comprised of three primary “units”. The first is the image processor (picture manipulation), the next is the HD video processor (encoding up to 720p and decoding up to 1080p), and finally we have the 3D graphics core which controls the 3D interface as well as running 3D content. The 3D portion is supposed to be able to run games like Quake 3 at handheld resolutions with full anti-aliasing and anisotropic filtering at around 40 fps.
NVIDIA also is producing the portion which controls the “southbridge” functionality. These are primarily the I/O operations and include USB support and IDE, as well as the memory and internal flash controller. The rest of the chip handles the display functionality as well as HDMI output.
The fully packaged chip is tiny. While the Tegra is not a direct competitor with Intel's Atom, it is also 1/10th the size of the Atom and its supporting chips.
The best part of this product is the amount of power it consumes. At idle it uses less than 100 mW, and when decoding a HD video stream at full power it is consuming around 2 watts total. On a single battery charge, NVIDIA expects around 25 hours of video playback and around 130 hours of audio playback. Of course, I am rather curious what kind of battery they are talking about for these applications. The first products are aimed at the large handheld market, as well as the smaller tablets and sub-notebooks.
NVIDIA is launching two products based in the Tegra line. The lower performance one runs at a core speed of 700 MHz, while the snappier version hits 800 MHz. The Tegra 600 runs at 700 MHz and can handle encoding a 720p video stream as well as decoding 720p at 30 fps. The faster Tegra 650 runs at 800 MHz and can encode at 720p while decoding at a max 1080p at 24 fps.
The products utilize LP DDR memory running at 166 MHz and 200 MHz speeds. This should give plenty of bandwidth for most applications that these products will run. These parts are not meant to play games like Crysis in high definition, but they certainly seem to be able to handle a fully functional 3D UI at decent resolutions.
The most impressive part of this product line is the size. All of this functionality is fit in a chip that is 144 mm square. NVIDIA did not disclose what process node that the product is going to be produced on, but we expect it to be either on the 65 nm or the 55 nm process. While it is slightly smaller than products such as the GeForce 8500, it does contain all of the functionality to run a basic system from one chip. And it seems to do it quite well when we considering the targeted market for these products. The size of the fully packaged chip is right around the size of a US dime (which has about the same surface area of an average sized fingernail).