CES 2015 officially begins on Tuesday, but NVIDIA have kicked off the event early with their now almost traditional Press Event which pre-empts the show. The same event last year was Tegra K1's first public outing, marking a new high-water mark in mobile computing power and efficiency, and also outed NVIDIA's current infatuation with smart automotive technologies. No surprise perhaps that this year was more of the same, only better
NVIDIA CEO Jen-Hsun Huang once again anchored the keynote, and hit the ground running with the unveiling of their new mobile computing platform: Tegra X1. Tegra K1 was based on the then cutting-edge Kepler GPU technology, and Tegra X1 goes one better by incorporating features of their new Maxwell desktop architecture. In so doing it aims to take an undisputed lead in mobile computing power and efficiency, a crown briefly relinquished with the release of Apple's A8 (although they still had a claim to it).
Tegra X1 incorporates both 256 Maxwell-based CUDA cores and an eight-core 64-bit CPU part for robust serial and parallel computing potential. Not only capable of 10-bit H.265 VP9 video, it also supports many of the very latest graphics technologies including DirectX 12, OpenGL 4.5, CUDA and OpenGL ES 3.1 and the Android Extension Pack; how relevant these technologies are remains to be seen but the headline figure you'll see is Tegra X1's raw teraflop computing power, the equivalent of a 2000-era supercomputer.
In order to demonstrate the power of Tegra X1 NVIDIA were running Epic's taxing Elemental Unreal Engine 4 demo. It's not clear exactly which configuration options were being used, but they claim that Elemental was ported to the platform with a minimum of fuss and bother. Either way, Tegra X1 is clearly the superior of Tegra K1, itself already an accomplished gaming GPU for mobile devices.
Of course, NVIDIA are keenly aware that the market for mobile tablets and laptops is already highly competitive, and so they continue to branch out into a reltively untapped market - computing for automotive applications. The first salvo's were fired last year, and this year NVIDIA bring their DRIVE development environment to the table along with two hardware SKUs for self-driving and in-car infotainment.
NVIDIA DRIVE PX is a dual-Tegra X1 computer with inputs for up to 12 cameras, and hence highly desirable for self-drive systems. By making full use of this data gathering a DRIVE PX-powered car has the potential to be safer and have more robust 'Auto-Valet' capabilities (in that it will park itself without any user input). The efficient nature of the Tegra X1 means that whilst highly capable, the additional power requirements of said functionality won't be onerous.
However DRIVE PX isn't simply for collating computer vision information. NVIDIA are using the processing power of the platform to enable what they're calling Deep Neural Network detection, a critical component to creating robust algorithms for self-drive. Contextual object recognition requires extensive neural network training which is highly time-consuming, and accelerating this process can shave huge chunks of time from development. Essentially, if you want a car to recognise what it sees and either alert the driver or adapt its self-driving accordingly, this learning is crucial.
By contrast NVIDIA's DRIVE CX, powered by Tegra X1 or K1, caters to the infotainment side of the coin. Able to control lighting systems with up to 16.8 million pixels (around 10 times the current best) it has the potential to turn pretty much any surface into a display for video and touch-control. In theory whole driver cockpits can be set up with interactive dashboards, providing data or external video in real time whilst simultaneously mimicking real-world materials.
By utilising NVIDIA DRIVE STUDIO, DRIVE PX and DRIVE CX allow the creation of complex digital cockpits that have the full range of sensor information at the drivers fingertips. Furthermore these designs can mimic the look of real-world materials to a more realistic extent than ever before. Then by using the power of NVIDIA DRIVE PX hardware developers can create cars which can auto-valet and eventually perhaps drive themselves as a matter of routine.
For more information on NVIDIA DRIVE and the TEGRA X1 check out the NVIDIA official blog, with more coverage no doubt on the way.