Lots of power, but for what?
From CES in Las Vegas literally comes a bombshell for the mobile market. NVIDIA has indeed launched its own SoC TEGRA X1 capable of 1 Teraflop processing power. The new 64-bit octa-core Superchip upsets the plans announced last year by the American manufacturer, introducing ERISTA technology a year in advance.
The new SoC will match the power of Maxwell GPU, (already used in the very recent Nvidia 970 and 980) with a processor ARM Cortex A57 or A53, all produced with a 20nm process.
The first, theA57 has 2MB of L2 cache shared on 4 cores, with 48KB / 32KB of L1s cache per core. L' A53 cluster has instead 512KB of L2 cache shared across the four cores and 32KB / 32KB of L1s cache (I $ + D $) per core.
NVIDIA will use proprietary connection technology and cluster migration rather than the more common technology of Global task scheduling which exposes all 8 cores to user applications. The NVIDIA solution is though cache coherent, this means that the SoC does not suffer from the power / performance problems common to other systems of this type.
The new TEGRA X1 is able to surpass even the Samsung LSI's Exynos 5433 as regards the performance per watt with a 1.4x increase in power with the same amount of energy consumed, while the gap widens, reaching 2x of performance per watt compared to the solution Apple A8X. All this is also due to the choice of customizing the consumption management system, no longer the standard ARM one, but built in house by NVIDIA.
Besides CPU and GPU also all other features of the X1 chipset compared to Tegra K1. There memory is now DDR4 which translates into a expansion of the memory band from 14.9 GB / s of the DDR3 solution to the current 25.6 GB / s. The maximum resolution also moves from 3200 × 2000 at 60 Hz to 3840 × 2160 at 60 Hz; HDMI 2.0 and HDCP 2.2 support has also been added which means 4K at 60fps is now supported. In terms of APIs they are supported DirectX 11.2 and OpenGL 4.5 and in the future, DirectX 12.
Specifically the new NVIDIA X1 SoC will guarantee a power of just over 1 teraflop which makes its performance in the graphics field, according to the manufacturer's statements, equal to those of an Xbox One and with 10 times lower consumption.
The figure appears slightly exaggerated if we consider that the processing power of the Xbox One GPU is equal to 1.33 teraflops (against 1.84 of the Sony solution with Playstation 4) but that nevertheless the power of the new SoC is a peak power, hardly in the field mobile, in fact, the approx Clock 1Ghz of this new processor, without dramatically affecting consumption and heating. More realistically (also seen how it went with the Tegra K1 which had to outclass the computing power of PS3 and X360 by twice), wanting to compare the new SoC to a console, the computing power under normal conditions of use of the new TEGRA X1 is comparable to that of PS3. The situation is different in the automotive field, where the desktop class performance of the new SoC can be released more easily.
In fact, NVIDIA has already announced that all this power will be destined for new automotive infotainement systems and the new generation tablets (probably the new Shields), however we ask ourselves and the question seems more legitimate than ever, if it is really necessary to access this enormous potential on the move. The answer could be the new Nvidia Drive PX an autoguiding system which, thanks to the processing power of two Tegra X1 units, combined with 12 cameras should allow a fully autonomous and low-cost autoguiding system. The first X1 solutions should arrive during the first half of the year. We just have to wait to evaluate the new products based on this technology.