TFlops needs to stop being some be-all-end-all metric. I have been PC gaming for years and years and one thing you quickly find out is that you cannot just look at TFlops and determine a system's capabilities (it gives a rough idea).
TFlops of the Tegra X2 is measured at FP16 (source), which is half that of FP32 that XB1 and PS4 are measured at (source needed for specifics, but industry standard is FP32, which is why nVidia mentioned they measured in FP16 as it is not the norm). So that roughly equates to 750 GFlops of FP32 performance (nearly half that of the XB1). (Pascal can run 2x FP16 instructions or 1x FP32 intructions per clock. source)
Then there is the CPU aspect. ARM vs. x86 is hard to compare directly these days, but x86 is almost always more powerful than ARM, even at similar clocks.
RAM is another aspect and with LPDDR4 for Tegra X2 @ 50 GB/s (source), this cannot compare at all with XB1's 68.3 GB/s and PS4's 176 GB/s (source).
Lastly, let's look at the CUDA cores - 256. While this is a new architecture, it is just Maxwell 2.0 so we can guestimate where this GPU lies. The only nVidia GPU we have close to that is an 920MX mobile GPU (comparisons), which has about 550 GFlops (source). Now sure, this is based on a 28nm Maxwell architecture, so it will definitely be faster, say closer to 750 GFlops? As for comparison, the XB1 GPU is similar to an AMD Radeon HD 7770 (specs) and the PS4 is similar to an AMD Radeon HD 7870 (specs). Therfore, there is no comparison at all as it is not even in the same class as the XB1 or PS4.
People please temper your hype and expectations. You will be dissapointed. I will not because this thing would still blow the Wii U out of the water, and I love my Wii U.
Last note: most of our specs for Tegra X2 are from nVidia's presentation of their Parker SOC for Autonomous Vehicles (source) which means Nintendo will probably have a different, customized design that may or may not be more powerful.
