I don't think this is a particularly fair test of ARM and x86 as architectures, though it is of Tegra 3 and Cloverview as specific systems. Cloverview just came out, but Tegra 3 is nearly 2 years old. Cloverview is on Intel's 32nm process node, whereas Tegra 3 is on TSMC's 40nm node (and remember that Intel's 45nm node actually had better performance than TSMC's 40nm).
And I've never quite understood NVidia's design choices with Tegra 3. All the other ARM SOC vendors used a low power version of the process node they were on for their ARM cores, but NVidia used the standard process version for 4 of their cores and the low power version only for the 5th core. And yet those four cores are clocked only slightly faster than the cores of their competitors, while their 5th core is clocked way lower.
Now, NVidia's GPUs are top notch and while it hasn't been that successful in recent generations of phones I assume that NVidia's existing Windows software compatibility is why it was selected for the Surface.
EDIT: I seem to remember that there were some big issues on the GPU front a couple of years ago with NVidia's design tools and TSMC's 40nm process having something of an impedance mismatch. The big improvements between NVidia's GPU's from that time to their current generation attest to these issues being sorted out.
It got Nvidia in the best selling tablet of the year, the Nexus 7. Technical merits aside, it was a win for Nvidia's bottom line.
I still don't get how embedded chipset manufacturers can keep competing with the largest graphics company in the world in the gpu space so easily though. Tegra 4 is supposed to be jumping from 12 up to 72 in the highest end model Kepler cores instead of old 8000 cores. If they manage to get some super efficient high performance graphics in mobile I could see a games revolution around it.
Especially if they put that in the Nexus7#2 next year.
"I still don't get how embedded chipset manufacturers can keep competing with the largest graphics company in the world in the gpu space so easily though."
Just for some perspective:
Samsung total revenue: $247 Billion
Apple total revenue: $157 Billion
Qualcomm total revenue: $19.1 Billion
Texas Instruments total revenue: $13.7 Billion
NVidia total revenue: $4.00 billion
NVidia is the relatively small fish in the pond here, and if they have relatively more graphics experience the others have relatively more experience with everything else in this space.
The Tegra 3 cores are actualy older 7000 series (prior to the whole CUDA revolution) cores, much the same as you have in a playstation3. Yes the Tegra4 will make a huge jump to teh kepler cores, so will be hard to compare core counts. Certainly will afford a very powerful platform. Tegra3 is how old now and works realy well.
And I've never quite understood NVidia's design choices with Tegra 3. All the other ARM SOC vendors used a low power version of the process node they were on for their ARM cores, but NVidia used the standard process version for 4 of their cores and the low power version only for the 5th core. And yet those four cores are clocked only slightly faster than the cores of their competitors, while their 5th core is clocked way lower.
Now, NVidia's GPUs are top notch and while it hasn't been that successful in recent generations of phones I assume that NVidia's existing Windows software compatibility is why it was selected for the Surface.
EDIT: I seem to remember that there were some big issues on the GPU front a couple of years ago with NVidia's design tools and TSMC's 40nm process having something of an impedance mismatch. The big improvements between NVidia's GPU's from that time to their current generation attest to these issues being sorted out.