Intel Corp (INTC)

[wbmw]

Isn't Tegra K1-32 still on 28nm planar process while Haswell is on Intel 22nm with the vaunted "TriGate" 3D transistors? ......what's up with that??

Process leadership is often necessary - but sometimes insufficient. Haswell is a PC design, and scales to 4GHz speeds, at the higher TDPs. Tegra K1 is great as a 6-8W processor, but you'll neither get great 3-5W performance if you scale-down, nor will you get >3GHz speeds if you scale-up.

There's also an argument to make about the architecture. Haswell reuses the graphics architecture of Ivy Bridge, while you can argue that nVidia's Kepler is quite a bit better. From what I can see with Broadwell, not only can Intel score better than Haswell 15W - but it does so at around a third of the power (I'll note that the 50k score could have been at the "config-up" TDP of 6W, instead of the nominal TDP of 4.5W). So I think Broadwell gets the benefit of both the process shrink, as well as the Gen8 architecture enhancements.

Of course, nVidia seems to have done miracles with the Maxwell design, showing between 1.5-2x the performance/watt of Kepler - and which they will integrate into Tegra next year. So it may take Gen9 and Skylake before Intel can get ahead of nVidia at their own game.