Nvidia announced its Project Denver CPU initiative back in 2011, and today, we learned when the program might bear fruit. During his GTC keynote, Nvidia CEO Jen-Hsun Huang revealed that a Denver-based Tegra processor is coming in 2015. Dubbed Parker, this SoC will combine 64-bit ARMv8 CPU cores with integrated graphics based on the next-gen Maxwell GPU architecture.
Parker will be fabbed using "3D" FinFET transistors, but it's unclear who will handle manufacturing. We don't have any other details about the chip other than the promise of a 100X increase in "performance" over the Tegra 2. It's unclear whether that boost refers to CPU performance, GPU performance, or a combination of the two.
Another Tegra chip is due before Parker arrives. This one is code-named Logan, and it will combine 32-bit ARM cores with a new GPU based on the current Kepler microarchitecture. Logan will enter production early next year, Huang says, and it will be compatible with both CUDA 5 and OpenGL 4.3. There's no word on the fabrication process for this puppy, but adamantium interconnects are presumably out of the question.
Huang apparently didn't want to wait until next year to combine CUDA support with an ARM-based processor, so he tasked Nvidia's engineers with creating Kayla. This tablet-sized circuit board pairs a Tegra 3 processor with a "brand-new, low-power GPU" that may be related to the Kepler derivative planned for Logan. The graphics processor boasts programmable geometry shaders and supports OpenGL, CUDA 5, and PhysX. It attaches to the Tegra chip via PCI Express.
We don't know how Kayla will be offered as an actual product, but a working example was shown running ray-tracing in real time. Kayla supports Linux, and I suspect it will appeal to folks looking to experiment with low-power GPU compute applications.