- The big increase in number of transistors comes because Nvidia is using a 16nm FinFET process. The chips will be fabricated by TSMC.
- The current generation, Maxwell, is based on a 28nm process. Pascal is expected to be commercially available in 2016.
- Pascal processors will support 3D memory and mixed prevision floating point calcuation, which are both useful for deep learning.
In 2016, Nvidia's Deep Learning GPU Will Have 17 Billion Transistors
7. 27. 15 by Kif Leswing