The byline might lead you to believe that nVIDIA’s upcoming chips will be made on a 20nm process but to be more specific the information DigiTimes received tells them that Kepler will be 28nm and Maxwell 22nm.  This shrinking of the process plays right into nVIDIA’s main focus during the GTC which is efficiency.  The charts we have seen are specifically displaying GFLOPs per watt and not a raw performance increase, something that requires shrinking the die.

"Nvidia CEO Jen-Hsun Huang at the company-hosted GTC 2010 event, said that it will announce its new GPU designs, Kepler at the end of 2011 and Maxwell in 2013 using 28nm and 22nm process from Taiwan-based Taiwan Semiconductor Manufacturing Company (TSMC).

Huang pointed out that both GPUs will have significant performance improvements in floating-point calculations, and will feature several CPU-like functions such as virtual memory.

Huang also announced that Nvidia is already cooperating with PGI to develop CUDA-x86, which allow a CUDA program to be recoded into a x86-based program, meaning PCs will be able to run CUDA software without requiring an Nvidia GPU, which will expand the ecosystem for CUDA programs and their developers."

Here is some more Tech News from around the web:

Tech Talk