Nvidia Pursuing Tensor Cores

http://www.tomshardware.com/news/nvidia-tensor-core-tesla-v100,34384.html

Tensor cores and Google's Tensorflow are digital approaches to solving problems in a significantly faster and more efficient analog fashion. Digital circuitry and programming is inferior in almost every respect to analog, but makes much greater error correction possible causing it to be the only viable way to go when you can always count on the chips themselves shrinking and making everything more efficient, but those same chips now have billions of parts that require endless error corrections right up to mass production. As the author points out at the end of the article, despite its advantages other hybrid approaches could prove more valuable in the long run. Google and Nvidia adopting tensors reflects the fact both are more invested in the programming aspect of the business than producing the fundamental hardware and can leave that aspect up to Intel, AMD, and others to fight over, while they continue to build the software ecology they require to dominate their niches.

However, what this announcement highlights is the fact that Python looks likely to replace any number of programming languages in the near future including cutting into MicroSoft territory, which is always good, and until the hardware issue is settled more clearly sometime in the next couple of years there's just no telling where this is all leading in the long run except to say everything is becoming more analog and adopting more scalar and open source architectures.