Google Tensor Processing Unit

If you have seen the Google’s I/O 2016, you might have known about Google Tensor Processing Unit, or TPU.

Why TPU is important

As machine learning guy will know, current industry trend is doing machine learning especially related to image processing one with GPU. A well known example is doing image convolution with CUDA. However GPU is mainly for graphic processing but not machine learning, therefore it is not functioning to the optimal.

And IMO this is the reason google releasing TPU. TPU will be optimized for doing machine learning training, and I foresee it will come with first class support with Tensorflow.


Google is aiming big for the A.I., and TPU is one of the strategy to make them stay in the main stream. And this is also a symbolic of that maker spirit is on the rise - traditionally it would be asking vendor such as Intel to produce these things.

Written on May 20, 2016