Google is set to release chip specialized for AI call TPU in October
 
Software developers are becoming more and more aware of the importance of having high grade hardware for development. In 2017, the GPU and graphics card market reached an all time high in sales due to the cryptocurrency craze. Now with rapid developments in the field of AI, there is an ever increasing number of programmers wanting to get in the game. Google is helping answer this call as it has announced that it will release their proprietary AI chip to the public called the Tensor Processing Unit or TPU.
 
The TPU is a specialized ASIC processor for accelerating the computation for neural network development. Similar to a GPU, it is very good at high volume of repetitive computations, but the difference lies in the TPU having higher IOPS(input/output operations per second) per watt and does not support rasterisation and texture mapping capabilities.
 
Machine learning applications are also heavy in vector and matrix computations. Up to this point, Google had publicly offered usage of their TPU through Google’s Cloud Machine Learning Engine. Just last month, they have announced their new initiative called Edge TPU which will allow developers to physically install the TPU chip onto their own machines. The TPU will come as a component that plugs right into the motherboard’s PCI express lanes.
 
In addition to this, there is a new set of services called the Cloud IoT Edge which is a software stack that allows the implementation of Google Cloud AI system to IoT devices. This comes with on the foot of the new light weight version of Tensor flow which uses less power and reduces latency.
 
It seems this all encompassing reach is Google’s attempt of owning the whole AI stack by keeping developers within their ecosystem. The AI services market has been heating up recently with Amazon currently owning most of the stack for cloud computing with completion from Microsoft’s Azure. Microsoft had earlier this year announced their own AI chips through what they call Project Brainwave. We have also seen small start ups such as Wave Computing entering the AI market. On the software field has multiple options out on the market as well with examples like Caffe2 and PyTorch.
 
These bold steps from Google will be sure to accelerate the number and development platforms for programmers to use to develop their own machine learning programs. The proliferation of GPUs have also made machine learning development more doable for the average developer, which may put additional challengers for Google to compete with. For those who are interested, Google cloud AI customers can apply for early access to Edge TPUs through this link:
Here

Original Article: https://www.cnet.com/news/google-to-let-you-pop-its-ai-chips-into-your-own-computer-as-of-october/?utm_source=reddit.com