All we know About Amazon’s New Machine Learning Chip “Amazon Inferentia Chip”

“Artificial Intelligence is the new electricity” – Andrew Ng

We are witnessing the unfolding of a golden era: The era of Artificial Intelligence. Questions like: Will AI destroy humans or not makes it more interesting and exciting.

Everyone wants to be a part of this revolution. Big TechGiants like Google, Facebook, Microsoft are already using this technology to empower their products and to bring more value to the table.

The AI-powered products are computation intensive which means they require a lot of computing power to function. The rule of thumb in this domain is- The more computation resources you have, the easier your life will turn out to be. A machine learning engineer has to wait for hours, days or even months too to train their machine learning models just only because of computation power involved in these things.

The future of this technology heavily depends on the development of dedicated and specific purpose hardware chips which are specifically made to do the computation required in these products. Companies like Nvidia and Intel are already manufacturing the chips to ease the process of development of AI-powered products. So other companies have to buy these chips from the chip vendors to make use of this technology for their product and services.

Last year in November 2018, Amazon entered the list of these manufacturers by announcing its own machine learning chip called INFERENTIA or amazon inferentia chip.

Now, of course, we have a lot of question like Why Amazon decided to manufacture their own chip, why they do not want to simply buy the chips from other OEMs, and what is the mystery behind the name amazon inferentia chip? To understand this, we have to dig deep into this machine learning world.

Phases involved in any machine learning project

Mainly, there are two steps involved in any machine learning powered product or service, Training, and Inference. Training is the process in which the machine is trained to learn the patterns in the given data, this is where the machine gets smarter by learning the complex mathematical function from the data. Training is a one-time process which means that you’ve to only train a machine once for a specific dataset and then it will be ready for inference phase. It’s just like how a professor teaches his or her students.

The inference is the phase where the AI-powered product provides value to others. The inference is the phase where people use that system for their use cases. This is where the product is actually used and it’s definitely not a one time process, there are millions of people making use of these products at the same time. The usefulness of that system is actually determined by how that system responds in the inference phase. And it’s like a student using the learned knowledge in the real world in various situations.

Till now, Amazon had been using the chips made by Nvidia and Intel for its AWS or Amazon Web Services. But now finally Amazon has come up with this new dedicated chip designed specifically for the inference phase, which is quite rightly named amazon inferentia chip.

The main thing about amazon inferentia chip is that it almost supports all of the major frameworks used in the deep learning community like Tensorflow (A deep learning framework by Google), Apache MXNet, and PyTorch(by Facebook), as well as models that use the ONNX format.

This feature of amazon inferentia chip is that it will help the deep learning practitioners a lot, as they can develop the models using any major framework. However, this chip will not be available directly for sales but Amazon Web Service will use this chip for providing computation resources to users. So it’s not going to directly harm the business of already established chip providers.

In the meantime, if you are keen to know more about AWS and cloud computing then you can explore the various sections of Cloud Computing With AWS From Scratch teaching all its concepts like AWS system administration, AWS infrastructure and much more.

Specifications of AWS Inferentia or Amazon Inferentia Chip:

This chip will provide hundreds of tera operations per second (TOPS) of inference. That will allow complex models like video captioning to make fast predictions during the inference phase. More than one Inferentia chips can also be integrated together to make even more fast predictions, in case a huge number of users are using that product at a time.

The Amazon Inferentia Chip will be available to users by the late 2019 and they can access it via Amazon Web Service. The use of this chip will ease the process of development of Artificial Intelligence-related products and services for researchers and will reduce the cost involved in these things.

The announcement of this chip by Amazon clearly signifies that all of the big tech giants are working hard to accelerate the growth of their companies by using this technology and they want to come up with their own dedicated hardware and software tools to develop these products as per their preferences.

How it can help me?

Can you observe the pattern here? The new revolution is in front of you. Don’t just witness it, be the part of it. Big tech giants are already using it for their use cases. Learn the required skills which are going to massively pay off in the future. Make use of software tools like Tensorflow and hardware products like Inferentia to develop more AI-powered products and services. If you’re quite familiar with machine learning and deep learning, you can make use of Inferentia(whenever it’s available) to deploy your models in the real world.

In order to understand and learn Machine Learning with TensorFlow, you can opt for this comprehensive practical guide teaching you TensorFlow and then all the essential algorithms of ML along with advanced ML.