With the introduction of its Cloud TPU processing chip, Google has become a major player in an AI industry set to drive innovation for the foreseeable future.
Google has recently announced a potentially game-changing development in AI technology. In his keynote at the 2017 Google I/O developer conference, CEO Sundar Pichai introduced the Cloud Tensor Processing Unit (Cloud TPU), a processing chip that draws its name from Google’s open-source TensorFlow machine learning software platform. According to Pichai, Cloud TPUs are the next step in his company’s pivot “from a mobile-first to an AI-first approach.”
AI has already shown tremendous promise in industries ranging from education to software development to marketing. Google is not only “rethinking all [its] products and applying machine learning and AI to solve user problems,” but also developing AI solutions for medical imaging, genomic analysis, and molecule discovery. In fact, Google believes that its new chip has such remarkable potential for good that it is making 1,000 Cloud TPU systems available to any researchers willing to share the details and outcomes of their work.
Redefining Processing Power with TPUs
Even a cursory examination of the technical specifications of the Cloud TPUs suggests Google’s enthusiasm is more than merited. Roughly 15 to 30 times faster and 30 to 80 times more power-efficient than standard CPUs and GPUs, Cloud TPUs provide developers with the kind of massive computing power that highly-intensive AI processes like machine learning need.
A single TPU offers 128 teraflops — each teraflop consisting of one trillion floating point computational operations per second — an astounding capacity, considering that a machine like the iPhone 6 offers around 100 gigaflops. In other words, one TPU delivers the computing power of 1,280 iPhones, and Pichai reports that Google can fit 256 TPUs into a single supercomputer, or “TPU pod.” The capacity of a single TPU pod? 11.5 petaflops. It’s no wonder Pichai touted this latest development as “an important advance in technical infrastructure for the AI era.”
Putting TPUs into Action
Impressive tech specs aside, Google’s “AI-first data centers” have already begun to produce promising results for many of the company’s signature products. For instance, Google’s translation programs rely upon machine learning and natural language processing algorithms that demand constant training. “To give you a sense,” Pichai explains, “each one of our machine translation models takes a training of over three billion words for a week on about 100 GPUs.”
Cloud TPUs have utterly redefined these operations. As Pichai marvels, when it comes to AI training, what would have previously taken a full day using 32 top-of-the-line GPUs now takes a single afternoon using one-eighth of a TPU pod. Developers have already been doing incredible work on Google’s TensorFlow platform, but the newfound power and accessibility of Cloud TPUs promise to initiate a bonafide paradigm shift in AI-based innovation.
Pichai’s closing remarks indicate the real gravity of this shift: “We are training neural nets to improve the accuracy of DNA sequencing. Deep Variant is a new tool from google.ai that identifies genetic variants more accurately than state-of-the-art methods, reducing errors in important applications; we can more accurately identify whether or not a patient has a genetic disease and can help with better diagnosis and treatment.”
For all the admittedly enjoyable and/or helpful voice and picture recognition apps, email completion assistants, and the like, the fact of the matter is that AI is driving the kind of innovation that changes and even saves lives. As such, the more efficient and effective we can make our AI, the better, and Google’s Cloud TPUs represent a confident step in the right direction.
And as machine learning technology prepares to completely transform other industries, it’s already having a tremendous impact on the world of marketing. Albert™, the world’s first completely AI-driven digital marketing platform, is automating the difficult, tedious work involved in cross-channel marketing campaigns. By completing the tasks involved in the planning, execution, and optimization of campaigns that effectively target individual consumers, Albert frees marketers to focus on what they do best: high-value creative and strategic work.