Brainwave Project: Real-time AI from the Microsoft Cloud?
To differentiate itself from its competitors in AI and the Cloud, Microsoft presented the Brainwave project on August 22nd.
The Redmond firm is heading headlong into the AI sector. After having developed Cortana, a voice assistant based on machine learning, Microsoft wants to facilitate the processing of information by AI in real time.
Presented at the Hot Chips 2017 show, the Brainwave project is based on a hardware solution communicating with Microsoft Azure. Concretely, the IT giant works with FPGA (Field Programmable Gate Array) chips . The latter, manufactured by Intel, are reprogrammable. The advantages of these chips lie in their on-the-fly computing capacities, lower latency times than a solution based on graphics cards and generally reduced power consumption.
The Brainwave project shows its muscles
They are placed in data centers to ensure the functioning of this distributed architecture, designed to speed up the processing of operations necessary in the context of computer vision or text to speech / speech to text, the main functionality of voice assistants.
To prove the effectiveness of its technology, Microsoft relies on benchmarks carried out within its own data centers. This technology reaches a power of 39.5 Tflops, or more than 130,000 operations per second.
Align with the competition
Based on deep learning, the Brainwave project aims to integrate the processing power of these chips dedicated to AI within the Microsoft Azure cloud.
Calculating artificial intelligence in real time is the challenge facing many industries to successfully transform themselves digitally. Auto factories and manufacturers are the first to need this computing power. The emergence of voice assistants is also one of the reasons for the launch of this Brainwave project. Opposite, Amazon also offers an FPGA solution, while Google relies on its in-house Tensor Processor Unit chips to upgrade its AI Alpha . The Redmond company therefore has every interest in quickly making this solution combining hardware and cloud computing functional.