Microsoft's AI chief talks about the speed up in AI performance being realized by its BrainWave platform.
Microsoft is using custom hardware to realize a 50-100x speed up in how quickly it can run AI algorithms that power its Bing search engine — and will make the tech available to all from next year.
The acceleration is being powered by the BrainWave platform, a network of customizable chips known as Field-Programmable Gate Arrays (FPGAs), tailored to efficiently handle deep neural networks.
"The power of the BrainWave platform are FPGAs, field-programmable gate arrays, which are really executing these AI algorithms in hardware," said Joseph Sirosh, corporate VP for artificial intelligence & research at Microsoft.
"What this means is approximately 50-100x the speed up. Also, there's significant cost savings, because you are executing these neural networks in the most efficient way possible in hardware."
A large amount of Bing search queries are served through BrainWave, with Sirosh saying the FPGAs are "especially good" for accelerating text-based applications.
Microsoft plans to roll out the AI-accelerating FPGAs found in BrainWave to its Azure cloud in 2018, from where they will be available to developers.
"It opens up a very large collection of new scenarios for AI applications because of the price-performance advantage that hardware-based execution gets," says Sirosh.
"The thing about FPGAs is that they are also very configurable, by their nature. So as AI evolves so rapidly, as new algorithms come in, it is possible for you to build the custom gate logic required to execute in hardware in that FPGA. So it can keep pace with AI development."
Microsoft has already deployed FPGAs across the vast majority of its new servers for applications like faster networking.
Read more on AI
- Google uses DeepMind AI to reduce energy use at data centers and save money
- Cray supercomputing comes to Microsoft Azure to boost AI workloads in the cloud
- Google launches open source system to make training deep learning models faster and easier
- How Google's DeepMind beat the game of Go, which is even more complex than chess
- Google weaves AI and machine learning into core products at I/O 2017
- Google AlphaGo AI clean sweeps European Go champion (ZDNet)
- AlphaGo defeats Go world champion in China (ZDNet)