Microsoft recently unveiled Brainwave, a field programmable gate array (FPGA) system, to provide record-breaking deep learning capabilities for cloud services.
The company yesterday just announced Brainwave for running super-fast, low-latency serving of AI. The new system Brainwave based on FPGA will enable developers to deploy the machine learning models onto the programmable silicon as well as get high performance over that developers could get from a GPU or CPU.
In July, the company ever claimed that it was integrating Bing and Azure with FPGA, and is also turning attention to the FPGA. Now, Brainwave is just the nest step on the evolution road.
According to Microsoft’s benchmarking, the company said that it displays that the platform is able to support for 39.5 Teraflops with no needs for batching on a huge gated recurrent. As for the early trsts, the company used the system with Intel Statix 10 FPGA.
Actually, the company has usually said the FPGA’s potential for AI, especially for marching learning as well as parallel computing. As a matter of fact, these chips are both easily programmable and efficient, which make them be good candidates.
To enhance its capability, Microsoft synthesized DNN or DPU processing units into the FPGA chips. In this means, Microsoft hopes to be able to promote its research as well as real-time processing capability.
Currently, Microsoft makes Brainwave system support the trained models which is created by using its CNTK framework and the TensorFlow framework from Google. According to Doug Burger, Microsoft Research’s distinguished engineer, the company is now utilizing Brainwave system across FPGAs installing in Microsoft’s data center. In the near future, Brainwave will make Microsoft’s services more rapidly support the AI features.
Although now the company has not offered a roadmap for the availability of Brainwave system for customers, it is working on making its third parties be able to bring Brainwave-based trained models through its Azure cloud.
The latest information Microsoft confirmed is that it is now developing Brainwave and making it be usable in Azure Cloud. Same to the FPGA integration make before, Microsoft claims that its customers can receive direct results from this new product, also including on Bing.
This new solution is highlighted in its super-fast performance speed, which allows near real-time AI speeds and deep learning.
Now, Brainwave is working with Microsoft CNTK and Google TensorFlow, but the company did not discover any specific data for the availability of Brainwave yet.