Deep learning acceleration platform from Microsoft targets real time artificial intelligence
Microsoft has announced the release of its Project Brainwave deep learning acceleration platform, which is a hardware platform build with three layers that is designed for real-time artificial intelligence processing.
Microsofthas announced the release of its Project Brainwave deep learningacceleration platform, which is a hardware platform build with three layers that is designed for real-time artificial intelligenceprocessing.
Introduced at Hot Chips 2017—a leading conference on microprocessors and microcomputers—Project Brainwave is built with three main layers:
- A distributed system architecture
- A hardware deep neural network (DNN) engine synthesized onto FPGAs
- A compiler and runtime for low-friction deployment of trained models
"We designed the system for real-time AI, which means the system processes requests as fast as it receives them, with ultra-low latency," wrote Doug Burger, Distinguished Engineer, Microsoft, in a blog post. "Real-time AI is becoming increasingly important as cloud infrastructures process live data streams, whether they be search queries, videos, sensor streams, or interactions with users."
Project Brainwave leverages the FPGA infrastructure that Microsoft has been deploying over the past few years. By attaching FPGAs directly to Microsoft’s datacenter network, the company can serve DNNs as hardware microservices, where a DNN can be mapped to a pool of remote FPGAs and called by a server with no software in the loop. This system architecture, according to the company, both reduces latency—, since the CPU does not need to process incoming requests—and allows very high throughput, with the FPGA processing requests as fast as the network can stream them.
Secondly, the hardware platform uses a "soft" DNN processing unit (or DPU), synthesized onto available FPGAs. Microsoft’s design combines both the ASIC digital signal processing blocks on the FPGAs and the synthesizable logic to provide a greater and more optimized number of functional units, according to the company. This approach exploits the FPGAs flexibility in two ways, the first of which is that highly customized, narrow-precision data types that increase performance without real losses in model accuracy have been defined. Second, the platform can incorporate research innovations into the hardware platform quickly (typically a few weeks), which is essential in this fast-moving space, according to Microsoft.
Lastly, Project Brainwave incorporates a software stack designed to support popular deep learning frameworks such as Microsoft Cognitive Toolkit and Google’s TensorFlow, with the plan to support more underway.
This is just the latest headline involving Microsoft’s focus on artificial intelligence technology. Earlier this year, Microsoft Ventures, Microsoft’s corporate venture fund, announced investments in two new artificial intelligence companies: Agolo, a company that develops AI-based summarization software, and Bonsai, a company that specializes in automating the management of complex machine learning algorithms. Additionally, the company recently announced that its next version of its wearable self-contained, holographic computer known as the HoloLens will incorporate an artificial intelligence coprocessor to enable deep learning capabilities on the device.
At Hot Chips, Microsoft employees demonstrated the device ported to Intel’s new 14 nm Stratix 10 FPGA. Read about this, as well as additional details about the device, on the Microsoft blog.