Deep learning has become the foundation of developing Artificial Intelligence systems and Microsoft wants to speed things up on that regard. At the Hot chips 2017, the tech giant announced Brainwave; an FPGA based neural system designed for ultra-low latency deep learning in the cloud. The early benchmarking for Brainwawe indicated that when using Intel Stratix 10 FPGAs, it can sustain 39.5 Teraflops on a large gated recurrent unit without any batching.
“I’m delighted to share more details in this post, since Project Brainwave achieves a major leap forward in both performance and flexibility for cloud-based serving of deep learning 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,” said Microsoft in its blog post.
Microsoft has been heavily investing both time and money into FPGA’s for quite some time now and the results are there to see. It has been deploying large clusters of the field-programmable gate arrays into its data centers with Algorithms written into FPGAs, making them both efficient and easily reprogrammable. This makes it the perfect option for machine learning, specifically parallel computing.
While FPGA’s are somewhat of a retro technology, there renaissance has been a bit of a surprise with new use cases popping out for them in the age of AI. An example for this is Mipsology, a FPGA obsessed startup that is working closely with Amazon to make the technology accessible over Amazon Web Services and other platforms.
The new system already supports Google’ TensorFlow and Microsoft’s very on Cognitive Toolkit. Microsoft is yet to announce when Brainwave will be available for Microsoft Azure customers. It seems to be setting very high and ambitious expectations for deep learning performance using the technology, so there is undoubtedly a lot more to come right around the corner.