NVIDIA and ARM have formed a partnership that will see a huge number of mobile and Internet of Things (IoT) devices integrated with deep learning inferencing. The tie-up will allow NVIDIA to incorporate its open-source NVIDIA Deep Learning Accelerator architecture into ARM's recently announced machine learning platform called Project Trillium. The goal of the partnership is to help companies that manufacture IoT chips introduce artificial intelligence (AI) to their product more seamlessly.
Project Trillium, which was announced last month, is a suite of technologies consisting of scalable processors, software, and architecture centered around machine learning, object detection, and neural networking. Project Trillium is meant to be initially combined into the ARM ML processors scaled to provide a range of benefits to mobile devices coupled with ARM’s more traditional chipset to provide a significant boost in performance. According to ARM, Project Trillium is capable of providing up to 4.6 trillion operations per second. Combined with NVIDIA's deep learning technology, the integrated platform is expected to offer a solution that could provide access to deep learning technologies to billions of IoT devices and other consumer electronics products.
The "NVDLA" architecture is based on the NVIDIA Xavier, the company's autonomous machine processor, with the goal of helping developers build deep learning inference accelerators. The system-on-a-chip for self-driving vehicles was unveiled in 2016 and built to achieve 20 Deep Learning Tera-Ops worth of 8-bit Integer operations while keeping its power requirements as low as 20 watts. NVIDIA's range of developer tools such as the upcoming versions of TensorRT also provides support for NVDLA. NVIDIA's deep learning push saw a significant boost last year when the company teamed up with Qwiklabs to train 100,000 developers in machine learning techniques through its Deep Learning Institute. As part of that effort, NVIDIA expanded all components of the program with assistance from other companies, with Qwiklabs providing aid in the process of growing the online version of the training platform.