As a leading developer of GPUs, NVIDIA has taken on the role of producing some of the most advanced processors for deep learning and AI applications currently available on the market.
As a leading developer of GPUs,
NVIDIA has taken on the role of producing some of the most advanced processors for deep learning and AI applications currently available on the market.
NVIDIA has long been a pioneer of GPUs, investing billions of dollars in R&D for each new processor generation. As a result, the company has enabled the industry to push the envelope of what’s possible with deep learning and other parallel computing applications. Today, NVIDIA’s GPUs have become a de-facto standard for modern AI and deep learning, currently used by all the major cloud service providers and system builders.
Deepu Talla, VP and GM of Autonomous Machines at NVIDIA explained that there are two key characteristics of the NVIDIA architecture that are critical for deep learning. “First, it must be high performing. The algorithms that do training and inference with neural networks are parallel in nature and they require a parallel architecture to deliver high performance efficiently. Second, it must be general-purpose programmable. Deep learning research continues to move forward, and any architecture that’s not generally programmable will not be able to adapt to the latest innovations,” he said.
In terms of the challenges faced when running deep-learning-based solutions, Talla pointed to workflow difficulties. “Deep learning is a new model for computing, in that the computer effectively writes its own software through the process of training a neural network, and if you don’t use the right tools you get suboptimal results,” he explained. “NVIDIA’s GPUs are designed specifically to solve parallel computing problems, like deep learning, which represent some of the toughest challenges in computing today. Additionally, NVIDIA builds an end-to-end solution for deep learning, which includes training and inferencing, that make it easy to develop and deploy cutting-edge solutions.”
The company introduced NVIDIA Metropolis, a comprehensive platform that includes hardware and software for the training and deployment of AI capabilities for smart and safe city applications.
“NVIDIA Metropolis is an end-to-end platform for intelligent video analytics that includes hardware and software for the training and deployment of deep neural networks. From NVIDIA GPU Cloud and DIGITS software running on an NVIDIA DGX supercomputer, to TensorRT and DeepStream running on NVIDIA Tesla and NVIDIA Jetson, NVIDIA Metropolis supports all of the latest frameworks and includes partner solutions for intelligent video analytics applications,” Talla said.
To address video surveillance needs, Talla offered NVIDIA Jetson, the company’s solution for AI at the edge. “It’s a credit-card sized computing module that operates under 10W, making it suitable for integration inside an NVR or near a camera,” he explained. “Jetson shares the same GPU architecture as the rest of NVIDIA’s products, and it’s part of the Metropolis platform. NVIDIA also provides NVDLA, a deep learning accelerator for interference that is free for all chip companies to integrate deep learning into their IoT and camera chips. It’s never been easier to develop and deploy solutions for smart and safe cities all the way to the camera.”