Nvidia, whose graphical processing units (GPUs) are being applied to machine learning, released its JetPack 2.3 software kit for download today. The company claims its software will double users’ deep learning performance on embedded systems.
JetPack is the software kit for Jetson TX1, Nvidia’s developer platform for deep learning on embedded systems. JetPack bundles and installs all system software, tools, libraries, and APIs.
Deep learning is fueling startups such as Deep Instinct, which claims to be the first company applying deep learning to cybersecurity, and Nervana Systems, which is being acquired by Intel. Deep learning is also at the heart of AlphaGo, the Google system that can beat humans at the game of Go.
Nvidia’s JetPack 2.3 includes a GPU engine that maximizes runtime performance for applications including image classification, segmentation, and object detection. When comparing JetPack 2.3 to other versions, Nvidia claims this engine doubles the performance of deep learning, and enables developers to deploy neural networks powered by the Jetson platform.
JetPack 2.3 also includes a deep-learning library, cuDNN 5.1, that provides implementations for convolutions, activation functions, and tensor transformations. Support for advanced network models such as long short-term memory (LSTM) and recurrent neural network (RNN) have also been included in this release.
JetPack also features a multimedia API, which provides a package of low-level APIs that are ideal for flexible development, Nvidia claims. The multimedia API also allows for video decode, encode, format conversion, and scaling functionality.
Nvidia said that the ability to run deep neural networks is key for intelligent machines to solve real world problems in areas like public safety, smart cities, manufacturing, disaster relief, agriculture transportation, and infrastructure inspection.