JetPack for L4T (Jetson Development Pack) is an on-demand all-in-one package that bundles and installs all software tools required to develop for the NVIDIA® Jetson Embedded Platform (including flashing your Jetson Development Kit with the latest OS images). JetPack includes host and target development tools, APIs and packages (OS images, tools, APIs, middleware, samples, documentation including compiling samples) to enable developers to jump start their development environment for developing with the Jetson Embedded Platform.

The JetPack 2.3.1 release supports both the latest Jetson TX1 Developer Kit, running 64-bit Ubuntu 16.04, and Jetson TK1 Developer Kit, running 32-bit Ubuntu 14.04. Note that the host PC needs to be running Ubuntu 14.04.

For a technical overview of JetPack 2.3, please refer to the Parallel Forall Blog.

View the full 2.3.1 Release Notes here.

Download Jetpack

Key Features in Jetpack 2.3.1

TensorRT TensorRT is a new tool for optimizing Deep Learning inference for production deployment. It reduces memory requirements and improves runtime performance by up to 2x, which translates to faster and more accurate Deep Learning applications. This version supports the prototxt model format with CNNs and Deconv networks (other model formats and network types are supported by cuDNN).
cuDNN 5.1 cuDNN provides high-performance building blocks for deep learning used by all deep learning frameworks, allowing for 2.7x faster training of networks with 3x3 convolutions such as VGG. Includes support for convolutions, activation functions and tensor transformations.

Improvements to developer tools and libraries:

  • System level bottleneck identification
  • Up to 2x faster NVCC compilation
  • Support for gcc 5.x host compiler
  • nvGRAPH for accelerated graph analytics
  • cuBLAS optimizations for matrices smaller than 512 and for batched operations
  • Optimized dense matrix x sparse vector routines
Multimedia API

The Jetson Multimedia API package provides access to low level APIs for flexible application development, including:

  • Camera API: enables per frame control over camera parameters and EGL stream outputs that allow efficient interoperation with GStreamer* and V4L2 pipelines.
  • V4L2 API: enables video decode, encode, format conversion and scaling functionality. V4L2 for encode opens up many features like bit rate control, quality presets, low latency encode, temporal tradeoff, motion vector maps, and more.

*GStreamer implementation from previous releases is still supported.

To access older versions of JetPack, please visit the JetPack Archive.