All Jetson Products

• Support for installation of JetPack components via Debian package archives.

  JetPack components are provided as Debian packages via a public APT server hosted by NVIDIA. Enables easier installation of JetPack components and upgrading to future versions.

• Includes TensorRT 6.0.1

• Includes cuDNN 7.6.3

• Support for CSI and Encode from within containers

• Includes OpenCV 4.1.1

• Developer preview of VPI (Vision Programming Interface)

  VPI is a software library that provides Computer Vision / Image Processing algorithms implemented on PVA¹ (Programmable Vision Accelerator), GPU² and CPU².

• Supports DeepStream 4.0.2

• Supports ISAAC SDK version 2019.3

Jetson AGX Xavier

• Support for DLA from within containers

• Performance optimized PVA support for developer preview of VPI

¹ PVA is available only on Jetson AGX Xavier series. ² GPU and CPU implementation is not performance optimized in this release. A future release will bring performance optimized GPU and CPU implementation.

L4T 32.3.1

• Support for simultaneously flashing multiple Jetson devices

• Support for configuring Wi-Fi during headless initial configuration

• Option to select the desired size of SD card APP partition during first boot on Jetson Nano

• Introducing Jetson-IO tool to configure 40-pin header

  New easy-to-use tool for configuring 40-pin header on Jetson developer kits

• Automatic addition of every new user to I2C group

• Increase USB 3.0 performance for Jetson TX1/Nano and TX2 family

TensorRT 6.0.1.10

• New Layers

  1. IResizeLayer – The IResizeLayer implements the resize operation on an input tensor.

  2. IShapeLayer – The IShapeLayer gets the shape of a tensor.

  3. PointWise fusion – Multiple adjacent pointwise layers can be fused into a single pointwise layer, to improve performance.

• New Operators

  1. 3-dimensional convolution – Performs a convolution operation with 3D filters on a 5D tensor.

  2. 3-dimensional deconvolution – Performs a deconvolution operation with 3D filters on a 5D tensor.

  3. 3-dimensional pooling – Performs a pooling operation with a 3D sliding window on a 5D tensor.

• New Operations

  1. TensorFlow - Added ResizeBilinear and ResizeNearest operations.

  2. ONNX - Added Resize operation.

• New Samples

  1. sampleDynamicReshape - Demonstrates how to use dynamic input dimensions in TensorRT by creating an engine for resizing dynamically shaped inputs to the correct size for an ONNX MNIST.

  2. sampleReformatFreeIO - Uses a Caffe model that was trained on the MNIST dataset and performs engine building and inference using TensorRT. Specifically, it shows how to use reformat free I/O tensors APIs to explicitly specify I/O formats to TensorFormat::kLINEAR, TensorFormat::kCHW2 and TensorFormat::kHWC8 for Float16 and INT8 precision.

  3. sampleUffPluginV2Ext - Implements the custom pooling layer for the MNIST model and demonstrates how to extend INT8 I/O for a plugin.

  4. sampleNMT - Demonstrates the implementation of Neural Machine Translation (NMT) based on a TensorFlow seq2seq model using the TensorRT API. The TensorFlow seq2seq model is an open sourced NMT project that uses deep neural networks to translate text from one language to another language.

• New Optimizations

  1. Dynamic Shapes - The size of a tensor can vary at runtime. IShuffleLayer, ISliceLayer, and the new IResizeLayer now have optional inputs that can specify runtime dimensions. IShapeLayer can get the dimensions of tensors at runtime, and some layers can compute new dimensions.

  2. Reformat free I/O - Network I/O tensors can be different to linear FP32. Formats of network I/O tensors now have APIs to be specified explicitly. The removal of reformatting is beneficial to many applications and specifically saves considerable memory traffic time.

  3. Layer Optimizations - Shuffle operations that are equivalent to identify operations on the underlying data will be omitted, if the input tensor is only used in the shuffle layer and the input and output tensors of this layer are not input and output tensors of the network. TensorRT no longer executes additional kernels or memory copies for such operations.

  4. New INT8 calibrator - MinMaxCalibrator - Preferred calibrator for NLP tasks. Supports per activation tensor scaling. Computes scales using per tensor absolute maximum value.

  5. Explicit Precision - One can manually configure a network to be an explicit precision network in TensorRT. This feature enables users to import pre-quantized models with explicit quantizing and dequantizing scale layers into TensorRT. Setting the network to be an explicit precision network implies that one will set the precision of all the network input tensors and layer output tensors in the network. TensorRT will not quantize the weights of any layer (including those running in lower precision). Instead, weights will simply be cast into the required precision.

cuDNN 7.6.3

• Improved depth-wise convolution

• Improved grouped convolution for cudnnConvolutionBackwardFilter()

• Performance of grouped convolution is enhanced for cudnnConvolutionForward and cudnnConvolutionBackwardFilter

CUDA 10.0.326

• CUDA version remains the same from JetPack 4.2.x

• NVIDIA Nsight Systems 2019.6 for application profiling across GPU and CPU.

  • Timeline enhancements to improve identification of tiny events

  • Option to disable CPU periodic sampling and thread state backtrace

  • Improved error reporting

• NVIDIA Nsight Graphics 2019.5 for graphics application debugging and profiling.

  • Adds support for new Vulkan extension

    VK_EXT_conditional_rendering

    • Ycbcr format extensions

    • Support for OpenGL 4.6 in frame debugging, C++ capture and frame profiling activities

  • Support for immediate mode

  • Improved visibility and detail in the Events View

• NVIDIA Nsight Compute 2019.3 for CUDA kernel profiling.

  • NVIDIA Nsight Compute version remains the same from JetPack 4.2.x