Data Science

Upgrade to the newest versions of NVIDIA CUDA-X libraries

AI-Generated Summary

  • cuDNN 7.5 introduces multi-head attention for accelerating Transformer models and improved depth-wise separable convolution for training Xception and MobileNet architectures.
  • CUDA 10.1 Update 1 delivers improved SpMM and SpMV kernel performance in cuSPARSE, Graph API support in cuFFT for CUDA Graphs, extended data types and heuristics in cuBLAS, and new nvJPEG APIs for stream parsing and multi-channel bitstreams.
  • TensorRT 5.1 GA adds optimization for DenseNet and TinyYOLO with over 20 new layers, the ability to update model weights without rebuilding engines, and INT8 deployment on NVIDIA Xavier-based AGX platforms using the DLA accelerator.
  • DALI 0.9 adds video pre-processing workflow support with a new optical flow operator and ROI-based JPEG decoding for high-resolution images.
  • Nsight Compute 2019.3 supports CUDA 10.1 Update 1 and the latest Turing GPUs, expands PC sampling and NVTX control, and improves workflows, UI, and command-line switches.

Next Steps

  • Download cuDNN 7.5 to access the latest deep neural network primitives.
  • Download CUDA 10.1 Update 1 for updated parallel computing tools and libraries.
  • Download TensorRT 5.1 GA to optimize deep learning inference workloads.
Powered by NVIDIA Nemotron. AI-generated content may summarize information incompletely. Verify important information. Learn more

Learn what’s new in the latest releases of cuDNN, CUDA, TensorRT, DALI, and Nsight Compute.

cuDNN 7.5

The NVIDIA CUDA Deep Neural Network library (cuDNN) is a GPU-accelerated library of primitives for deep neural networks. This version of cuDNN includes:

  • Multi-head attention for accelerating popular models such as Transformer
  • Improved depth-wise separable convolution for training models such as Xception and Mobilenet

Download

CUDA 10.1 Update 1

CUDA is the parallel computing platform and programming model for general-purpose computing on NVIDIA GPUs. This version of CUDA includes:

  • Improved SpMM/SpMV kernel performance in cuSPARSE for sparse applications in high-performance computing (HPC) and machine learning
  • Graph API support in cuFFT to allow use of FFT kernels in CUDA Graphs
  • Extended data type support and improved heuristics in cuBLAS for all HPC and machine learning applications
  • APIs for stream parsing, memory control, decoding, and multi-channel bitstreams in nvJPEG

Download

TensorRT 5.1 GA

NVIDIA TensorRT is a platform for high-performance deep learning inference. This version of TensorRT includes:

  • Optimization of new models such as DenseNet and TinyYOLO with support for over 20 new layers, activations, and operations in TensorFlow and ONNX
  • The ability to speed up deployment by updating model weights in an existing engine without rebuilding them
  • Application deployment in INT8 precision on NVIDIA Xavier™-based NVIDIA AGX™ platforms using the NVIDIA DLA accelerator

Download

DALI 0.9

The NVIDIA Data Loading Library (DALI) is a portable, open-source GPU-accelerated library for decoding and augmenting images and videos to accelerate deep learning applications. This version of DALI includes:

  • Video pre-processing workflow support with new optical flow operator
  • ROI-based JPEG decoding for high-resolution images

Download

Nsight Compute 2019.3

NVIDIA Nsight Compute is an interactive kernel profiler for CUDA applications. This version of Nsight compute includes:

  • Support for CUDA 10.1 Update 1 and the latest NVIDIA Turing GPUs
  • Expanded PC sampling and NVTX control
  • Workflow and UI improvements along with extended command line switches

Download

Discuss (0)

Tags