The NVIDIA CUDA Deep Neural Network library (cuDNN) is a GPU-accelerated library of primitives for deep neural networks.
Today, the NVIDIA team released the latest version of NVIDIA cuDNN – version 7.5.
What’s New in cuDNN 7.5
Deep learning frameworks using cuDNN 7.5 and later, can leverage new features and performance of the Volta and Turing architectures to deliver faster training performance. cuDNN 7.5 highlights include:
- Up to 3x faster training of ResNet-50 and GNMT on Tesla V100 vs. Tesla P100
- Improved depth-wise separable convolution for training models such as Xception and Mobilenet
- Multi-Head Attention for accelerating popular models such as Transformer
- New tensor folding APIs for accelerated performance on models such as Mask R-CNN, GANs and DeepSpeech2
Read the latest cuDNN release notes for a detailed list of new features and enhancements.