NVIDIA Deep Learning SDK（NVIDIA深度学习相关的各种库）

原文链接

• Mixed precision in AI frameworks (Automatic Mixed Precision): 混合精度计算，最高3倍加速比，利用Tensor Cores；（Get upto 3X speedup running on Tensor Cores With just a few lines of code added to your existing training script）
• Deep Learning Primitives (cuDNN): 深度学习GPU加速的标配；（High-performance building blocks for deep neural network applications including convolutions, activation functions, and tensor transformations）
• Input Data Processing (DALI): 并行度高的数据加载和数据增强库（主要针对图像、视频）；（An open source data loading and augmentation library that is fast, portable and flexible）
• Multi-GPU Communication (NCCL): 组播通信神器，double-tree实现；（Collective communication routines, such as all-gather, reduce, and broadcast that accelerate multi-GPU deep learning training）
• Deep Learning Inference Engine (TensorRT): 推理神器；（High-performance deep learning inference runtime for production deployment）（TensorFlow-to-ONNX-to-TensorRT例子）
• Deep Learning for Video Analytics (DeepStream SDK): High-level C++ API and runtime for GPU-accelerated transcoding and deep learning inference
• Optical Flow for Video Inference (Optical Flow SDK): Set of high-level APIs that expose the latest hardware capability of Turing GPUs dedicated for computing the optical flow of pixels between images. Also useful for calculating stereo disparity and depth estimation.
• High level SDK for tuning domain specific DNNs (Transfer Learning Toolkit): 迁移学习；（Enabling end to end Deep Learning workflows for industries）
• AI enabled Annotation for Medical Imaging (AI-Assisted Annotation SDK): 没权限打开？？；（AI-assisted annotation for medical imaging related data labeling）
• Deep Learning GPU Training System (DIGITS): 网页版的数据集、模型、训练可视化工具（和TensorGou很像），在计算框架等核心组件外围包的一层可视化而已；（Rapidly train highly accurate deep neural network (DNNs) for image classification, segmentation and object detection tasks）
• Linear Algebra (cuBLAS): GPU矩阵计算标配；（GPU-accelerated BLAS functionality that delivers 6x to 17x faster performance than CPU-only BLAS libraries）
• Sparse Matrix Operations (cuSPARSE): 稀疏矩阵计算标配（模型权重剪枝那里真用到了）；（GPU-accelerated linear algebra subroutines for sparse matrices that deliver up to 8x faster performance than CPU BLAS (MKL), ideal for applications such as natural language processing）