✳flâneur — a map of the web's best reading

Memory-Limited Layers User's Guide - NVIDIA Docs

docs.nvidia.com · 2,732 words · saved by 1 readers

This guide describes the performance of memory-limited layers including batch normalization, activations, and pooling. It also provides tips for understanding and reducing the time spent on these layers within a network. The following quick start checklist provides specific tips for layers whose performance is limited by memory accesses. Explore the available implementations of each layer in the NVIDIA cuDNN API Reference or your framework. Often the best way to improve performance is to choose a more efficient implementation. For example, persistent implementations of batch normalization require fewer loads from memory. Be aware of the number of memory accesses required for each layer. Performance of a memory-bound calculation is simply based on the number of inputs, outputs, and weights that need to be loaded and/or stored per pass. We don’t have recommended parameter tweaks for these layers. Be aware of the impact of each layer on the overall training step performance. Memory-bound

Memory-Limited Layers User's Guide - NVIDIA Docs Menu Docs Hub NVIDIA Deep Learning Performance Submit Search Submit Search NVIDIA Docs Hub Homepage Deep Learning Performance NVIDIA Deep Learning Performance Memory-Limited Layers User's Guide Download PDF Memory-Limited Layers User's Guide Abstract This guide describes the performance of memory-limited layers including batch normalization, activations, and pooling. It also provides tips for understanding and reducing the time spent on these layers within a network. 1. Quick Start Checklist The following quick start checklist provides specific

Explore this link on the map →

saved by

related reading