Workspace limit size (in MB) for selecting cuDNN convolution algorithms. cuDNN’s internal functions achieve the fastest matching algorithm within this memory limit. Typically, faster algorithms can be selected within larger workspaces, but at the expense of significantly increasing memory space. Users need to make a trade-off between memory and speed.

#PaddlePaddle #User #Guide #cudnn

Leave a Comment

Your email address will not be published. Required fields are marked *