cuDNN Release Notes v7.0.3
cuDNN Release Notes v7.0.3 (PDF)
Key Features and Enhancements
- Forward Grouped Convolutions where input channel per groups is 1, 2 or 4 and hardware is Volta or Pascal.
cudnnTransformTensor()where input and output tensor is packed.Note:
This is an improved fallback, improvements will not be seen in all cases.
The following are known issues in this release:
CUDA_ERROR_ILLEGAL_ADDRESS. This issue affects input images of just one 1 pixel in width and certain
The following issues have been fixed in this release:
TensorOpproduce incorrect results for half and INT8 inputs for various use cases.
cudnnPoolingBackward()can produce incorrect values for rare cases of non-deterministic MAX pooling with
window_width > 256. These rare cases are when the maximum element in a window is duplicated horizontally (along width) by a stride of
k. The behavior is now fixed to accumulate derivatives for the duplicate that is left-most.
cudnnGetConvolutionForwardWorkspaceSize()produces incorrect workspace size for algorithm
FFT_TILINGfor 1d convolutions. This only occurs for large sized convolutions where intermediate calculations produce values greater than 2^31 (2 to the power of 31).
cudnnPooling*()functions for small
channels * height * width < 4).