Currently, cudnn tensor descriptor has two interface: cudnnSetTensor4dDescriptor and cudnnSetTensorNdDescriptor.
I’m a little confused about the performance loss if I use second interface for 4D data.
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| Some question abort cudnnSetTensorNdDescriptor and cudnnSetTensor4dDescriptor. | 0 | 594 | November 21, 2017 | |
| cudnnSetTensor4dDescriptor & cudnnSetTensorNdDescriptor parameters opposite params | 0 | 1575 | May 31, 2018 | |
| 4d descriptors vs Nd descriptors | 0 | 486 | June 24, 2019 | |
| Stride parameters of cudnnSetTensor4dDescriptorEx and cudnnGetTensor4dDescriptor | 0 | 891 | July 4, 2016 | |
| cuDNN: "cudnnTensor4dDescriptor_t" for sigmoid component in fully connected layer | 1 | 1544 | November 12, 2014 | |
| is it normal for creating cudnn handle that takes about 1.5 second? | 3 | 480 | October 31, 2019 | |
| invalid resource handle for cudnnSetTensor4dDescriptorEx | 1 | 753 | July 2, 2016 | |
| Data organization of Tensor in cudnn | 0 | 611 | December 7, 2016 | |
| Tensor packing and cryptic errors | 6 | 1086 | June 8, 2022 | |
| CUDNN_STATUS_NOT_SUPPORTED when call cudnnOpTensor() | 2 | 786 | December 31, 2021 |