Hello! When (if it’s planned at all) can we expect high performance NHWC kernels in cuDNN?
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| cuDNN6.0: NCHW vs. NHWC | 0 | 2050 | May 22, 2017 | |
| foldedNhwcToNhwcKernel<__half, __half, float, bool=1> shows up on profiler timeline | 1 | 651 | June 5, 2020 | |
| Convolution incompatible with NHWC | 3 | 139 | August 30, 2024 | |
| cudnnPoolingForward() tensor format support | 1 | 596 | January 30, 2018 | |
| cudnnSetTensor and cudnnAddTensor are extremely slow for NHWC Tensors | 0 | 577 | March 18, 2018 | |
| cuDNN: NHWC format not supported unless manually specified? | 4 | 2878 | September 8, 2015 | |
| Unsolvable, see posted link below why: CUDNN_TENSOR_NCHW vs. CUDNN_TENSOR_NHWC | 3 | 1957 | March 9, 2017 | |
| NHWC vs NCHW convolution | 4 | 5067 | January 29, 2020 | |
| Many problems with the new batch normalization API | 1 | 835 | February 8, 2019 | |
| Which algo should be passed for cudnnConvolutionForward() when TensorCore and NHWC ? | 1 | 1382 | October 25, 2018 |