surroundings:
Anaconda3-5.2.0
python3.6.5
The initial tensorflow version is 1.8.0, cuda9.0, when upgrading tensorflow to 1.10.0, the torch1.1.0 running error occurs: RuntimeError: cuDNN error: CUDNN_STATUS_EXECUTION_FAILED. At first I thought it was because some content of cuda was updated when upgrading tensorflow. So I tried several combinations and failed to solve the problem. Instead, CUDA reported an error. The relevant version of the library (**.dll) could not be found. Later, the reason was found to be caused by the version mismatch. Finally, the effective combination was proved as follows:
tensorflow1.14.0+torch1.2.0+CUDA10.0
My personal test is valid, here is a link to Baidu Cloud, and friends who need it can download it directly:
- Anaconda3-5.2.0
Link: https://pan.baidu.com/s/15zsZQZ_-4Ein4lvNAOFCkg Extraction code: vgk1
- tensorflow-gpu 1.14.0
Link: https://pan.baidu.com/s/1HsRu3Z4QsUEqg6AEkBebmQ Extraction code: bmcb
- pytorch-gpu 1.2.0
Link: https://pan.baidu.com/s/1i_MZoP-aPPsgNQMhlF_5Jg Extraction code: gjs2
- CUDA10.0 + CUDnn
Link: https://pan.baidu.com/s/1kG_5i7LgToxVeL0G4_LujA Extraction code: jwfw
- CUDA10.2 + CUDnn
Link: https://pan.baidu.com/s/1U-R1EmJlEfICKJL6ASlejw Extraction code: utd7
Please indicate the source for reprinting, hoping to help those in need