在尝试通过Tensorflow/Anaconda使用NVIDIA GPU计算工具包时,我遇到了CUDA安装后各种dll文件丢失的问题。这就是TF所抱怨的:
2021-06-13 15:57:48.349493: W tensorflow/stream_executor/platform/default/dso_loader.cc:60] Could not load dynamic library 'cublas64_11.dll'; dlerror: cublas64_11.dll not found
2021-06-13 15:57:48.349942: W tensorflow/stream_executor/platform/default/dso_loader.cc:60] Could not load dynamic library 'cublasLt64_11.dll'; dlerror: cublasLt64_11.dll not found
2021-06-13 15:57:48.350349: W tensorflow/stream_executor/platform/default/dso_loader.cc:60] Could not load dynamic library 'cufft64_10.dll'; dlerror: cufft64_10.dll not found
2021-06-13 15:57:48.350752: W tensorflow/stream_executor/platform/default/dso_loader.cc:60] Could not load dynamic library 'curand64_10.dll'; dlerror: curand64_10.dll not found
2021-06-13 15:57:48.351155: W tensorflow/stream_executor/platform/default/dso_loader.cc:60] Could not load dynamic library 'cusolver64_10.dll'; dlerror: cusolver64_10.dll not found
2021-06-13 15:57:48.351555: W tensorflow/stream_executor/platform/default/dso_loader.cc:60] Could not load dynamic library 'cusparse64_11.dll'; dlerror: cusparse64_11.dll not found最初,cudart64_110.dll也丢失了,但我手动下载了它并将其放入C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.3\bin中。这不是环境变量(即path)的问题-在安装CUDA之后,上述DLL将不可用,无论是在上述工具包文件夹中,还是在其他任何地方。我也不能简单地安装以前版本的CUDA,因为安装工具提到我已经安装了较新的版本。
在这些DLL不可用方面,我遗漏了什么?
发布于 2021-06-14 03:12:57
TF需要CUDA 11.0,正如评论中友好地指出的那样。奇怪的是,我的RTX 3070的性能比我的i9-10850K (训练一个批量大小为128的密集序列模型)要差得多,但这是我需要研究的一个单独的主题。
https://stackoverflow.com/questions/67959662
复制相似问题