How to install tensorflow2.0 in cuda9?
lucasjinreal opened this issue · comments
Any way to specific a cuda version with tensorflow2.0 alpha version? To reinstall cuda is really risky, cause there are many other projects based on caffe which does not support an advanced cuda
@jinfagang Is there any version of Tensorflow already installed on your system?
@jinfagang TF2.0 was built with CUDA10.0. But, I think you can use CUDA9.0 with TF2.0. You need to make sure that the CUDA and cuDNN are referenced correctly. Thanks!
@jvishnuvardhan Sorry, I'm not the correct person to be assigned
@jinfagang You need to build TF2.0 from source, then you can use CUDA 9.0 which you already have it. Please don't use binaries as they are built with CUDA10.0 and some of the dependencies will be looking for CUDA 10.0 paths. Please let us know how it progresses. Thanks!
@jinfagang Please check the resource here. Thanks!
@jviereck Thanks, I never built tensorflow successfully yet, most reason is that it downloads too many dependencies and some using google which blocked in China................ I meant if there any wheel just like pytorch does provide difference version built on cuda 9 with cudnn 7
@jinfagang Do you have access to https://github.com/tensorflow/tensorflow/tags and https://pypi.org/project/tensorflow-gpu/? You could download from those locations. Thanks!
Closing due to lack of recent activity. Please update the issue when new information becomes available, and we will open a new issue. Thanks!
I don't find the tf2.0 package built with cuda9.0 from https://github.com/tensorflow/tensorflow/tags and https://pypi.org/project/tensorflow-gpu/. And it's troublesome for me to install tf by building. Finally, I give up and adopt the shell from https://github.com/phohenecker/switch-cuda to switch cuda version as a temporary solution.
@jvishnuvardhan you are right. I have succeeded to build TF2.0 with CUDA9 and cudnn6.7 from source. Now it's working grateful. Thanks.
as a temporary solution.
I don't find the tf2.0 package built with cuda9.0 from https://github.com/tensorflow/tensorflow/tags and https://pypi.org/project/tensorflow-gpu/. And it's troublesome for me to install tf by building. Finally, I give up and adopt the shell from https://github.com/phohenecker/switch-cuda to switch cuda version as a temporary solution.
Tensorflow2.0 officially built with cuda10.0. There are community built binaries are there but not sure whether they have anything you are looking. Please check some here. Thanks
hi @SmileTM will you share the binary for us? thanks.
@sudonto ，hi
This binary may not be useful to you , because it was modified based on my configuration.
if you need the binary, you must bazel the binary in your configuration.
you can refer my method.
hello, I meet some probleam by your blog, can I add your . qq . or can u share the tensorflow 2.0 which build by cuda9.0?
hi @SmileTM will you share the binary for us? thanks.
@sudonto
The binary in the here. But i am not sure ,it works well in your machine.
If it works in your machine, please reply me.
https://github.com/SmileTM/Tensorflow2.0-GPU-CUDA9.0