Hi,
I have laptop with Nvidia Geforce GTX 1050ti. I have to run tensorflow in this for deep learning applications. Can you please suggest me which version of cuda, cudnn and tensorflow will work in this.
Thanks,
Deepak
1 Like
Hello,
I am also starting with tensorflow and have the exact same card on my laptop.
Lucky for you I went through the setup process already so the following is what I’ve found out:
Latest stable tensorflow is 1.12.0 and it does not support Cuda 10, only up to Cuda 9.
If you wish to use Cuda 10 with tensorflow you have to install the nightly build, at the time of writing 1.13.0 but I do not recommend it since it simply didn’t work for me when trying out some simple stuff.
cuDNN versions are matching to which version of Cuda you had installed, Nvidia has an online tool to allow you to download the correct version.
Tensorflow Python PIP package names:
Stable (gpu) - pip install tensorflow-gpu
Nightly gpu) - pip install tf-nightly-gpu
Stable (cpu) - pip install tensorflow
Nighly (cpu) - pip install tf-nightly
Another thing worth mentioning, for me installing Cuda 10 on windows installed a driver that windows didn’t want to accept because of digital signatures, I had to reinstall the driver from GeForce download page so it might be the reason why tf-nightly-gpu couldn’t work right but I doubt it.
I’d appreciate if some one more knowledgeable would comment on this issue, I am a beginner as well.