recommendation for cuDNN for Titan Xp + CUDA 8.0 + Ubuntu 14.04 + TensorFlow

Accelerated Computing CUDA CUDA Setup and Installation
#1

Hi, I’m barely installed CUDA 8.0 on VNC when I started learning ML.

I’d like to install cuDNN for Tensorflow, but I can’t find any recommendation cuDNN for Titan Xp.

I’m using CUDA 8.0, Ubuntu14.04, driver 384.11

I’ll use TensorFlow as well. I found that TF support cuDNN v5, so I have to install v5 ?
Would be prob if I install higher version?

Can anybody help me? :0

Or can I automatically install cuDNN with some package manager?

With Thanks,
Ashley.

#2

You will want to use whatever cuDNN your tensorflow binary is linked against, if you are doing a binary install.

Typically, recent TF versions like 1.4.x are built/linked against CUDA 8 GA2 and cuDNN 6.

If you build your own TF, you can use other combinations.

#3

Thanks!

I istalled cuDNN v.7 before seeing ur comment,

Unfortunately, I cannot import tf.

There’ was “ImportError: libcudnn.so.6: cannot open shared object file: No file or directory.”

so I’ll try switch v.7 to v.6

By the way,

I’m using Anaconda (conda 4.3.30) at global directory(at server, PATH is opt/anaconda3)

Do I have to use python after “activate tensorflow” ? (I mean, virtual env.)

Thanks again,