Issues creating a working KNIME deep learning gpu environment

Hi Everyone,
I’m using the AutoML workflow and I was able to create the TF and Keras cpu based environments through the KNIME GUI successfully. However when trying to do the same for the gpu based environments for deep learning. Under Keras it says Keras lib not properly installed and same for tensorflow… Tensorflow library not properly installed.

Any ideas what I am doing wrong?

Thanks in advance,

Hi @j_ochoada
Have you created a complete new environment (e.g. with conda) for this GPU setup? If you want to use this second environment, you must change to this environment via Preferences > KNIME > Python Deep Learning.

Tensorflow does require different packages to be installed between the CPU and GPU version and you cannot mix the two. That’s why a second environment is necessary.
Please check for more information.

You can see more on GPU support by Tensorflow itself on (GPU support is available on Windows and Linux and requires a GPU-card with a CUDA architecture).


Hello @JanDuo,

just to add info that when linking documentation you can use latest instead of date (2019-06 in your case) to always point to latest publish documentation :wink: (if that is what you are after)



Hi @ipazin,
Thanks for the advice! I wasn’t even aware of this feature :slight_smile:
I followed a chain of links to get to this documentation (which I find difficult to find when starting at


Thank you @JanDuo and @ipazin. To clarify/elaborate yes I was following the 2019-06 page and had successfully created two separate env for the cpu through the gui. The same process failed for the gpu with errors (for dl env) that keras was improperly installed. I’ve uploaded a partial screenshot. This is a brand new conda install version 4.84 that knime was trying to bulid environments from. I’ll take a look at the latest to see if anything has changed. Thanks!

@JanDuo and @ipazin

I just wanted to follow up as I believe I have found the solution to what was causing the issue. So in my case my KNIME install is in my local linux directory and I am launching it on the cluster with x forwarding for the GUI. I have choices on the cluster and some nodes have gpu and others do not. I typically specify a cpu only queue and didn’t think anything of it as I figured the gpu environments only cared about the conda env. It appears actually it is checking to ensure that the env is operational wherever the work is taking place. So in short if I re launch KNIME on a node with gpu’s everything works perfectly.

Probably something that everyone else takes for granted but it took me a while to fully understand. Thank you both for your help!



This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.