Hello, @U04VAD6RL! Thank for your work! Which version of cuda and cudnn I should use for building mxnet from source? It doesn't work on cuda 9.0 and cudnn 7.0. Also, I pull compiled jar from clojars, install cuda 8.0 and cudnn 6.0, run gan example and failed. I can post stacktrace later.

Thanks @U0NBJDK4N so much for giving it a try! Here are the details of the jars https://cwiki.apache.org/confluence/display/MXNET/MXNet-Scala+Release+Process

Cpu version is working fine, btw

for linux it is CUDA 9.0 and CUDNN 7.0.5

wondering if you are able to clone the repo and run lein test ?

the stack trace might be helpful too

lein test working without errors, but only on cpu( in deps I choose [org.apache.mxnet/mxnet-full_2.11-linux-x86_64-gpu "1.2.0"]

Can you post an error you get from the gpu enabled build? It might give some more info

yep, i’m on it

lein test running without error, both on CPU and GPU versions, on a Docker container based on nVidia's nvidia/cuda:9.0-cudnn7-devel-ubuntu16.04 base image. It needs a dependency that's not packaged in Ubuntu (neither in 16.04 nor in any of the latter versions), but once you install it, it works as expected. I'm opening an issue in Github with all the relevant details. Edit: forgot to mention that the we run the Docker container on a Linux host (in case that makes a difference).

Thanks a lot for all of your help troubleshooting