I am having trouble getting Deeplearning4j to use cuDNN for training. I am using CUDA 11.2 and cuDNN 9.3. It seems that CUDA is successfully installed, but not cuDNN. If anyone can assist. I feel like I have used most online resources, but it just won’t work. Thank you in advance.
Thank you for the reply. I am still struggling and would like to know what versions of CUDA and cuDNN are recommended when using deeplearning4j?
@CrossFawn1 could you be more specific? I gave you the guide there and in the dependency it shows both the cudnn version to use and the cuda version. You can see cuda 11.6 right here:
https://repo1.maven.org/maven2/org/nd4j/nd4j-cuda-11.6/1.0.0-M2.1/
Were you just hoping a random version of cudnn/cuda was going to work? I’d love to help you if you could tell me about what you actually tried.
Ok, so am I correct in saying that to use cudnn all I need to do is add this:
org.nd4j
nd4j-cuda-11.6-platform
1.0.0-M2.1
windows-x86_64-cudnn
dependency? according to the cudnn section on the deeplearing4j docs.
If so, I tried adding it to my POM.xml and got the following:
Could not find artifact org.nd4j:nd4j-cuda-11.6-platform:jar:windows-x86_64-cudnn:1.0.0-M2.1 in central (https//repo.maven.apache.org/maven2)
use nd4j-cuda-11.6 and nd4j-cuda-11.6-windows-x86_64-cudnn.
-platform is a generic multi platform builder.
The one without a classifier contains the classes. The one with contains the native platform code.