E external/xla/xla/stream_executor/cuda/cuda_dnn.cc:407] There was an error before creating cudnn handle (302): cudaGetErrorName symbol not found. : cudaGetErrorString symbol not found.

[TRo202]

Hi.

Previously I had no problem using JAX on my PC, that has CUDA version of 11.

Now, I'm trying to use JAX on my laptop, whose CUDA version is 12.2.

So I installed JAX with the code below
pip install --upgrade "jax[cuda12_pip]" -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html

It was installed successfully, but when I used JAX, I got the following error:

2024-06-07 03:14:00.848518: E external/xla/xla/stream_executor/cuda/cuda_dnn.cc:407] There was an error before creating cudnn handle (302): cudaGetErrorName symbol not found. : cudaGetErrorString symbol not found
...
jax._src.traceback_util.UnfilteredStackTrace: jaxlib.xla_extension.XlaRuntimeError: FAILED_PRECONDITION: DNN library initialization failed. Look at the errors above for more details.

But when I checked

from jax.lib import xla_bridge
print(xla_bridge.get_backend().platform)

I did get gpu

I uninstalled JAX and JAXlib again and installed again with
pip install "jax[cuda]"==0.4.13 -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html
but I still got the same error.

I tried to check the cuDNN version
cat /usr/local/cuda/include/cudnn_version.h | grep CUDNN_MAJOR -A 2
but I got
cat: /usr/local/cuda/include/cudnn_version.h: No such file or directory
So I printed cuDNN version manually and I got

>>> import torch
>>> print("cuDNN version:", torch.backends.cudnn.version())
cuDNN version: 8902

CUDA version, cuDNN version, and JAX version seem to have no problem, but I don't know why I'm getting this error.

I've been struggling with this for hours, and I would greatly appreciate if someone could help me with this. Thanks in advance.

[prateekiiest]

Same here

[TRo202]

Since I couldn't get any help regarding this issue, I just had to install the lower version (11)...

[Topdp]

Have you solved it now? This question has been bothering me for many days.

[TRo202]

No... I just installed the lower version of CUDA (11.3)