-4

I was trying to run a cycleGAN-keras code from GitHub, which recommended on:

  • ubuntu 16.04;
  • python 3.6;
  • keras 2.1.2;
  • Tensorflow 1.0.1;
  • NVIDIA GPU+CUDA8.0 CuDNN6 OR CuDNN5.

But my GPU is RTX3090 (CUDA Version: 11.4) and the driver cannot be modified. CUDA10 and below are not supported in Tensorflow1., so I did this to build CUDA11 to fit Tensorflow1.:

conda create -n kr_py36 python=3.6
conda activate kr_py36
pip install nvidia-pyindex
pip install nvidia-tensorflow[horovod]
pip install nvidia-tensorboard==1.15

but when i run: netD_A = n_layer_discriminator(image_size) i got error:

InternalError: cudaGetDevice() failed. Status: CUDA driver version is insufficient for CUDA runtime version

I think the correspondence between versions of cuda, TensorFlow, cudnn must be wrong. I'm a computer noob and stuck in there for almost two days. I would appreciate it if you could solve this problem.

NiRui_961203
  • 1
  • 2
  • Please clarify your specific problem or provide additional details to highlight exactly what you need. As it's currently written, it's hard to tell exactly what you're asking. – Community Aug 15 '23 at 15:36
  • `and the driver cannot be modified.` well you are pretty much stuck then. Something in what you are building demands a newer driver. – Robert Crovella Aug 20 '23 at 19:34

0 Answers0