Fail to use 4070Ti

General Discussion
,
#1

Hello,

I’m currently using a 4070Ti GPU with an AMD 5800x CPU on Windows 11.

My installed CUDA version is 11.7 and the cuDNN version is 8.8.

I’ve encountered an issue when utilizing TensorFlow versions 2.12 or 2.11 - no GPU devices are detected (the function ‘tf.config.list_physical_devices(‘GPU’)’ returns an empty list).

When I switch to TensorFlow version 2.10, I receive an error message while training a model :

Could not locate cublasLt64_12.dll. Please make sure it is in your library path!

Is there a possible method to enable the usage of a 4070Ti GPU in TensorFlow?

Thank you in advance for your assistance.

#3

@Jue_Wu,

Welcome to the Tensorflow Forum!

According to the tested build configurations CUDA 11.8 and cuDNN 8.6 are compatible with latest TF version.

The NVIDIA GeForce RTX 4070 Ti supporting start with CUDA 12.x. Hence i recommend you to build from source.

I’ve encountered an issue when utilizing TensorFlow versions 2.12 or 2.11 - no GPU devices are detected (the function ‘tf.config.list_physical_devices(‘GPU’)’ returns an empty list).

To work with the latest versions of Tensorflow on windows you can follow WSL2 instrcutions. Please refer to the instructions here Install TensorFlow with pip.

Thank you!

1 Like