biyer
biyer

Reputation: 11

Sagemaker Notebook ml.g4dn.xlarge instance do not list GPU device

I have created a sagemaker notebook with instance type ' ml.g4dn.xlarge' with amazonei_tensorflow2_p36 kernel. with tensorflow version 2.3.2, tf.config.list_physical_devices('GPU') returns PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')

However, when I change the tensorflow version to 2.4.1 (installed with !pip install --upgrade pip && pip install tensorflow==2.4.1 && pip install tensorflow-gpu==2.4.1), tf.config.list_physical_devices('GPU') returns []

How can I use GPU with tensorflow version 2.4.1?

Upvotes: 1

Views: 1710

Answers (1)

rok
rok

Reputation: 2785

I cannot find any reference in the documentation, but I suspect the way you are updating Tensorflow is not totally supported by Sagemaker. Sagemaker usually uses prebuilt containers, that already contains tensorflow, but also CUDA. If you just update Tensorflow there is no guarantee that will be still compatible with the installed CUDA.

Upvotes: 1

Related Questions