GPU not detected by tensorflow in SM Studio

Question

I'm currently using SageMaker Studio with kernel "Python 3 (TensorFlow 2.3 Python 3.7 GPU Optimized)".  
Tensorflow doesn't detect the GPU on both ml.g4dn.xlarge and ml.g4dn.2xlarge instances (with 1 GPU).  
I would appreciate any advice.  
  
import tensorflow as tf  
tf.config.list_physical_devices('GPU')  
: \[]  
  
Edited by: haganHL on Jul 29, 2021 11:55 PM

Answer

Downgrading to the Python 3 (TensorFlow 2.1 Python 3.6 GPU Optimized) kernel allowed the GPU to be recognized. However, I had to recode my notebook to support that kernel. What a shame. What a pain.  
  
Edited by: jmlineb on Aug 18, 2021 8:00 AM

Answer

I am having the very same problem with the very same kernel using the very same instance types. It worked perfectly a month ago. Attempting to downgrade my kernel to see if my notebook will still work.  
  
Edited by: jmlineb on Aug 18, 2021 7:47 AM  
  
Edited by: jmlineb on Aug 18, 2021 8:11 AM

Answer

Hi, I'm having the same problem here. I can't believe that Amazon hasn't fixed this - we could really use support for GPUs on new Tensorflow versions!

GPU not detected by tensorflow in SM Studio

Relevant content