Open MarioSegallaMoreira opened 4 years ago
Hi! Did you solve the issue yet? If not, please check if your GPU is available using tf.test.is_gpu_available(). If it returns false, then:
Hi! Did you solve the issue yet? If not, please check if your GPU is available using tf.test.is_gpu_available(). If it returns false, then:
- Check if some other process is using your GPU
- Check your BIOS or settings to see if you have somehow switched to integrated graphics. Let us know if you solved the issue
hello @divyanshusharma1709 , I'm facing the same isue. I'm working with AWS ec2; I have checked if gpu is available, and it does.
What might be the problem?
I've the same issue using tensorflow/tensorflow:1.15.2-gpu-py3-jupyter docker image.
I've the same issue using tensorflow/tensorflow:1.15.2-gpu-py3-jupyter docker image.
Not sure if this should be a new issue, but I found a possible reason for this:
On https://www.tensorflow.org/install/gpu?hl=nb it is reported that tensorflow 1.15 requires CUDA 10.1.
But I've looked through the Dockerfile for tensorflow/tensorflow:1.15.2-gpu-py3-jupyter
on https://hub.docker.com/layers/tensorflow/tensorflow/1.15.2-gpu-py3-jupyter/images/sha256-2c2ddc9780724ee528757f44beb16dac302a09ee7eb4e333b7dd85404597fdd9?context=explore and it seems that it installs CUDA 10.0 and not CUDA 10.1
If I'm mistaken, can someone please explain to me how? Because I have the same issue and I have yet to try if uninstalling CUDA 10.0 and installing 10.1 will fix my image. If I'm not, it would be great, if someone who has the authority to do so would replace 10.0 with 10.1 in the Dockerfile
Hello! I'm doing the object_detection_tutorial.ipynb and when I run the following command it gets this error: I did the protoc install and added all three pythonpath variables from models;models\slim;models\research.
cuda: 10.1 cudnn: 7.6.5