search

googlecolab / colabtools

Python libraries for Google Colaboratory
Apache License 2.0
2.17k stars 701 forks source link

Colab not Utilizing Multiple GPUs on Custom Colab VM #2909

Open ni-metre opened 2 years ago

ni-metre commented 2 years ago

Describe the current behavior I can create and deploy a Colab VM with increased memory and 2 GPUs. I can connect a Colab notebook to the custom runtime (following these instructions. However, it seems like 1 of the 2 GPUs from the custom deployment are being utilized. i.e. When I run tf.test.gpu_device_name(), I only see/device:GPU:0, and when I compile and run my model using tf.distribute.Strategy, I only see one GPU ('0') being used.

Describe the expected behavior I expect to see 2 GPUs listed when running tf.test.gpu_device_name() when connected to a Colab VM deployment launched with 2 GPUs.

What web browser you are using Chrome

cperry-goog commented 1 year ago

This slipped past me sorry. We only support a single GPU connection today, this is a great feature request. Tracked at b/262734546

ben-howt commented 1 year ago

Hi @cperry-goog , could you point me to any update on this request?

In the absence of a fix, could I suggest adding a note when provisioning VMs for Colab in cloud console, that only 1 GPU will be available to the Colab instance?

Kuni27 commented 1 month ago

Hi @cperry-goog, I was trying to find out if we can use multiple GPUs in Google Collab today and stumbled across this discussion. Can you please let me know if this feature is available in the Google Collab yet?

aditipanda commented 1 month ago

Hi @cperry-goog, I was trying to find out if we can use multiple GPUs in Google Collab today and stumbled across this discussion. Can you please let me know if this feature is available in the Google Collab yet?

I'm also looking to use multiple GPUs on a colab-pro notebook. Is it supported yet?

cperry-goog commented 1 month ago

Not yet on our managed runtimes.