Look into accessibility of further GPU options

[rk1a]

Other options aside from the FZI server could be:

• University compute
  • Tübingen doesn't work with docker
• Kaggle
  • check whether works with docker
• Google Colab
  • check whether works with docker
• Lambda GPUs
• Microsoft Azure
  • need to pay, but works with docker
  • could be good to test out our code on the validation instance
  • From AICrowd "submission evaluation and training will most likely use Azure NC6 instances with 6 vCPUs, 56GB of RAM and a K80 with 12GB of VRAM

[gekaklam]

At the list on the left side of the page there's a number of resources. This was from an old iteration of the Fast.ai course.

https://course19.fast.ai/start_gradient.html

[lauritowal]

Could someone of you @gekaklam or @rk1a check which of those are useful for us?

[rk1a]

Other options aside from the FZI server could be:

• University compute

  • Tübingen doesn't work with docker

• Kaggle

  • check whether works with docker

• Google Colab

  • check whether works with docker
• Lambda GPUs

• Microsoft Azure

  • need to pay, but works with docker

  • could be good to test out our code on the validation instance

  • From AICrowd "submission evaluation and training will most likely use Azure NC6 instances with 6 vCPUs, 56GB of RAM and a K80 with 12GB of VRAM

• Kaggle does not support Docker or installing dependencies like JDK
• Colab supports dependencies, but not docker; also installing takes a long time..
• Lambda seems to suppor Docker:https://lambdalabs.com/blog/set-up-a-tensorflow-gpu-docker-container-using-lambda-stack-dockerfile/