【Cloud GPU】How to use RunPod.io

This time, I introduce Runpod cloud GPU service.
https://www.runpod.io/

1. What is Runpod?

RunPod is one of the rental cloud computing services, we can rent high-performance machines with pay-per-use.
I use this when need high computational resources in kaggle competition.

We can use about ~80GB VRAM / ~250GB RAM.

2. How to use

1. visit the web page and sign up to runpod.
2. Charge the money.
3. Choice the GPU from "pods" tab, and deploy the pod.

4. Connect to pod. basically, I use ssh.(also can use jupyterlab or TCP)
   when use ssh, you have to generate your public/private key pair.

   ssh-keygen -t ed25519 -C "[e-mail]"
   

   and register public key to settings.
   

5. Connect to the pod by following the connect instructions in the pod details.
   

2.1 Error handling

If you get the Could not establish connection to "XXXXXXXXXXXXXXXXXXX": Could not resolve hostname. error in vscode, try to do this.
Adding the below to .ssh/config file:
・Open config file(or open from file manually)

nano ~/.ssh/config

・Add the config

~/.ssh/config

Host myserver
    HostName 190.68.000.00
    User root
    Port 22085
    IdentityFile ~/.ssh/id_ed00000

finally, try to Cmd + Shift + P, Remote-SSH: Connect to Host..., and select myserver.
This error is because the SSH extension in VSCode expects the hostname and port to be specified separately.

3. Notes

1. While running pod, the fee is paid automatically.
2. When turn off the pod, all of the data will lost, recommend you move to another place like github.
3. The pod while turn off requires a little bit fee, when you delete the pod completely, the charges will stop

4. Summary

Cloud machine has so high performance. When you need the machine power, please consider using Cloud GPU survices.

Reference

[1] RunPod Official