Skip to main content

​
Scenario 1

For tasks that don’t require a GPU, such as coding, debugging, or data transfer, you can shut down and restart in “no-GPU mode.” This mode uses 0.5 CPU cores, 2GB memory, and no GPU, costing $0.1/hour. Data remains unaffected, and you can switch back to normal mode later.
No-GPU mode releases the GPU and sets it to an idle state. If the GPU is rented by other users during normal startup, there may be insufficient idle GPUs available.

​
Scenario 2

Start with one GPU for debugging and validation, then scale up to multiple GPUs for parallel training to save costs. Refer to Scale Configuration for details.

​
Scenario 3

If you’re unsure how long your code will run and want to shut down the machine immediately after completion, you can use the shutdown command (preferably with the full path: /usr/bin/shutdown).
Save your program’s logs, as they will no longer be visible after automatic shutdown.
Example Commands:
bash
# If your original command is:
python train.py

# You can append the shutdown command:
python train.py; /usr/bin/shutdown      # Using `;` ensures shutdown runs regardless of the previous command’s success.
python train.py && /usr/bin/shutdown    # Using `&&` ensures shutdown runs only if the previous command succeeds. Choose based on your needs.
Alternatively, you can call shutdown from within your Python code:
python
import os

if __name__ == "__main__":
    # Your code here
    os.system("/usr/bin/shutdown")