Hello everyone,
First time posting here 
We have a working XGBoost regression model which we are hyperparameter optimizing in a paid Google Colab notebook with GPU support. It is working fine but taking a lot of time. We are thinking to go for the fastest publicly available cloud multi GPU instances and have decided to try out the 8x A100 multi-GPU options both from AWS (p4d.24xlarge or p4de.24xlarge) and GCP. Has anyone successfully utilized 8x A100 GPUs with XGBoost? If so, are there any caveats, comments, concerns and tips? Any links to tutorials or papers? Many thanks in advance! Currently we are using the Python interface.
Warm regards,