Hyperparameter tuning XGBoost on GPU



I need to reduce the time that RandomSearchCV takes to perform tuning of hyper parameters for XGBoost.

It takes around 36 hours for ~ 5 lakh training set with 300 features for tuning to complete and another 12 hours for training the model.

Would getting GPUs help to reduce the time? From forums I am also getting to understand that sklearn’s RandomSearchCV has no GPU support.

Has anyone trained XGBoost using GPUs? I would welcome any thoughts/ideas from anyone who has solved similar problem.