will the speed same if i convert both xgboost model and lighgbm model to .so use treelite?

Currently, I have tested to convert lightgbm model or xgboost model to .so file use treelite & tl2cgen, but i find that even though the depth(7) and tree number(500) are both same for two model, the .so converted from lightgbm is much faster than xgboost when predict one row, is this normal? If so, how can i speed up the xgboost for one row predict