A graph convolutional neural network (GCNN) was trained with millions of molecules to accurately predict molecular photo-optical properties by scaling data loading and training to over 1,500 GPUs on the Summit and Perlmutter supercomputers at the OLCF a