Abstract
We present CONQUEST, the first GPU-accelerated conformer search algorithm which uses the Spin- Component-Scaled MP2 level-of-theory. CONQUEST’s global minima outperform competing methods on the GMTKN55 data set, and its top-5 local minima are of a higher quality than those previously reported in lit- erature. Additionally, for medium-sized ligands, CON- QUEST is more accurate than competing approaches. Finally, due its GPU acceleration, CONQUEST can find structures within comparable times to semi-empirical methods and can scale to multiple nodes further reduc- ing runtime.