The GAMESS ab initio quantum chemistry software package has been maintained for high performance on a wide variety of hardware architectures over the past few decades. In spite of this resilience, the rapid approach of exascale computing brings substantial and unfamiliar challenges. Most methods in computational chemistry are not easily adaptable to the trend of manycore processors with high degrees of explicit parallelism. Those methods which are easily adapted are rarely implemented with hardware features such as wide vector lanes and high bandwidth memory in mind.
 
In this talk, I will first describe the methods in GAMESS which have been successful on modern leadership-class systems. Next, I will describe our efforts to improve the performance of these methods for modern hardware architectures. Finally, I will discuss our analysis of the performance and energy-to-solution of GAMESS calculations for novel low-power hardware architectures.