The OpenACC programming model represent an effective and flexible tool to accelerate large and complex software packages. By means of compiler directives the work can be distributed between CPUs and GPUs in order to leverage the impressive computational capabilities of modern GPUs.<br/>In this contribution we describe the current status of a long-term collaboration between Gaussian, Inc. and NVIDIA aimed to enable the Gaussian program to take advantage of heterogeneous CPU+GPU hardware configurations, which are becoming more and more common in academia, supercomputing centers and in industry.<br/>We will show a degree of performance improvement which (i) is well balanced among the many kinds of calculations the program can perform (HF and DFT ground and excited states energy, geometries, frequencies and properties), (ii) is undoubtedly significant in terms of raw speed-up, and (iii) is affordable in the sense that it justifies the acquisition of servers that include both CPUs and GPUs, rather than only CPUs.

Performance of the Gaussian program on GPUs using the OpenACC programming model: Achieving a balanced, significant and affordable speed-up with respect to CPU-only architectures

255th National Meeting (2018)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.