TACOMA GPU Port

Project Description

The proposed project has several benefits.
First, the project will provide key information that will guide future decisions by GE regarding HPC investments, in particular the role of GPUs. The promise of GPU is quite clear as are the published performance gains on research codes. However, GE is unlikely to make multi-million dollar investments in computer hardware unless the performance benefits are known. This creates a classic chicken and egg dilemma that access to National Laboratory resources can help resolve.Second, GE’s internal compute capability cannot compete with the resources available at the US National Laboratories and GE resources must by necessity be dedicated to high TRL production design. Accesses to the resources at the Oak Ridge National Laboratory provide the opportunity to investigate computationally expensive computations of low TRL concepts. Given that Titan now consists of CPUs and GPUs, TACOMA will need to be ported to the GPU in order to enable GE to compete for grants, e.g. INCITE, for Supercomputer time.Third, the programming model used to run scientific codes on the GPU has been evolving. The release of OpenACC in 2011 provides a viable path forward for complex legacy industrial codes to take advantage of GPUs. We anticipate many learnings regarding what works well and perhaps not so well. Such knowledge would be useful to others considering porting a large scale legacy solver and will be useful in guiding the evolution of OpenACC and OpenMP capabilities and standards. We intend to talk publically about these learnings.