The tremendous successes that GPUs have had in accelerating molecular simulations must continue to be matched by advances in their application to challenging simulation preparation, analysis, and visualization tasks. We will describe how the latest developments in the molecular visualization tool VMD exploit GPUs using exciting new features of CUDA, OpenACC, EGL, and OptiX to accelerate key science tasks on clouds, clusters, and petascale computers. We will summarize our early experiences and performance results on GPU-accelerated OpenPOWER platforms with an eye toward the challenges and opportunities posed by the upcoming DOE Summit and Sierra systems.
An accelerated OpenAcc based discrete ordinate (DO) radiation solver is now available in ANSYS FLUENT R18.0. The accelerated solver uses the OpenACC programming model with the PGI compiler and speeds up the radiation computations on NVIDIA's GPGPUs. The OpenACC programming model is effectively used for simultaneously computing on CPU and GPU thereby utilizing the maximum computing power a GPU equipped machine.
HPC compiler directives were originally used as hints to enable optimizations. OpenMP extended directives into de facto language extensions. What are the differences between OpenMP and OpenACC directives? How do they relate to parallel constructs in standard languages? This talk explores these issues and a practical path forward for using HPC compiler directives.