Programming GPU-Accelerated POWER Systems with OpenACC on the Nimbix Cloud
March 20, 2019

TUTORIAL: Programming GPU-Accelerated POWER Systems with OpenACC (L9112) on the Nimbix Cloud

In the tutorial, attendees will learn how to leverage the full performance offered by GPU-equipped IBM POWER systems – the architecture which powers the world’s fastest supercomputers Summit and Sierra. The POWER architecture features a powerful CPU which is connected to a number of very powerful NVIDIA GPUs (up to 6) by a tight and fast interconnect called NVLink. OpenACC, a directive-based programming model, can be used to accelerate an application to use all parts of the system. In the tutorial, the POWER architecture and its features are presented. OpenACC is introduced in small steps and used to accelerate a simple serial application in guided hands-on exercises. Within the exercises, the attendees can explore the programming model and the acceleration opportunities on POWER nodes in the Nimbix cloud.

NVIDIA GTC tutorial

Image Copyright OpenACC / Julia Levites

Learn about & register for GTC here
March 18-21, 2019
Silicon Valley

The tutorial is for any programmer who has reached the limits of their serial or CPU-parallel application and thinks about taking the leap to the realm of massive parallelism. OpenACC offers a simple model to target GPUs in a portable and backwards-compatible way. It’s a good fit also on the POWER architecture. The tutorial is for you, if you have no GPU programming experience (but always wanted to try this famous technology), if you want to get to know OpenACC (and find out how it’s different from OpenMP), if you’re curious about the POWER+GPU combination (but don’t know exactly what’s that all about); or if you are keen to run your very first program on the POWER9/Volta platform (but never got the grant for a project).

Andreas Herten graduated in physics before he started as a scientific researcher at Jülich Supercomputing Centre (JSC) of Forschungszentrum Jülich – one of the largest research centers in Europe. He consults on and optimizes HPC applications which want to leverage GPUs for their scientific workload. He co-organizes trainings in and beyond JSC (usually CUDA, OpenACC, but not only) and is involved in GPU Hackathons, week-long intensive GPU coding events. He also co-administrates JURON, a prototype-like supercomputer based on POWER8+ processors located in JSC. He likes Python and LaTeX and would script everything if he could.