Navigation and service

Exascale application preparation services

The first Exaflop-capable supercomputers are imminent, and Juelich Supercomputing Centre is preparing accordingly with a wide range of associated activities [reference to NIC-Symposium paper will be added]. These include the installation and integration of the large GPU-based JUWELS Booster module expected to enter production by the end of 2020.

Apart from coming architectural changes, why should you care? Better scalability of your applications enables you to tackle bigger scientific problems. Imagine possibilities arising from much larger system sizes, increased resolution, or more detailed problems because execution time or memory allocations are less of a constraint.

Talking about time, efficient large scale jobs also mean you do not have to wait for your results to become ready at the end of the project period after months of running jobs. Ideal scalability assumed, your turnaround time will shrink with job size and you are ready to start the next project earlier. You might get your results from a single job.

The current JUWELS Cluster now accepts jobs with up to 1024 nodes as part of standard scheduling. Please refer to the page listing specific hints for tuning large jobs and known (and resolved) issues.

Beyond that, to prepare users of JSC computer systems and their applications, our extensive education, training and support services continue to be expanded to assist with the efficient use of current HPC systems and get ready for upcoming systems.

Activities to especially help large scale jobs:

Big Days/Week

Following the tradition of the previous "Big Week" event series on JUQUEEN , starting from February 2020 JSC will regularly offer "Big Days" on JUWELS. During a “Big Day”, the system will be reserved exclusively for the execution of capability sized jobs. In the future, the Big Day may extend to a Big Week again and also include other systems like the upcoming JUWELS Booster module. For more details please refer to More information

Scaling Workshops

JSC has a long tradition of scaling workshops to help users run on large fractions of any given system at that time. In the past this resulted in applications running with up to 1.8M threads on JUQUEEN, as shown e.g. by the High-Q-Club. The next annual workshop dedicated to assisting qualified teams of application developers with scaling and optimising their codes on JUWELS CPUs and GPUs is scheduled for 25-29 May 2020. More information

Additional Assistance

For assistance preparing for any of these events, contact your project advisor or request a free performance assessment of your code from the Performance Optimisation and Productivity (POP) Centre of Excellence.

Contact

Please contact the SC support at sc@fz-juelich.de for questions concerning any of these events.