GCS large-scale projects are characterised by projects that require a large amount of computing time over
longer periods of time.
In detail, projects are classified as
large-scale if they require at least 2% of the systems' annual production
in terms of estimated availability.
This requirement can be translated into different values.
For the supercomputers Hawk and SuperMUC-NG we give the requirement in terms of core-hours.
Thus, the projects' minimum requirements should be the following for these two systems:
- >= 25000 node-h on Hunter or
- >= 45 Mcore-h on SuperMUC-NG
JUWELS consists of two different modules with different hardware architectures.
Thus, the minimum computing time requirements for GCS-LS projects cannot be coherently specified by use of core-hours,
but EFLOP, based on the system's peak performance, is
used instead.
The projects' minimum requirements summed over all requested modules should be:
- >= 45000 EFLOP per year on JUWELS
Please refer to the
JUWELS fact sheet for further information.
Projects which do not fall into the category
large scale are called
GCS regular projects.