Testbed Support for GridPP member institutes
> [mailto:[log in to unmask]] On Behalf Of Andrew McNab said:
> Logical CPUs means hyperthreading changes it. If it's per-core,
> then it should be sufficient.
The fundamental relation is that LogicalCPUs*HEPSPEC should be the maximum deliverable power of the subcluster. In general I would say that LogicalCPUs/PhysicalCPUs should equal the number of job slots or the number of cores*hyperthreads per CPU, whichever is smaller, and the HEPSPEC should be calculated by running as many copies of the benchmark as that value.
Stephen
|