FAQs

[stgeke]

• What's the relationship to nek5000

nekRS is the successor of nek5000

• How to get started?

See https://github.com/Nek5000/nekRS/blob/master/README.md

• What meshing options are available?

See https://nek5000.github.io/NekDoc/quickstart.html#meshing

• What element types are supported.

At the moment only hex elements.

• What polynomial order N is recommended?

N=5-9 (code supports N=1-10), although N=7 is typically preferred from a performance standpoint.

• What GPUs are supported?

Support is currently available for GPUs from NVIDIA, AMD, and Intel. Preference is given to HPC-optimized GPUs for better performance, as opposed to desktop versions.

• How about CPU support?

The code is not limited to GPUs and runs on CPUs too.

• What local problem size is required for good parallel scaling?

This depends somewhat on the considered case and hardware. Typically values for a 80% parallel efficiency range from 3-5M gridpoints (defined as n/P = E * N^3).

• Can I still use my nek5000 case file?

Yes, but userbc, uservp, userf and userq cannot be imported automatically and need to be re-implemented natively.

• Is there any documentation available?
  We're currently in the process of creating a user guide, but it's not available yet. In the meantime, feel free to check out the examples.

• Why does a segmentation fault occur while executing pack/unpack device + deviceBuffer MPI using pw?

The MPI implementation doesn't seem to support device buffers.

• Why does the code crash during device initialization saying invalid device?

An invalid device ID was requested. Note that each MPI rank utilizes one GPU. By default, the local rank is used as the device ID. On most HPC system the closest GPU (to a given rank) is mapped to 0 (through the job scheduling system or other settings). In this case you want to overwrite the default by specifying --device-id 0 as a command line argument.

• What examples are available

See https://github.com/Nek5000/nekRS/tree/master/examples
...