You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In order to compile the full GACODE package on a new machine, a version of the **platform files** must be selected. You can use existing ones or create your own. After cloning the repository, you can query the existing platforms using::
27
+
In order to compile the full GACODE package on a new machine, a version of the **platform files** must be selected. You can use existing ones or create your own. After cloning the repository, you can query the existing platforms using
26
28
27
-
$ ./gacode/shared/bin/gacode_platforms
29
+
.. code::
30
+
31
+
$ ./gacode/shared/bin/gacode_platforms
28
32
29
33
If you decide to create new platform files (say, for a new platform MINE) then create the following files, starting from one of the existing ones (say, CONDA_OMPI_GNU)::
The (a) Multi-platform strong-scaling comparison for CGYRO test case \texttt{nl03}, showing wallclock time vs. number of nodes. Frontier is by far the best performer on both a per-node and maximum performance basis. (b) Kernel-level analysis.
182
-
Left (darker) bars indicate compute time; right (faded) bars indicate the communication time. Data is normalized to the total time, such that total bar area is constant (1.0). Lower compute-to-communication ratio on GPU systems reflects the extremely high performance of the GPUs. Note the significant improvement in communication management from Summit/Perlmutter to Frontier.}}
178
+
The (a) Multi-platform strong-scaling comparison for CGYRO test case \texttt{nl03}, showing wallclock time vs. number of nodes. Frontier is by far the best performer on both a per-node and maximum performance basis. (b) Kernel-level analysis. Left (darker) bars indicate compute time; right (faded) bars indicate the communication time. Data is normalized to the total time, such that total bar area is constant (1.0). Lower compute-to-communication ratio on GPU systems reflects the extremely high performance of the GPUs. Note the significant improvement in communication management from Summit/Perlmutter to Frontier.
0 commit comments