mirror of
https://github.com/JHUAPL/kaiju.git
synced 2026-01-09 23:27:57 -05:00
53 lines
1.5 KiB
ReStructuredText
53 lines
1.5 KiB
ReStructuredText
Computational Costs
|
|
===================
|
|
|
|
GAMERA-HELIO+GL model
|
|
---------------------
|
|
|
|
Estimates of computational costs in cpu-hours (NCAR) or SBUs (NASA) for
|
|
different model resolutions based on Helio runs
|
|
|
|
On the Derecho supercomputer:
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
|
|
AMD Epyc Milan (128 cores)
|
|
~~~~~~~~~~~~~~~~~~~~~~~~~~
|
|
|
|
|
|
256x128x256 - 64 ranks on 8 nodes
|
|
160 Hour Spinup + 100 Hrs CME propagation - ~1000 cpu-hours
|
|
|
|
512x256x512 - 256 ranks on 32 nodes
|
|
160 Hour Spinup + 50 Hrs CME propagation - ~8-10,000 cpu-hours
|
|
|
|
1024x512x1024 - 512 ranks on 64 nodes
|
|
160 Hour Spinup + 50 hours of CME propagation time - ~85k cpu-hours
|
|
|
|
2048x1024x2048 - 1024 ranks on 128 nodes
|
|
160 Hour Spinup ~490k cpu-hours, 32 hours of CME propagation
|
|
~125k cpu-hours
|
|
|
|
On the Pleiades supercomputer:
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
|
|
Cascade Lake Nodes (40 cores, 80 threads) Using HT and AVX-512
|
|
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
|
|
|
|
256x128x256 - 64 ranks on 8 nodes
|
|
160 Hour Spinup + 100 Hrs CME propagation - ~200 SBUs
|
|
|
|
512x256x512 - 256 ranks on 16 nodes
|
|
160 Hour Spinup + 100 Hrs CME propagation - ~3200 SBUs
|
|
|
|
1024x512x1024 - 512 ranks on 32 nodes
|
|
160 Hour Spinup + 50 hours of CME propagation time - ~52k SBUs
|
|
|
|
Broadwell Nodes (28 cores, 28 threads)
|
|
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
|
|
|
|
256x128x256 - 64 ranks on 16 nodes
|
|
160 Hour Spinup + 100 Hrs CME propagation - ~300 SBUs
|
|
|
|
512x256x512 - 256 ranks on 64 nodes
|
|
160 Hour Spinup + 100 Hrs CME propagation - ~4000 SBUs Broadwell Nodes
|