Scheduled Downtime
On Friday 21 April 2023 @ 5pm MT, this website will be down for maintenance and expected to return online the morning of 24 April 2023 at the latest

NWSC-3 benchmark timing results

This post was from a previous version of the WRF&MPAS-A Support Forum. New replies have been disabled and if you have follow up questions related to this post, then please start a new thread from the forum home page.

This might not be exactly a post-processing question but I wasn't sure in which subsection to post it. Taking a look at the CSEP times for the 30 km benchmark shows
My initial expectation was that there results were measured at Cheyenne but the numbers are not clicking (for me). Would it be possible to get some information on the hardware and conditions used to perform these benchmarks? Thanks.
I'm not familiar with the CSEP benchmark. Do you have a link to where you obtained this spreadsheet, or do you know who performed these benchmarks?
Sorry for the delay. The link is and is filed under UCAR_RFP000074. My initial assumption was that someone in your team performed the benchmarks but this seems to not be the case; hence, I don't know who actually did it.
Thanks very much for the link -- now I know that "CSEP" stands for Cheyenne Sustained Equivalent Performance!

My guess is that someone in NCAR's Computational and Information Systems Lab produced these numbers. As you're probably aware, Cheyenne does not have any sort of accelerators -- just Broadwell CPUs -- so my guess is that the columns "Cores" and "(seconds)" under the "Cheyenne" heading are simply the benchmark cases running on Cheyenne's regular compute nodes with the code compiled "as-is" with the Intel C and Fortran compilers (and probably SGI's MPT library for MPI).
Yours are similar to my first thoughts but there's more to it. As you mention, Cheyenne doesn't have accelerators so did they use some other HW or just estimated assuming some ratio? However, my foremost concern is with the full nodes results, which are supposed to be w/o accelerators. The last result points to a node (potentially a Cheyenne node) but even using Skylake processors, the time looks like too low for 36 cores (here, my assumption if that this time is the output at the end of the 'log.atmosphere.0000.out' file).
My question boils down to: Are there reliable benchmark results that can be used for comparison? Thanks.