Odynews (Q1 2023)

A new GUI to enhance HPC experience in the public cloud

We are pleased to announce the availability of ‘GUI for Numerical Predictions in the public cloud: Graviton3 & 3E.’ A preview is now available for the 7th generation of AWS EC2 instances powered by Graviton3 with a free trial* of 10 days; it includes CMAQ and WRF. The GUI allows model users to download data, preprocess and run the apps within the same graphical environment. A GIS environment facilitates visualization and enhances user experience. The GUI also has the capability to copy files directly to and from S3 buckets so there is no need for external storage or other on-premises resources, if so desired.

The GUI as an AMI is now available (https://aws.amazon.com/marketplace/pp/prodview-fcy4llnqba5gk) and a SasS version will be available later in the year.The GUI facilitates direct (and faster) download of GFS and NAM data to be used by WRF.

The AMI includes all the files necessary to run the CMAQv5.4-U.S. Northeast (2018) and CMAQv5.3.3.3-U.S. Southeast (2016) benchmarks. The following screenshots show the GUI running the CMAQv5.4-U.S. Northeast (2018) and the interface with S3 buckets.

*The free trial applies only to AMI fees; AWS will charge all other costs at the applicable rates.

Benchmarks for CMAQv5.4

With the release of CMAQv5.4 at the end of October, we thought that having new performance measurements is important not only to understand any difference between versions but also for future reference as CSPs introduce new hardware. Along with the new version of the code, CMAS released a new benchmark covering the U.S. northeast area. This benchmark is slightly larger (105 x 100) than the U.S. southeast area (80 x 100), which has been the standard benchmark for the last few years. The resolution is still the same (12 km) and even though the increase in area is only 31.25%, wall times increase in the 65-75% range. Therefore, the new domain dimensions would render a direct comparison quantitatively moot but, from a qualitative perspective, the measurements reflect similar trends. We have measured wall times for the U.S. northeast 2018 benchmark with the most advanced AWS 6th and 7th generation instances. The figure shows wall times for a simulation covering a period of 24 hours with hourly outputs:

N.E. benchmark results

A more demanding test case for CMAQv5.4 is the new 2018-12US1 CONUS (continental U.S.) benchmark with a grid of 459 by 299 points and a 12 km resolution. The input files from are available from the AWS OpenData S3 bucket s3:::cmas-cmaq-modeling-platform-2018 and include 2 different mechanisms (CB6r5 and CRACMM). The figure shows some measurements with the new instances: 

CONUS benchmark results

In a similar fashion to our early measurements from last year, the performance of Graviton3 was very impressive with computational performance very close to that of x86_64 processors, but at a substantially lower price.