Seeking ideas for benchmarking hardware platforms for ArcGIS Pro use

[PeterWang]

My employer builds bespoke high performance computing private cloud platforms for use by energy companies who run Geophysical and Reservoir Simulation software. We'd really like to learn how to tune  our racks optimally for ArcGIS Pro as well, as every one of our customers has Arc Desktop.

 What are some ArcGIS Pro Geoprocessing tools which heavily utilize multi-threaded CPU resources, which consume RAM, and are heavy on disk I/O as the process runs? That's where we want to tune. Are there any Geoprocesses which do the computations in GPU as opposed to CPU? If so, we'd want to examine those as well.

When I think of GIS analogs to the kind of work we do now with seismic data, I think of large raster datasets, or large LiDAR point clouds. Seismic datasets often run into 200+ GB per project... that's not uncommon... if you know of any kind of public-domain GIS data that might be as taxing as seismic, I'd like to hear about it. We already have some 200+ GB USGS 3D stacked seismic cubes... can we convert those into a truly 3D GIS dataset?

This GIS Newbie thanks you!

[Robert_LeClair · Accepted solution]

Peter - only a guess on may part as I'm not savvy on Deep Learning in ArcGIS Pro but based upon my limited reading I would say the Deep Learning tools in ArcGIS Pro would heavily rely upon multi-threaded CPU/GPU resources, consume RAM, etc..  You can read more about Deep Learning here and here - good luck!

[PeterWang]

Great suggestion...many of our clients are interested in Deep Learning generally, so including those algos as part of the hardware benchmarking would be useful to them. Thank you!

[PeterWang]

The Deep Learning really sucks up RAM, CPU, and there's lots of disk activity as well. It's a good all-around test. I downloaded a pre-defined deep learning model and am searching for building outlines on a GeoTIFF in my city.