I am an engineer and do a lot of hydraulic simulation. My primary modeling system is HEC-RAS developed by the US Army Corps of Engineers. The software was recently updated to support 2D hydraulic simulations. 2D simulations can be extremely complex and take hours to complete on a typical workstation. HEC-RAS is a 64bit, multi-threaded application that seems to be well optimized. To get a better handle on what systems we should run the simulations on I have started benchmarking systems with one of the sample models provided with HEC-RAS. It can take anywhere from 1 to 3 hours to complete depending on systems.
Having tested several different machines, a pattern seems to be evolving that more cores are better than faster cores and a really fast hard drive is a plus. Some examples are:
• E5-1650 V4 in AsRock X99 board w/ quad channel DDR4 at 2400 and NVME – 49 minutes
• E3-1505M V5 in HP LT, DDR4 2400 and NVME - 58 Minutes
• W3880 DDR3 1333, HP Z400 w/ Seagate SSHD – 1 hour 22 Minutes
• W3565 DDR3 1333, HP Z400 w/ WD HHD – 2 hours 9 Minutes
• FX-8350 DDR 1333, on GIGABYTE GA-990FXA-UD5 with SATA SSD - 1 hour 11 Minutes
All times are based off multiple runs.
Last night I ran a test on a HP Z640 with an E3-1630 V3, DDR4 2133 Quad Channel and a WD Blue drive. It completed in 48 Minutes. There went my pattern.
The differences between this machine and the E5-1650 V4 above are:
• The E3 has two less core than the E5
• The E3 is 0.1 GH fast base but O.2 GH slower boost
• The E3 is one generation older than the E5
• The E3 has slower ram (both are ECC Registered)
• The E3 is running windows 7 instead of windows 8.1
• The E3 system uses a HHD (WD Blue) and the E5 uses a NVME (Crucial M500)
• The E3 has a C612 Chipset and the E5 is X99
Everything seems to point to the E5 system having the advantage. Could the chipset have made a difference?
11/21 Correction: HP Z640 has an E5-1630 not an E3. Sorry for any confusion.
i Have not used HEC-RAS in years and Im sure its changed and evolved quite a bit but seems like HEC-RAS wants cores. I think i find the most suprising from your test results are the AMD FX-8350 is 13 min behind the xeon E3. I think the program might be CPU cache limited or something. I woukd really love to see what a Ryzen CPU or a dual xeon E5-2670 would do.
After install, run this model for testing: C:\Users…\Documents\HEC-RAS\Example Projects\2D Unsteady Flow Hydraulics\BaldEagleCrkMulti2D
Open model, then from “Run” menu select “Unsteady Flow Analysis” with the following settings (see image). Run time is output at completion of simulation. Run at least twice.
For lols I ran the test on my system and got 1 hour and 5 minutes. My system is a Xeon E3-1231v3 3.4Ghz base 3.8Ghz base with 4x4GB 1600mhz DDR3 and a Plextor M8Pe 512GB NVMe ssd.
I also noticed that when running the program, my cpu utilization never exceeded like 60%. It would flat line at 60% in hardware monitor and Task Manager for long stretchs of time before dipping momentarily and then returning to 60%. I just installed the program and the demos and ran it, with no other setup.
Not sure which numbers are the good ones here, so I took a screenshot (this is the first run):
Oh, CPU utilization didn’t fall under 97% at any time I looked.
Ryzen 1700X 3.95GHz, 32GB (8x4) at 2933MHz 14-15-15-15-34, Samsung 960 EVO NVMe.
I’ll run a second time in a bit.
Edit: Second run. Complete Process time dropped to 42:26 which I suspected might happen due to Zen’s architecture regarding repeat calculation optimizations (they call it neural net prediction).
Hope this helps! I’d be real curious if someone with a Threadripper would run it.
