Learning Forum

[gemoore_uw]

Could anybody provide me details on how to report a performance metric for a given Ansys HFSS solution, so that collaborators could compare their simulation times on their machines to mine?nFor example, on my Windows 10 system (a dual-processor Xeon E5-2682 running at 2.5GHz with 16 cores each and 128GB of RAM) the total elapsed CPU Time read from the Ansys HFSS Solutions:Profile tab is 42:21 min, with a real time execution time of 20:16min. However, these results are specific to my system, and I'd like to generalize the results of my simulation to other systems. Also, I don't think I can multiply the CPU Time by the Clock Rate, since I don't know how Ansys was utilizing parallel processing.nI was thinking that I could use Resource Monitor to determine the excess processing while Ansys was operating, and then multiply the CPU Time by the averaged Clock Rate minus background processing, to result in CPU Clock Cycles, which seems more generalizable.nAre there better figures of merit for reporting processing load for a given simulation?nThanks for the assistance!n

[rtk]

nCould you please share the profile data information that you are referring to in this post?. nPlease impinge the image/screenshots in the comment rather than uploading it as file.Best Regards,nn

[AndyJP]

HFSS is very specific to the system. It may differ even between systems with similar specs. (like Dell V.S. HP)nHFSS can sometines create a high load which can hit a RAM bus bottleneck, or CPU thermal package limit. And the license manager system is vveeeeerrrrryyyy slow and uneliable, which can add extra 15% penalty to a large set of small simulations (custom optimization scripts) depending on your license management.nnGeneral considerations for reaching top performance in HFSS are 4-6 cores per worker, 1-2 workers per CPU shared cache. Maximum per-core clocks and then maximum cache size as a secondary parameter. nAvoiding multi-CPU workstations, but using fast and low-latency multichannel memory (4-6 channels per CPU). It is better using an HPC cluster of independent workers, than feeding the sweep to one multi-CPU machine.nI do not have ideas on RAM bank/rank configuration yet.nIn general, i7 systems with disaled TDP/boostdown (like ASUS boards offer) and OC-oriented low latency unbuffered RAM offer much more performance. The same way unbuffered RAM offers more reliability problems, which you should consider when populating the board with more than 4 sticks.n