[Yi]

Recalling the significant performance improvement from v7.2 to v8.0, I was a bit surprised to see the overall trend of the simulation speed of EnergyPlus is getting slower. The chart below shows the same models take more than 20% of time to run on average in v8.2 than in v7.0. Luckily the progress of computer hardware during the period is more than enough to make up for this.

[Yi]

Updated chart showing 95% confidence interval of mean simulation time, as requested by Larry B.

---

Also the benchmark suite is here: http://www.jeplus.org/wiki/doku.php?id=e...s:examples

[joel.travis]

Yi,  

Thank you for posting your results.  I did a similar analysis based on 80 models I had on my system.  The model run-time ranged from 9 minutes to 20 seconds.  For both 8.0 and 8.2 I set E+ to only run one model at a time to guarantee that processor allocation would not affect run-time.  My findings were that for model runs that took longer than 2 minutes the average performance improvement was ~30%.  The time-penalty came when the run time was under a 30 seconds.  For example, a model run that was 20 seconds took 30 seconds.  I was including outputs for all end-uses for each time period so this would increase the input/output time.  I would expect this penalty could be mitigated if the number of outputs were reduced.  My findings are inline with the comments Amir Roth said on the Linkedin discussion that the input/output time is slower but the computational speed is faster.  Per Amir's comment I am interested in rerunning my test on a computer with an SSD drive.  My previous test was done on a 5400rpm speed hard drive.  With that said, I think that the 8.2 update does help by roughly 30% for where the E+ run-time is the most painful (large models that have a run-time that is larger than 1 to 2 minutes)

Thanks for kicking off this discussion.

Results are included in the link and attachment below:
Image
   

[Yi]

Hi Joel, that's very good news! Hope the results from a set of larger models I am still waiting for will confirm your findings. Cheers, Yi

[Yi]

New results using the DOC models. This is such a mystery now...

[Yi]

E+ benchmark results updated to v8.3