Difference in Sniper results Version 6.1 - 8.0

[Jake Merritt]

Hi everyone,

I recently ran some simulations on Spec2006 benchmarks using the newest version of Sniper and compared the results to some results my lab produced a couple years ago using Sniper version 6.1. I did my best to ensure the experimental setup was as close as possible between the two experiments. I got some results that I don't quite understand. For some benchmarks (libquantum, hmmer), my IPC over time plots between 8.0 ("new") and 6.1("old") match quite closely, for others (lbm, mcf) the IPC_old > IPC_new, and for another subsection (omnetpp, sjeng), IPC_new > IPC_old. I was wondeing if anyone knew what changes to snipersim could explain these differences. This is a link to the charts I am talking about: 

https://docs.google.com/presentation/d/1YNh6HeoA2WjeTei_3FDBu_O6xmlg5H1dRYUSayh7fB0/edit?usp=sharing

Best,

Jake