Hello team,
since 463 and since a few days the simulations of allmost everything (artefacts, talents, BiB) took approximately twice time as before. E. g. a simulation that should take 60 seconds approx, and were done within 50 seconds in the past, takes 140 seconds, now.
https://www.askmrrobot.com/wow/simulator/report/b87fcdee6b6f45ce9847f5e009b28cb2
I’m running the client @ Windows 7, 64-bit, AMD FX 6300 Six-Core w. 3,5 GHz - not the fastest one tbh., but well, it did the job fine until few days ago.
Cheers
ChessLion
Strange, as far as I know, nothing significant was changed in the simulator… you didn’t happen to change the mode for your client to use less threads, or have anything else running on your computer… anything like that?
I am not sure if this is something to do with the patches microsoft released for Meltdown and Specter. It affects the performance of Intel CPUs and as far as I know it has some implications for AMD CPU (reported blue screens and slowdowns).
@yellowfive Does any of the libraries you use for the client uses speculative execution? If so, the performance will be affected.
It was my understanding that the speculative execution they were talking about in all that stuff is done at a very low level, e.g. branch prediction and stuff, processor-level logic. Also, I believe the performance impact of the patches was largely exaggerated… and also not as big a deal on consumer computers, but I’d have to look up details on that to be certain.
That said, I checked the commits and there was only one change to the simulator in this update that would have no significant performance impact, and another change that might slow down pantheon trinkets a tad (we have enough data now to take a better guess at how the triggering mechanic works).
The patch only slows down switching into and out of kernel mode (to stop Meltdown). This could mean accessing the file system, sending and receiving traffic through ports, etc. There might be a performance hit to the auto-updator, but for running sims itself I would assume the main bottleneck is all the math being done while inside user mode with only the occasional call to the OS kernel.
