+++ This bug was initially created as a clone of Bug #834003 +++ This is similar to bug 834003. The idea was suggested by bsmith on dev.platform. Vladan, any chance you could take a look into this, please? Thanks!

This is a comparison of the Jan 23rd Nightly built with MSVC 2012 (no PGO) against the same code built with MSVC 2010 (with and without PGO). I ran Talos 3 times for each build to generate 9 saved Telemetry pings. All Talos runs were performed after a system reboot. I used the same setup as bug 834003 The Nightly built with MSVC2012 (without PGO) performs roughly the same as the Nightly built by MSVC2010 (without PGO), and noticeably worse than the Nightly built by MSVC2010 (with PGO). The two non-PGO builds under-perform the PGO build on mostly the same histograms: XUL_FOREGROUND_REFLOW_MS, HTML_FOREGROUND_REFLOW_MS, CYCLE_COLLECTOR, FX_SESSION_RESTORE_COLLECT_DATA_MS, TOTAL_CONTENT_PAGE_LOAD_TIME, FX_PAGE_LOAD_MS, GC_MS For the majority of histograms, noise was not a major factor. The mean value of a histogram from a Talos run was less than 5% away from the mean value across all 3 runs. A few tests were more sensitive to noise, and I made a note of it in the attached results.

Thanks a lot for the analysis, Vladan.