All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=733590

Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?keys=agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgIDg5rHcsgkM,agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgIDg5uCmuQoM,agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgIDgpoegpAsM,agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgIDg5orKtAkM,agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgIDg5siztwsM,agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgIDg5qCgqwsM,agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgIDg5ra6sAoM,agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgIDg5qySsgkM


Bot(s) for this bug's original alert(s):

webrtc-mac-large-tests
webrtc-win-large-tests

Over to Jerome to verify that the perf changes (~12% more memory, ~2% more CPU) are reasonable.

~2% more cpu usage is expected since the denoiser is enabled.

The memory usage is probably caused by the extra ref frame buffers allocated by the denoiser. It allocates another 4 reference frame buffers. 
It should be safe to remove the alt-ref such that the memory usage can be reduced by 1/4.
I'll make a change to address it next week.

The change to remove alt-ref from vp9 denoiser has been rolled in. The graph shows the memory usage is reduced by approximately 1/4 of the memory usage increase in this issue.