Maybe related to https://chromium-review.googlesource.com/1199627
otherwise also please see if you can find the cause.

Here's the regression range:
https://chromium.googlesource.com/chromium/src/+log/70.0.3538.0..71.0.3545.0?pretty=fuller&n=10000

That CL definitely could increase memory, but it should only be while in the tap switcher itself (we could go from having 2 images to 4, but I don't think this is drastically dissimilar to how it was before).

Looks like phones are worse off than tablets:
https://uma.googleplex.com/p/chrome/timeline_v2/?sid=93fdb24cf6d2c8fef7dab512abc9ba82

How did you get the regression range by the way?  If you split the canary by versions, could you find one where it jumped up?

I did some memory dumps locally and didn't see any bitmaps from this CL lingering.  +twellington as the NTP had the bulk of the bitmaps on the stack from my local testing, so would like them to take a cursory look through the  regression range.

Sorry I might have been wrong about the regression range. The memory seems to increase over time between 3545 and 3550 on canary. But, i can't tell for sure if it's not any version before 3545..

https://uma.googleplex.com/p/chrome/timeline_v2/?sid=7b08fee45c80bd34170f012368bab9a3
It also shows 400KB regression on canary..

Maybe one of these cls is responsible:
https://chromium-review.googlesource.com/1199768
https://chromium-review.googlesource.com/1217362

Could it be Duet going into experimentation? It looks like the Finch config for Canary targets 71.0.3543.0+ and started rolling out around Sept 5 or so.

When I split by Duet experiment group, the Enabled group line is ~500k higher.


https://uma.googleplex.com/p/chrome/timeline_v2/?sid=d32b311e1391384812d21c3954506889

This is expected, we're retaining a second bitmap that is the size of the bottom toolbar for animation (same thing we do for the top toolbar). This size will differ based on the resolution of the device.

We might be able to improve in this in the future (possibly individual icons on a solid color layer) but I won't want to implement that until we are sure duet is shipping.

In that case I will assign this to you and keep it open.
It also shows reduction in page load count, so not easy to compare metrics.
https://uma.googleplex.com/p/chrome/variations/?sid=a34cc9058af3761db436a468d516d1fd