We can very likely do this, although the change won't necessarily be smooth (I'd expect a momentary jank as we re-produce all tiles).

This will likely reduce memory by something like 4-8MB, so not sure if it's worth the visual degredation. I'd want to better understand how often we're hitting OOM on 1GB devices.

Also, while I think this is doable, I'd suggest staying in 16-bit mode for the lifetime of a page once we drop to it (jumping back and forth is a harder heuristic).

We have AndroidMemoryNotificationForeground histogram that records memory pressure events reported by Android. For 1GiB phones we have this:

                            User Count    Usage     Count
TrimMemoryRunningModerate   5,105,517     29.58%    35,008,783
TrimMemoryRunningLow        8,139,950     47.16%    75,711,803
TrimMemoryRunningCritical   10,992,364    63.69%    230,997,220
LowMemory                   4,224,929     24.48%    16,696,903

https://uma.googleplex.com/p/chrome/histograms/?endDate=20180313&dayCount=1&histograms=MemoryAndroid.NotificationForeground&fixupData=true&uniqueUsers=true&showMax=true&filters=platform%2Ceq%2CA%2Cchannel%2Ceq%2C4%2Chw_ram%2Cge%2C512%2Chw_ram%2Clt%2C1024%2Cisofficial%2Ceq%2CTrue&implicitFilters=isofficial

I.e. 11M users generated 231M CRITICAL events, ~20 per user on average.

FWIW, it doesn't seem like all these critical events result in foreground OOMs: https://uma.googleplex.com/p/chrome/histograms/?endDate=20180313&dayCount=1&histograms=MemoryAndroid.NotificationForeground%2CTab.RendererDetailedExitStatus&fixupData=true&uniqueUsers=true&showMax=true&filters=platform%2Ceq%2CA%2Cchannel%2Ceq%2C4%2Chw_ram%2Cge%2C512%2Chw_ram%2Clt%2C1024%2Cisofficial%2Ceq%2CTrue&implicitFilters=isofficial

On the other hand, what we don't know is whether the user experience is significantly impacted by slowness, even though they don't OOM.

We don have a pretty high renderer hang count for this category of devices: https://uma.googleplex.com/p/chrome/histograms/?endDate=20180313&dayCount=1&histograms=MemoryAndroid.NotificationForeground%2CStability.Counts%2CTab.RendererDetailedExitStatus&fixupData=true&uniqueUsers=true&showMax=true&filters=platform%2Ceq%2CA%2Cchannel%2Ceq%2C4%2Chw_ram%2Cge%2C512%2Chw_ram%2Clt%2C1024%2Cisofficial%2Ceq%2CTrue&implicitFilters=isofficial

I've recently learned that increasing memory pressure on device causes an increase in renderer hangs.