Remember Nov-2021 video reviews all claiming that M1 Pro / Max are extremely powerful browser-wise? As they easily drive dozens Chrome tabs open (but not visible 🙃), some even running Youtube videos, and the system is still responsive?
Is it actually a relevant benchmark for a heavy Chrome user? How can we properly measure tabs/memory performance of Apple Silicon?
My approach is to measure how many tabs can be kept loaded in memory without slowing down the system. Admittedly, such a test is over-stretching for a regular browser user.
Specifically:
I compared 32GB Pro side-by-side to 64GB Max (32 GPU cores), both 16" at Normal Power Mode.
TLDR:
Swap used on 32GB Pro vs RAM used on 64GB Max as the number of tabs grow:
Here is how number of tabs affect the system responsiveness on 32GB Pro. Each stage assumes some stabilisation period after initial load:
To ensure that content is always shown from RAM / swap, I disabled network connection for the 10-window setup; content in all tabs remained fully available (both on Pro and Max).
I used Chrome extensions: Copy All Urls to mass-open tabs; Revolver to auto-switch tabs every 1 second.
Chrome Version 97.0.4692.71 (stable release channel).
During the experiment, I had OmniFocus and TOR with ~10 tabs open all the time--which also took a tiny fraction of RAM.
Is it actually a relevant benchmark for a heavy Chrome user? How can we properly measure tabs/memory performance of Apple Silicon?
My approach is to measure how many tabs can be kept loaded in memory without slowing down the system. Admittedly, such a test is over-stretching for a regular browser user.
Specifically:
- open multiple, non-overlapping Chrome windows, all on the same screen
- launch multiple dozen tabs in each (I used personal bookmarks, mostly from Twitter)
- !! make each window switch tab every second--effectively forcing Chrome to keep entire session in memory / swap all the time
- keep adding tabs, keep track of swap used; see when memory pressure becomes yellow and/or system becomes less responsive
I compared 32GB Pro side-by-side to 64GB Max (32 GPU cores), both 16" at Normal Power Mode.
TLDR:
- at 380 tabs, the 32GB is fully responsive; 640 tabs makes system almost completely busy, seems close to a usable limit
- 64GB remains absolutely fluid up to 640 tabs (and will likely sustain several 100s more).
Swap used on 32GB Pro vs RAM used on 64GB Max as the number of tabs grow:
Here is how number of tabs affect the system responsiveness on 32GB Pro. Each stage assumes some stabilisation period after initial load:
- at 3 windows totalling 379 tabs, all of them are shown immediately after switch, system fully responsive, memory pressure is mostly green with occasional yellow
- 4 windows (407 tabs): memory pressure is always yellow;
- 5 windows (447 tabs): system becomes less responsive while new tabs are opened
- 6 windows (469 tabs): each tab appears with a fraction-of-second delay after switch, looks like it is read from swap most of the time
- 7 windows (514 tabs): on most of switches, 1-2 tabs appear blank as if SSD swap starts becoming a bottleneck
- 9 windows (608 tabs): on most of switches, 3-5 tabs appear blank; closing any single Chrome window takes 15-20 seconds
- 10 windows (641 tabs): 50%-70% of tabs appear blank after each switch, even after 30 minutes of stabilisation (SSD swap seems a real bottleneck now). Memory pressure becomes occasionally red. CPU cores are used around 90%, fans run at 2300-2700 rpm; total power reaches 48W. Opening the last 30 tabs takes 55 seconds (vs 5 seconds on 64GB Max). Clicking system menu takes few seconds for iStat icons. Looks like the system is completely busy with retrieving tabs from swap and displaying them, in a loop.
To ensure that content is always shown from RAM / swap, I disabled network connection for the 10-window setup; content in all tabs remained fully available (both on Pro and Max).
I used Chrome extensions: Copy All Urls to mass-open tabs; Revolver to auto-switch tabs every 1 second.
Chrome Version 97.0.4692.71 (stable release channel).
During the experiment, I had OmniFocus and TOR with ~10 tabs open all the time--which also took a tiny fraction of RAM.