Over the past couple of years, many of you asked us to de-emphasize (or completely remove) the memory usage tests from the Web Browser Grand Prix because memory is there to be used, after all. While we can't argue with that assertion, the fact remains that some browsers use far less memory than others in the exact same workload. But what we could not see on our modern test system was how the browsers scale their usage to the available hardware. In our last installment, we used a decade-old Windows XP test system. That scenario demonstrated that some of the worst memory hogs under Windows 7 dramatically scale back total memory usage on the older hardware, but still display the same content. So, memory usage is tied to the test system's available resources.
Therefore, memory usage is no longer being factored into the final scoring. However, the 40-tab test is still needed in order to achieve the memory management and page load reliability scores, and to get a general sense of browser responsiveness under load. Overall memory efficiency is now gauged by the difference between a browser's single-tab memory usage and -39-tab memory management total. The browser with the lowest score is able to return the most physical memory back to the operating system without actually closing the application itself, but simply by decreasing workload (closing tabs).
Chrome 20 keeps a tight grip on Google's memory efficiency lead, only hanging onto 94 MB more RAM than its pre-40-tab total. IE9 doesn't disappoint either, keeping just 117 MB to place second. Firefox 13 earns a very close third place, retaining far less memory than previous versions. Safari still holds onto 331 MB, placing it in fourth, while Opera 12 is in last place after closing 39 tabs..
The charts below contain the single-tab and 40-tab memory usage, as well as the -39-tab and -39-tab plus two-minute memory management tests.
IE9 uses half as much memory as most of the competition with only one tab open. Firefox has always had the lowest 40-tab memory usage total, but version 13 takes its single-tab total down to just 61 MB, which is right in line with Safari and Opera. What the composite score does not show is the speed at which the different browsers return memory back to the operating system. Chrome is the only contender to do this instantaneously. While Firefox and IE9 drop usage totals a great deal, they can take a minute to do so.
2. in the 40 tab test, try working in a tab during the loading of the 40 tabs. you will find lots of difference between browsers. FF hangs, Opera and Chrome remain fluid.
3. how about a test where a browser is using 1GB+ RAM and you are trying to open/close tabs. Then see the UI responsiveness. most browsers can easily handle 800MB RAM. but which browser easily handles 1.2GB+ RAM ?
i tested this and found that during a HTML5 benchmark, IE9 had the least CPU usage, and most GPU usage amongst all the browsers.
maybe you should do a few memory benchmarks with ABP installed just to realistically judge what 99.99% of FF users go through.
ABP works wonderful on Firefox, i RARELY see any ad. While I have used ABP on Chrome BUT its doesn't block half the ads.
I know its Not Google's fault, its just that ABP developers are putting more effort with Firefox.
So for me, Firefox > Chrome.
IE9 does so good on HTML5 HWA accelerated benchmarks because its able to offload more of the processing to the GPU. i tested this and found that during a HTML5 benchmark, IE9 had the least CPU usage, and most GPU usage amongst all the browsers.Really interesting, what utility do you use for measuring GPU usage?
How many firefox users dont use ADblockPlus ? very very less.Also ABP developer is a regular contributor to the Firefox source code.maybe you should do a few memory benchmarks with ABP installed just to realistically judge what 99.99% of FF users go through.I'd estimate ABP usage on FF at around 5% or less based on ABP and FF usage statistics. Besides, that would give FF an unfair advantage.
MSI afterburner for GPU. windows task manager for CPU.
i sent a mail regarding this to Chris. but maybe i sent it too late for this article...
but if you run 4 instance of dromaeo in 4 tabs, the CPU usage is still 25% (using only 1 core).
so chrome is not completely multiprocessing.
in IE10 beta, if you run 4 instances of dromaeo benchmark in 4 tabs, it uses all the for cores. so we can expect better multiprocessing from IE10 and win8 :)
Is Dromaeo (the DOM portion) working in Chrome for you? I could not get it to finish in Chrome or Safari on any of my Windows machines.
BTW, i run chrome dev version. so that could make a difference.