“Oh sure, the following tests aren’t as scientific as putting all the browsers in a ring and seeing which one is left standing after the fight, but it’s close.” More here.
What’s The World’s Fastest Browser?
About The Author
Eugenia Loli
Ex-programmer, ex-editor in chief at OSAlert.com, now a visual artist/filmmaker.
Follow me on Twitter @EugeniaLoli
For me, it doesn’t matter if the browser renders a webpage in 4.76 seconds or 4.89. What’s much, much more noticeable on my PC is the GUI speed, that’s where I can see the biggest difference between web browsers’ speed and that’s where Firefox (or Mozilla/Seamonkey) is by far the slowest piece of software I use due to the massive overhead of the XUL based GUI (several times slower than “native” GUI, like pure GTK+ or Qt or the Windows GUI). Plus, I’d really like to know why the Firefox GUI is so much slower on Linux/BSD than it is on Windows (I know this is generally the case with many cross-platform apps, the Opera GUI is also several time slower on Linux, but again, why?).
Which is .12 seconds of my life I’ll never get back waiting for a stinking browser to render, adds up, that’s why it matters. I don’t think Firefox is based on XUL, Mozilla is based on XUL, that’s one of the big differences. XUL applications have to be adapted to firefox.
I don’t think Firefox is based on XUL,
You think wrong. Firefox’s ugliness and poor feel, as well as its laggard speed, can be largely attributed to XUL. Compare to K-Meleon or other native Gecko-based browsers to see the difference.
XUL needs to be taken out back and shot.
XUL vs. native toolkit has nothing to do with look and layout of the browser. All XUL is is an abstraction layer for the native toolkit. I’m right there with you if you want to talk about GUI responsiveness (which is something firefox could do well to improve on). Apps that use the native toolkit directly are going to be faster than ones that use an abstraction layer. But if you want to say that firefox is ugly because it’s run through XUL is just not accurate.
Though abstracting several toolkits means that certain generalisations must be made, and inconsistencies will occur when compared to a native UI.
For instance, if Win32 supported something GTK didn’t (or the other way round), support would need to be hacked in, and this might look somewhat ugly or not match themes or something. The same occurs with Swing and SWT on Java, or wxwindows.
Damn, thought it already had gone native, what was the point of rewriting mozilla then?
“I don’t think Firefox is based on XUL”
You think wrong.
happens.
Did go into the http://scragz.com/tech/mozilla/test-rendering-time page (the link is in the news)
Results:
Apple Safari 2.0.4 on Dual G5 2.0Ghz PowerPC – Mac OS 10.4.7 Tiger
1st. time rendering: 3.386000156402588 seconds
2nd. time rendering: 2.934999942779541 seconds
This I agree with. I can’t stand sluggishness and thus avoid Linux for daily work.
Damn, this was supposed to be a reply :\
Edited 2006-09-08 05:04
Too outlandish, rather poor effort. Trolls should be at least a little bit believeable to work.
Lynx!
Indeed, I have to agree.
Who needs graphics and CSS1,2&3 support when all you want is speedy browsing.
– Jesse McNelis
Yeah, baby: less is the new more!
and don’t forget links2, elinks, w3m, and if you wan’t something “beefier” dillo!
You just can’t go wrong with the classics!
Plus, when you’re using lynx, a modem internet connection is actually useable!
Don’t people have anything to do than benchmarking browser speed? Loading a lot of data (not common) is not the best way to test rendering speed. Complex pages with CSS 2.1, Javascript and XHTML, and images are more suited to be benchmarked because they are close to the real world, not some stupid and fixed HTML rows. A better benchmark should be to generate the page dynamically with javascript in the browser and measure that.
Edited 2006-09-08 06:10
For example, this page is slow on some browsers, and significantly faster on others, while scrolling:
pong2.berlios.de
That’s due to transparency effects.
CSS 2.1 and XHTML are not the problem. A well crafted page using these is a smaller file than an older page with several levels of nested tables. CSS reduces complexity.
But you are right about Javascript. Specifically, I think advertising schemes like intellitxt.com and kontera.com that double underline random words to insert advertising links are the real guilty parties on a lot of forum sites.
Edited typo.
Edited 2006-09-08 20:17
On my 800Mhz G3 iBook i get between 9.3 – 10.6 on firefox and safari.
Maybe I should take this as a hint to upgrade…
You do not need to upgrade yet! In your case this benchmark likely measured your download speed and not the rendering speed.
On my MacBookPro I get abount 8 seconds in Opera and Safari. If I open the HTML file from the hard drive it takes abount 0.3 seconds in Safari and about 2 seconds in Opera.
BTW the file has a size of 796 KB and my max. download speed is 1024kbit/s.
I’ve executed the same test for the most of the latest browsers on MS WIN XP platform.
Resume:
* Swift 0.1 is a winner.
* MS IE 7 is a looser.
Details:
http://leoz.org/blog/?p=16
Why are nonsense articles like this one keep being pusblished on OSAlert.com ?
Not only is the default (and often overlooked) Gnome browser Epiphany rendering almost as fast as Swiftfox
Come on !
Both are using the very same Gekko engine…
Leo.
“Why are nonsense articles like this one keep being pusblished on OSAlert.com ?”
There should be a OSNonSense.com site to publish these articles. So any editor that needs to publish crap can go there and satisfy themselves. Of course OSAlert main page would be mostly empty.
Edited 2006-09-08 08:33
I was surprised that the author’s results had IE coming out faster than Opera, so I tried it for myself under Windows XP.
#1 Opera 9.01: 6.01
#2 IE 6.0.2900: 6.54
#3 Firefox 1.5.0.6: 8.68
Not to mention Opera’s Javascript implementation is much, much faster than either browser. The only downsides to Opera is the kinda slow loading time, and rendering problems on some websites.
This is my own test:
http://scragz.com/tech/mozilla/test-rendering-time
PII 498MHz
RAM 384MB
System – Kubuntu 6.0.6; KDE3.5.2
Test__IE6*_____O 9.01____Konq
1.____ 15.07 ____ 16.40 ____ 28.06
2.____ 14.86 ____ 15.81 ____ 24.36
3.____ 6.12 ____ 15.90 ____ 25.47
4.____ 14.54 ____ 18.83 ____ 40.79
5.____ 14.80 ____ 17.23 ____ 24.38
6.____ 14.80 ____ 16.51 ____ 23.14
7.____ 15.48 ____ 16.39 ____ 23.10
8.____ 14.61 ____ 16.21 ____ 28.44
9.____ 14.82 ____ 16.33 ____ 23.43
10.___ 14.61 ____ 17.17 ____ 35.84
11.___ 15.58 ____ 15.99 ____ 26.92
12.___ 14.76 ____ 15.73 ____ 42.11
——————————————————
Avg**: 14.84s ____ 16.39s ____ 28.08s
* – http://www.tatanka.com.br/ies4linux/index-en.html
** – (A-a(min)-a(max))/(C-2)
To me, this whole browser rendering speed debate is really splitting hairs. What matters to me is how long it takes from the time I click on the short-cut to when it’s actually usable to start browsing.
What matters to me is how long it takes from the time I click on the short-cut to when it’s actually usable to start browsing.
BS, you are only saying that becasue IE seemed faster than FF. If they benchmarked startup speeds and IE won a bunch of people would say “who cares, all that really matters is render speed”.
While we are on topic though, I noticed that IE 7 takes seemingly forever to start.
No, actually I don’t particularly care about how much tenths of a second another browser renders. I care about how long it takes for me to start a browser or tab instance.
I’m almost always doing multiple things online at once, so render speed isn’t a factor, and unless it was hidiously slow (and I’m talking in the 5-10+ second mark) I don’t care.
I don’t know your configuration, but on my computer, IE7 b2 loads a great deal faster than Firefox does.
Edited 2006-09-08 12:23
Internet Explorer isn’t actually an executable. If you go under program files and delete iexplore.exe, you can still get internet explorer by typing a URL into an explorer window.
IE starts up much faster on windows systems because the DLLs are always preloaded in the background. This is due in part to the fact IE is integrated into the core of windows and loaded with explorer.
Really? My favorite thing about IE7 is that it loads up literally instantly for me, while IE6 took a second or a few, depending.
Yesterday an article about what Konqueror is about, and today an article that states that Konqueror beats all when it comes to performance. KDE’s marketing-machine running? Ah well it is a great browser anyway and has come a long way the last 1,5 year (probably apples’ webkit has something to do with it as well: more people working on (almost) the same code). Too bad so many sites want you to change the ID to firefox to get all the html-code, not really Konqueror’s fault, because it can run the sites when the ID is changed. I hope at least google is going to recognise that konqueror can show their sites, they’re supposed to be open minded towards open source.
It is the reason why I like Konqueror so much as a Web-Browser:
– Speedy startup compared to Firefox (at least for us KDE users).
– Fast rendering.
However, there still are some sites, who insist on using bad java/javascript/flash code. For these I will start up firefox (and kmail, to make their webmaster become aware of their errors
).
Would it not be better to publish an article on what browser rendered the largest percentage of web pages correctly? I mean a second or two, difference. You can often start reading the top of the page before the bottom has appeard.
It would probably be some crappy version of IE that comes out best, but it would be useful to see what the others are like in comparison. Konqeror does some useful things I prefer over firefox, but in my experience, Firefox is generally more likely to render things “correctly” (not in the Acid2 test type meaning of correctly – more like how the page is expected to look).
Where are the results for Opera on Linux? On Windows, he shows Opera is a bit faster than FireFox, but Opera on Linux is apparently not tested.
In my experience Opera has a significant speed advantage over both IE and Firefox, but the important difference isn’t in the rendering speed that’s tested here. An extra second here and there due to rendering speed is completely insignificant to me, most of the time I open pages in the background and don’t even see them load. The important thing is responsiveness and efficient use of system resources, I don’t want the browser to slow to a crawl or consume all my PC’s resources when I push it heavily.
In Opera I tend to keep interesting pages open rather bookmarking and closing them, and often open a large number of pages, for example using the Links Panel to simultaneously open all the interesting links on a news site. They load in the background while I read another page and are there for me to flick through using ctrl+tab. Obviously YMMV, but I find that to be a fast and efficient way of browsing through large quantities of information on the web.
Opera can cope with that and stay responsive and stable, something that isn’t true of the other browsers I’ve tried. Being able to browse heavily without interruptions, using a GUI that I find more efficient and elegant, saves me a lot more time than slightly faster rendering.
Well put.
There are also other factors involved with browsing efficiency and speed.
– Opera’s fastback cache system is the best in my experience
– Opera has a lot more shortcuts. You can easily add more through mouse/keyboard shortcut configuration which is limitless.
– Opera’s UI is more responsive than at least Firefox (XUL overhead. XUL is way more abstract than Opera’s themeing engine). IE is a toss-up.
There’s more too.
In Opera I tend to keep interesting pages open rather bookmarking and closing them, and often open a large number of pages…
I do that with Firefox. I often have ten pages open for a couple weeks. Some of them are articles I think I will eventually read (but don’t). But the problem with Firefox is that it tends to accumulate zombies, which consumes resources and slows me down.
Back when I was using Windows on a regular basis, I really loved Opera because it was a speed demon compared to anything I had ever seen. Especially the back button! If I remember, when I close all my browsers, I’ll use Opera next time.
Whatever the answer may it be sure isn’t in this article.
In all the cases I have experienced, Opera is the winner in rendering speed. It is noticably faster on all of my systems. No benchmark needed.
Just for kicks I tried this in IBrowse 2.3 which is 68k code running emulated on an 800MHz 750GX cpu (G3).
I performed the test locally to eliminate loading (which cut around 2 seoncds off the test if you’re curious) and ended up at 5,2 seconds.
AWeb 3.5.08 didn’t perform quite as well, as it ended up at 12,9 seconds.
I on my gentoo, compiled firefox was always slower than binary firefox. I tried swiftfox, and it was the fastest. This is because of disabled pango on swiftfox.
If I run firefox with “MOZ_DISABLE_PANGO=1 firefox” my builded firefox is faster than both. And compiled with –as-needed flag it start much faster than official and swiftfox
I on my gentoo, compiled firefox was always slower than binary firefox. I tried swiftfox, and it was the fastest. This is because of disabled pango on swiftfox.
If I run firefox with “MOZ_DISABLE_PANGO=1 firefox” my builded firefox is faster than both. And compiled with –as-needed flag it start much faster than official and swiftfox
I would take that as a hint that pango is not built optimally. Perhaps you should check your flags for pango and try rebuilding it.
Actually I think it’s just that Pango is pretty slow. A lot of /big/ speed improvements have been made in the latest version, however (I’ve seen figures around 35% for speed improvements).
Actually I think it’s just that Pango is pretty slow. A lot of /big/ speed improvements have been made in the latest version, however (I’ve seen figures around 35% for speed improvements).
That wouldn’t explain why the binary of Firefox, which uses pango, is faster than a compiled version of Firefox.
The threading/reply system on OSAlert sucks. I’m reposting under the right thread.
Edited 2006-09-09 04:23
…is hv3: http://tkhtml.tcl.tk/hv3.html
It is not at all ready, but seems very nice indeed! Passes ACID2, is fast and cheap on memory.
Check it out, if not for serious use, at least for testing. I think it really deserves more developers.
It’s just a bunch of text in one font. What about images, complex layouts, CSS, javascript, HTTP performance (pipelining and multiple connections)? Surely that is more important these days.
Edited 2006-09-08 13:21
You hit the nail right on the head!
I would like to add that all sorts of benchmark are somewhat useless unless the browsers’ performance would vary by an order of magnitude.
The order in which the page gets rendered is far more important than the rendering speed.
A browser might be fast as hell but if he puts an image in a place where a link was before so that I keep clicking on images instead of links, well, this pretty much renders the fastest browser useless.
> KDE’s marketing-machine running?
no, actually. both articles came as welcome surprises to us. marketing teams are wonderful and important things, but happy satisfied users that spontaneously sing your praises are even better
This was like THE most crapy test:)
1333Mhz G4 powerbook
Safari – 0,14 (No shit)
Firefox – 4,5
I have an old 450MHz PII with 384M RAM running XP pro.
On legacy hardware such as this, the difference in rendering is night and day in favour of IE.
Dillo (http://www.dillo.org/) is the faster browser.
I use Konqueror about 98% of time, in 1.5% I use the firefox and in 0.5% I use the explorer (with Wine or with VMWared win2000).
I think that Konqueror is the best existing browser now.
…is the word “unscientific”. There are so, so many variables that make a browser “slow” or “fast” that testing the way the author did is very disingenuous. In fact, all it’s really accomplished is starting a flamefest, both here and all over. Articles like that always do.
For instance, I was able to get *wildly* different results from Firefox by simply enabling and disabling extensions:
Rendering with the following extensions:
Talkback, Adblock, Colorzilla, Web Developer, Html Validator, XPather, Adblock Filterset.G Updater, Tamper Data
=6.5
Rendering with NO extensions:
=3.92199
I’m positive that IE with various BHOs will also affect its rendering speed as well. Did the author have any toolbars installed? He doesn’t say…
Seriously, you have to take this kind of test with a grain of salt and not let it get you pissed off. Clearly the author knew it would cause a firestorm. My bet is that if IE had “lost” it wouldn’t have even been posted. Same for Firefox articles in the same vein.
Some other thoughts:
-IE7 hasn’t been released fully yet…who knows how fast it will be. It could blow the pants off everything, or be slow as mud
-Mozilla depends on Pango, which is being optimized in an ongoing effort to make Gnome and GTK+ faster
Imo safari is the fastest of all browsers.
I’ve never felt a faster browser. Imo on the second rang, Konqueror.
OK, Safari uses KDEs KHTML, so it’s clear ^^
Links is the fastest browser by far.
I have to say this confirms what i have believed for some time that konqueror has been the fastest full featured browser for *nix since the 3.5.x series.
I used firefox for a long time but ever since kde 3.5.x has come out I have pretty much used konqueror exclusively except for the rare case that i find a page that won’t work on konqueror.
the reason for all of this? It is fricking fast, it is noticably faster then firefox on about 90% of the pages i go to and is right on par with Opera and seems to handle plugins better.
Edited 2006-09-08 20:05
I’m more interested in the following questions:
(1) Will it run on the platform(s) I use?
(2) Will it display the web pages I want to view?
(3) Will web pages lose any functionality?
(4) Why won’t Santa give me his list of naughty girls?
Edited:
Sorry… I mean (4) How secure is the browser during normal use?
Edited 2006-09-08 20:30
Here is my emerge info:
emerge -pv mozilla-firefox
These are the packages that would be merged, in order:
Calculating dependencies… done!
[ebuild R ] www-client/mozilla-firefox-1.5.0.6 USE=”gnome ipv6 java xinerama -debug -mozdevelop -xprint” LINGUAS=”-ar -ca -cs -da -de -el -en_GB -es -es_AR -es_ES -fi -fr -ga -ga_IE -he -hu -it -ja -ko -mk -nb -nb_NO -nl -pl -pt_BR -ro -ru -sk -sl -sv -sv_SE -tr -zh_CN -zh_TW” 34,721 kB
Total size of downloads: 34,721 kB
There is no use flag for pango. It can’t be compiled without pango. But it can be disabled on runing with MOZ_DISABLE_PANGO=1 firefox, or i have added to /etc/profiles export MOZ_DISABLE_PANGO=1 firefox
Presuming that the tests this fellow ran were not limited by bandwidth (ie, he ran them on a locally cached copy of the rendering page), I decided to run the same test here on Firefox. I have Ubuntu Edgy installed, with the latest updates, so I’m running Firefox 2.0b1. I have an Athlon64 3200+ with 512 MB RAM (32 MB shared with the onboard video), so it’s quite a bit slower than this guy’s test rig, however the same test took me only 1.74 seconds to render (averaged over five tests).
There wasn’t any apparent difference in rendering time between opening the file and between refreshing it once open. It’s probably also worth noting that I did not bother “cleaning up” my environment before running these tests; I still have Rhythmbox playing in the background, with Liferea, Evolution, XChat-Gnome, Nautilus, and Gnome-Terminal open also.
This either proves one of two things:
1. The test this guy performed is useless as a general benchmark because rendering speed is highly dependent on something specific and quite variable between systems.
2. Firefox 2.01b is a LOT faster than Firefox 1.5x.
Or 3, I suppose: a bit of both.
Ymmv…