Firefox 3.1 Beta 1 is pretty damn fast

[tolien]

The result is what it says on the tin - the average (of a series of runs) of the time it takes the browser to complete a benchmark suite. Obviously JS takes longer to execute on a slower machine, so you'd get a slower time - that's why you can't directly compare the times between machines.
There hasn't been a merge from the TraceMonkey branch since beta 1 was cut, so you're not really gaining much (in terms of JS performance) over a more stable beta.

The setting I mentioned in #2 turns on the JIT bits of the JavaScript engine for content, i.e. webpages. The other option for Chrome does the same for the browser's own JS (i.e. bits of the browser itself, extensions etc.) and is a bit more unstable (and unlikely to be enabled on the released 3.1).

Edit: Just updated the graph to include the last WebKit nightly too.

 

[mrk]

In that Sunspider test I got 981ms with 1.3% deviation

Is this good or bad?

Nobody knows!

 

[tolien]

On it's own, the value's useless. All it really tells you is your machine was twice as fast as mine (x2 3800+ at stock, 2GB PC3200), assuming the difference in Firefox versions makes little to no difference.

Obviously JS takes longer to execute on a slower machine, so you'd get a slower time - that's why you can't directly compare the times between machines.

One benefit of having a much faster machine is you will probably be able to run the tests on IE without tripping the "this script's been running since the dawn of time, maybe you'd like to stop it" prompt.

 

[mrk]

I just ran it in IE7 and got 18660.2ms +/- 5.0% deviation

The two scores together give me a prime example of why never to use IE

 

[mrk]

Thought I'd test it in person just to see and whoah, 3.1 is very fast.

 

[tolien]

Try Sunspider with 2.0.0.17, 3.0.3 and the 3.1 nightlies and the advertising (Firefox 3 faster than 2 etc) speaks for itself:

I didn't think the gap was quite so large though

 

[LeperousDust]

I dunno, I think they're pretty close

http://cgi.tolien.co.uk/pics/ocuk/sunspider.jpg
And the relative difference:
[img]http://cgi.tolien.co.uk/pics/ocuk/sunspider-relative.jpg
([url=http://webkit.org/perf/sunspider-0.9/sunspider.html]Sunspider[/url], average of 5 runs, lower is better, Firefox nightly is 20081017033517, TM = javascript.options.jit.chrome and javascript.options.jit.content both true, Opera 9.60, Chrome 0.2.149.30)

Expect javascript.options.jit.content to be set to true by default by the time 3.1 ships.[/QUOTE]

So you've got Nighties and Betas, and "Opera"? Which opera, which build :p? 9.6 is out and released now and the 9.6 weeklies are pretty damn awesome (if a little more unstable than usual i've found). Nice to see Chrome kicked Fx up the arse into doing something though, Opera's always had an easy time with the "competition".

 

[mrk]

I used 9.6 in my test above

 

[tolien]

Which opera, which build ? 9.6 is out and released now and the 9.6 weeklies are pretty damn awesome (if a little more unstable than usual i've found).

I did give the version numbers for all the browsers in the part you've even quoted:

Opera 9.60

I'm well aware of the issues with testing a release version versus a nightly (which is roughly identical to a beta), but if I couldn't find anything newer in a few minutes of Googling and looking round the Opera site there's not much I can do...

Nice to see Chrome kicked Fx up the arse into doing something though, Opera's always had an easy time with the "competition".

Firefox 3 was really finished (as far as anything that was going to significantly impact JS perf) at the start of the year (Profile Guided Optimisation being switched on for official builds in February was the last thing I think), and it's still (~1.3x) faster than Opera 9.6, not to mention that TraceMonkey (in some form) was on the cards for a while before Chrome appeared.
Maybe V8/TraceMonkey/SquirrelFish Extreme will kick Opera, or more importantly IE, up the arse to do something...

Edit: While I'm at it, here's the same graphs for Mozilla's Dromaeo JS benchmark. Same drill, average of 5 runs and the same browser versions as last time.

There's a degree of irony in WebKit beating Firefox at Mozilla's own test

 

[zzzJonny]

I've been looking around and the speed increases sound good. Is the 3.1 beta stable?

 

[mrk]

I've been looking around and the speed increases sound good. Is the 3.1 beta stable?

I'm using it 24/7 at home and also now all day at work and I don't tend to use stuff that crashes even every now and then

 

[LeperousDust]

I did give the version numbers for all the browsers in the part you've even quoted:

Sorry, yeah i totally missed that because it was labeled differently on the graph. It does seem they're doing a fantastic job, and its slightly annoying because i always liked running Opera when everyone was raving about Fx and it was the underdog If Fx is stable and released as quick as this i don't see any reason why i shouldn't give it another real go. &

 

[tolien]

I guess I could've been clearer - I missed version numbers off the graph because I was trying to keep the legend as small as possible.

The interesting bit IMO will be what happens next - there's talk of another minor release, and doing small-ticket items on a short cycle until Moz2/Fx4. Until IE gets its thumb from its arse (IE8 beta 2 is ~6x slower than Fx3.1), it's virtually irrelevant though

 

[willhub]

Whats that TM and how do you get it? Also I tried adding glasser 1.0.16 and it said it was not supported

 

[tolien]

You'll probably have to override compatibility on Glasser - either install Nightly Tester Tools/Mr Tech's Local Install, or add extensions.checkCompatibility and set it to false (with the proviso that it might burn the house down/cause a plague of locusts/eat children etc). As for TM:

about:config -> javascript.options.jit.content to turn on the more stable of the TraceMonkey options too.

 

[zzzJonny]

I just installed it and I'm noticing a slight speed increase too.

 

[mrk]

I installed the latest Minefield release today and all my extensions once forced using Nightly tester Tools (inc Adblock Plus) work fine. Also get 92/100 in Acid 3 now and it seems to be faster than 3.1 BETA as well!

 

[tolien]

Adblock Plus is probably stopping you making 93.

 

[mrk]

Can't get rid of that

 

[tolien]

So disable the tabs on elements (which IIRC was identified as one of the causes) or whitelist the Acid3 site?

Ctrl-F5 might fix it too - there are GC issues which occasionally break the test. Patches welcome