Firefox 68 Performance Is Looking Good With WebRender On Linux

[Michael_S]

Chrome is still #1, another fail for rust

Firefox was 20% the speed of Rust or worse a few years ago, well before Rust was implemented. The problem isn't the language, it's just less efficient code.

[Rallos Zek]

Chrome is still #1, another fail for rust

Too bad Chrome uses 4 times the ram Firefox does.

[Sonadow]

good luck trying to compile it from source (hint needs newer Rust): https://www.youtube.com/watch?v=3MdTu8dSorY

That's an inconvenience that is so common it's not even worth mentioning anymore.

I had to compile the latest stable version of rust yesterday (v1.34.2) as the last version I built in January (v1.31) was too old for Firefox 67. And today Rust v1.35 just got released. By my estimates, a stable Rust release is only good for building two or three major Firefox releases before it needs to be replaced.

[re:fi.64]

Curious to see how it now compares to Chrome.

[HadrienG]

That's an inconvenience that is so common it's not even worth mentioning anymore.

I had to compile the latest stable version of rust yesterday (v1.34.2) as the last version I built in January (v1.31) was too old for Firefox 67. And today Rust v1.35 just got released. By my estimates, a stable Rust release is only good for building two or three major Firefox releases before it needs to be replaced.

I would suggest giving a chance to rust's official binary distribution channel, rustup.

I was initially skeptical too. Would rather stick with as few trusted software distribution channels as possible if I had the choice. But the big Linux distributions have long failed us at the task of distributing up-to-date software (with the notable exception of rolling releases, which tend to have QA issues instead), and with modern devops practices enabling much faster software release cycles without quality loss, the cracks in their software distribution model are seriously starting to show.

Maybe, one of these days, major players like Ubuntu or RedHat will get the Tumbleweed memo and abandon the fundamentally broken 6-month stable release paradigm in favor of taking rolling release QA seriously. Until then, for rust users, rustup is the most solid path to sanity available.

[gedgon]

Curious to see how it now compares to Chrome.

MotionMark, i5 2500, RX 460, 4k, (firefox 68 on the left, chromium 74 on the right)

[Weasel]

Firefox was 20% the speed of Rust or worse a few years ago, well before Rust was implemented. The problem isn't the language, it's just less efficient code.

Yeah, but people who use Rust have differently-wired brain. Only such brain would use Rust. And they tend to write this kind of inefficient code. And well... forced bounds checking and fat pointers...

[wpupkin]

First message was a dumb joke, but ... look, already reserved:

PID     VSZ     COMMAND
6835    1.53G   /opt/firefox/firefox-bin
14156   2.52G   /opt/firefox/firefox
14218   1.53G   /opt/firefox/firefox-bin
14265   20.44G  /opt/firefox/firefox-bin
26884   1.37G   /opt/firefox/firefox-bin

<< dumb joke #2 END

For migrated default profile, not reproducible on new profile.

[Marc.2377]

Firefox was 20% the speed of Rust or worse a few years ago, well before Rust was implemented. The problem isn't the language, it's just less efficient code.

and the language.

[oleid]

Yeah, but people who use Rust have differently-wired brain. Only such brain would use Rust. And they tend to write this kind of inefficient code. And well... forced bounds checking and fat pointers...

Haha, our favorite C troll again. Who doesn't know what it's talking about.

An advanced rust developer knows how to avoid safely avoid bound checks. And hey, bound checks are good, by the way. At least when doing safety critical stuff. Ada uses them as well, and that's what used to drive the space shuttle and other things which may not fail.

And fat pointers? Nobody forces you to use Rc or Arc. Simply use references. Or plain C pointers, if you have to.

But the fat pointers, like they are implemented in rust make more sense than their c++ counterparts. As the refcount is on the same cache line.