Announcement

Collapse
No announcement yet.

Firefox 76 + 77 Beta Web Browser Benchmarks On Ubuntu Linux

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #11
    Originally posted by birdie View Post

    A compiler can increase performance only so far. Refactoring, using assembler (SSE/AVX), using new approaches/frameworks, manual optimizations - that's what really affects performance.
    Oh yes, of course. I was thinking in terms of aggregating all those components. The best compiler on earth can't turn not so great code into gold. =) Everything you say is true re: optimization.

    Still, for Linux, Rustification is helping on that front and the newer Rust has some benefits there [1].

    https://blog.mozilla.org/nnethercote/
    Last edited by kozman; 11 May 2020, 01:26 PM.

    Comment


    • #12
      Originally posted by AJenbo View Post

      Hasn't LLVM 10 actually been showing degradations in performance compared to LLVM 9?
      Yes, on a couple things if I remember too. Not super major though. Maybe that perf gap will be fixed in 10.0.1?

      Comment


      • #13
        Originally posted by rene View Post
        I usually use Firefox on my Ryzen 3950x workstation at work (https://www.youtube.com/watch?v=NUq39Jz5ZJI) but just last night I plugged that SSD into my Intel Core thin and light from 5 years ago at home and man was that lagging. Can't even browse the web with Firefox on such aging low-power silicon :-/
        Working just fine on my 3770.

        Comment


        • #14
          Originally posted by kozman View Post

          Yes, on a couple things if I remember too. Not super major though. Maybe that perf gap will be fixed in 10.0.1?
          I wouldn't hold my breath. LLVM seems to mostly not care about Rust. If the regressions don't show for C/C++ then it probably won't be touched.

          Comment


          • #15
            Originally posted by kozman View Post

            Oh yes, of course. I was thinking in terms of aggregating all those components. The best compiler on earth can't turn not so great code into gold. =) Everything you say is true re: optimization.

            Still, for Linux, Rustification is helping on that front and the newer Rust has some benefits there [1].

            https://blog.mozilla.org/nnethercote/
            If I've understood correctly, the article is referring to performance improvements of the rust compiler itself, not generated binary performance, is that correct?

            Comment


            • #16
              Originally posted by TemplarGR View Post

              IIRC webrender compositor does not work on Linux at the moment, it doesn't do anything if you enable it.
              Works fine on my laptop, has worked for years (well, it was a little buggy in the early days, but is quite stable nowadays). Just make sure `gfx.webender.all` is set to true. Totally worth enabling, much better performance.

              Comment


              • #17
                Originally posted by QwertyChouskie View Post

                Works fine on my laptop, has worked for years (well, it was a little buggy in the early days, but is quite stable nowadays). Just make sure `gfx.webender.all` is set to true. Totally worth enabling, much better performance.
                I think he was not talking about Webrender in general but Webrender compositor integration (`gfx.webrender.compositor`). That's the next step, necessary to become as/more power efficient as the pre-WR renderer (on Win/MacOS - on Linux there was no real compositor integration so far*). On Linux this will only be possible on Wayland, as X11 lacks certain features for it (for real this time, as opposed to e.g. video acceleration which would be possible on X11).

                See https://bugzilla.mozilla.org/show_bug.cgi?id=1617498 (tracking bug) and https://mozillagfx.wordpress.com/201...ore-animation/ for what compositor integration meant for the OpenGL renderer on MacOS (similar changes for WR currently get rolled out on Windows, MacOS is also close AFAIK).

                *: there is partial damage and opaque region support, both on X11 and Wayland, although the former is currently limited to the software renderer (see https://bugzilla.mozilla.org/show_bug.cgi?id=1620076 for WR support) - proper compositor integration would e.g. make use of subsurfaces, viewports and similar advanced stuff.
                Last edited by treba; 11 May 2020, 09:28 PM.

                Comment


                • #18
                  Originally posted by Mez' View Post
                  My laptop is 8 years old and comes with an AMD A4-3300M, 8 Gb RAM and I upgraded the main storage to a lacklustre Sandisk Ultra 256Gb SSD. I currently have 40+29 (I let you do the math ) tabs open and it works very well considering. Loading page is neither fast nor slow. But certainly not lagging.

                  Don't know how a Ryzen 3950x could lag, seems pretty absurd to me.
                  It would be, but that's not what he said. He said a low powered 5 year old machine was slow in comparison to the 3950 box.

                  Comment


                  • #19
                    Originally posted by timrichardson View Post

                    It would be, but that's not what he said. He said a low powered 5 year old machine was slow in comparison to the 3950 box.
                    Right. Indeed.
                    I got confused by the SSD. I don't come back from work with a SSD in my pocket every day...

                    In any case, I'm sure his Intel core thin and light from 5 years ago relatively destroys my A4-3300m low-budget Llano APU from 8 years ago. He should still be fine overall.
                    Last edited by Mez'; 11 May 2020, 10:20 PM.

                    Comment


                    • #20
                      Trying to summarize these findings, to remember this better. Overall no significant differences detected. Real world performance may show differences. All tests here were with Wayland, not the current default: X11 displays.
                      All tests were done without having a "standard working load" of tabs and extensions. Most users will run perhaps a few tabs, and a few extensions. Fr any worksation then that deviates from the test conditions here, the tests might tell a very different story.

                      Comment

                      Working...
                      X