Announcement

Collapse
No announcement yet.

Fedora 40 Looks To Provide Optimized x86_64 Binaries For Different HWCAPs

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

  • Fedora 40 Looks To Provide Optimized x86_64 Binaries For Different HWCAPs

    Phoronix: Fedora 40 Looks To Provide Optimized x86_64 Binaries For Different HWCAPs

    In addition to Ubuntu exploring the possibility of x86_64-v3 builds/packages, a proposal has been raised for Fedora Linux with its current Fedora 40 cycle to provide the ability to offer optimized x86_64 (AMD64) binaries based upon the CPU's x86_64 micro-architecture feature level...

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    Good. Practically every benchmark I've seen shows that using better than V1 builds usually give a free 5-10% performance improvement across the board with some things getting massive 50-100% improvements. That's regarding open source software that can be recompiled.

    For gamers and proprietary software users, optimized OS builds usually don't seem to help as much in regards to increased performance in the double digit percentage range unless they rely on system libraries and not self-shipped libraries. Anecdotally, I've noticed that I tend to get better minimums when I optimize my gaming stack; better lows and averages. Depending on how you look at it, that's actually better than getting higher maximums since it can smooth out areas where the FPS tanks.

    Granted, I haven't been keeping up with how much of an improvement that various tweaks make since moving to a combination of Zen 4 and RDNA 2. Practically every game I have hits 1440p60 with High/Ultra settings with my hardware so I don't have the need to chase the optimization dragon that much. Chasing that dragon is what CachyOS and other distributions that offer optimized builds are for.

    Comment


    • #3
      CachyOS compiles almost all packages with optimized (v2,v3,v4) compiler flags. I noticed distributions like OpenSuSE, which rely on HWCAPS, don't ship optimized builds of every package and only a few are actually tuned.

      Really wish CachyOS would team up with Artix (dinit) to provide a optimized SystemD-free distro..
      Last edited by Kjell; 30 December 2023, 11:45 AM.

      Comment


      • #4
        Also for Silverblue?

        Comment


        • #5
          Originally posted by Kjell View Post
          Really wish CachyOS would team up with Artix (dinit) to provide a optimized SystemD-free distro..
          What's the benefit of not shipping systemd? Better incompatibility?

          Comment


          • #6
            Originally posted by skeevy420 View Post
            Good. Practically every benchmark I've seen shows that using better than V1 builds usually give a free 5-10% performance improvement across the board with some things getting massive 50-100% improvements.
            I hope they upgrade the minimum requirements.
            I'm seeing this problem breaching from a different angle as well. Today's compilers can produce advanced languages like Zig (not to bring it into discussion) which ship with SIMD vectors built into the language with the builtin function @Vector() (A vector is a group of booleans, Integers, Floats, or Pointers which are operated on in parallel, using SIMD instructions if possible).

            Comment


            • #7
              Originally posted by eagleoneraptor View Post
              What's the benefit of not shipping systemd?
              It's like a malignant tumor, slowly encompassing all of Linux' userspace services and infrastructure. Its only goal seems to be alleviating support burden for RHEL and their customers, but what if you have a use case which doesn't fit their worldview?

              Another potential issue is the interdependence between services forcing all-or-nothing upgrades, which could leave people in the unenviable position of having to decide between bug fixes vs. security fixes, for instance.

              I don't hate dealing with systemd, but I'll bet many subjects under past monarchies didn't necessarily hate their day-to-day life, either. That doesn't mean they loved the monarchy, or that their lives couldn't have been even better under a different system of government.
              Last edited by coder; 30 December 2023, 12:35 PM.

              Comment


              • #8
                Originally posted by coder View Post
                It's like a malignant tumor, slowly encompassing all of Linux' userspace services and infrastructure. Its only goal seems to be alleviating support burden for RHEL and their customers, but what if you have a use case which doesn't fit their worldview?

                Another potential issue is the interdependence between services forcing all-or-nothing upgrades, which could leave people in the unenviable position of having to decide between bug fixes vs. security fixes, for instance.

                I don't hate dealing with systemd, but I'll bet many subjects under past monarchies didn't necessarily hate their day-to-day life, either. That doesn't mean they loved the monarchy, or that their lives couldn't have been even better under a different system of government.
                Are you suggesting that monarchies and init systems can be compared? Or that standardization of distros hasn't categorically been a positive thing, with widespread compatibility and services that actually start up and shut down on the desktop? Or that systemd isn't an umbrella project with many different subprojects, e.g. GNU?

                Comment


                • #9
                  Originally posted by Kjell View Post
                  CachyOS compiles almost all packages with optimized (v2,v3,v4) compiler flags. I noticed distributions like OpenSuSE, which rely on HWCAPS, don't ship optimized builds of every package and only a few are actually tuned.

                  Really wish CachyOS would team up with Artix (dinit) to provide a optimized SystemD-free distro..
                  The biggest problem with CachyOS is that not everything is done with 2/3/4 so you have to also add the Arch repos and then layer them in the correct order. IMHO, CachyOS needs to unshackle itself from Arch.​

                  eagleoneraptor

                  Systemd has a lot of bells and whistles for something that starts and stops software. It does lot of stuff behind the scenes in ways that a lot of old Linux and Unix users just don't like. IMHO, a big problem that systemd has is that it's the name of the overall project of related utilities and also what people say when they really mean the specific systemd-init part of the project that initializes the system like SysV, OpenRC, or Dinit.

                  While systemd-init can be used without any of the rest of systemd, that's practically unheard of in actual practice since all the systemd software is designed to be used with one another and that cooperation and synergy is what takes it from being a simple init that you manage yourself to a comprehensive init system that can handle service dependencies so it doesn't have to rely on a sys admin to script everything the system requires in the correct order. That can induce a lot of bloat and background tasks running if you don't know how to prune it down....kind of like Windows or macOS....

                  Comment


                  • #10
                    Originally posted by AlanTuring69 View Post
                    Are you suggesting that monarchies and init systems can be compared?
                    Oh, it's a lot more than an init system. That was just the camel's nose under the tent.

                    Originally posted by AlanTuring69 View Post
                    Or that standardization of distros hasn't categorically been a positive thing
                    There are different ways to do standardization. You could convene a consortium, like POSIX (or maybe create a Working Group under the Linux Foundation), and define a set of protocols and standards. Or, you can go the monolithic route and just give people an all-or-nothing choice. Redhat chose the latter.

                    As far as standards go, it's no more a standard than Windows.

                    Comment

                    Working...
                    X