Announcement

Collapse
No announcement yet.

The Fastest Linux OS For AMD Ryzen Zen 3? It's Still Intel Clear Linux

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

  • #11
    Originally posted by duby229 View Post

    Clear linux does a whole lot more than just compiler flags tweaks. They make a bunch of modifications. It would be impossible to duplicate all that Clear linux does because quite a lot of it is (or at least was) hardcoded.

    EDIT: I haven't followed it in quite some time though, so I don't know if they ever got to distributing patches for all the modifications they make? They didn't used to. Many of the modifications to non-GPL projects they just simply didn't make patches available. Like I said I don't know if they ever corrected that.

    EDIT: Which is one of the reasons I think the GPL is a superior license. All of their kernel modifications are available and always have been because the GPL requires it. But to duplicate Clear linux in full at least used to be impossible.
    I did not know all of that...but it would still be interesting if AMD developed their own version. It would be even better if they did that and up streamed their modifications.
    GOD is REAL unless declared as an INTEGER.

    Comment


    • #12
      Originally posted by duby229 View Post
      EDIT: I haven't followed it in quite some time though, so I don't know if they ever got to distributing patches for all the modifications they make? They didn't used to. Many of the modifications to non-GPL projects they just simply didn't make patches available. Like I said I don't know if they ever corrected that.
      Their patches are made available on this github page: https://github.com/clearlinux-pkgs/linux/

      I personally recommend using this actively maintained repository: https://github.com/sirlucjan/kernel-...ux-patches-sep

      In terms of replicating/duplicating ClearLinux, in my experience I've found LTO greatly increases compiling time making it a dealbreaker for me. And PGO seems like a lot of work unless you have a really good use case for it.

      Other than that, I tweak away. -O3 -march=native, compiler flags, kernel config tweaks, performance governor, BFQ, something from xanmod, something from ClearLinux, out-of-tree patches like Valve's fsync, futex2, and tons others. In my opinion, it makes a huge difference but obviously only if you're an enthusiast. Nice thing about ClearLinux is that it makes a lot of that available out the box (and more) as a distro. I wouldn't use CL as a daily driver because I prefer Debian/Ubuntu but I'm a sucker for making the system as fast it can be.

      What I don't do anymore is mess with heavily experimental out-of-tree stuff like schedulers (CFS by Ingo Molnar is still the king imo) or things like uksm which manipulate the memory and possibly introduce instability.

      Comment


      • #13
        Originally posted by perpetually high View Post

        Their patches are made available on this github page: https://github.com/clearlinux-pkgs/linux/

        I personally recommend using this actively maintained repository: https://github.com/sirlucjan/kernel-...ux-patches-sep

        In terms of replicating/duplicating ClearLinux, in my experience I've found LTO greatly increases compiling time making it a dealbreaker for me. And PGO seems like a lot of work unless you have a really good use case for it.

        Other than that, I tweak away. -O3 -march=native, compiler flags, kernel config tweaks, performance governor, BFQ, something from xanmod, something from ClearLinux, out-of-tree patches like Valve's fsync, futex2, and tons others. In my opinion, it makes a huge difference but obviously only if you're an enthusiast. Nice thing about ClearLinux is that it makes a lot of that available out the box (and more) as a distro. I wouldn't use CL as a daily driver because I prefer Debian/Ubuntu but I'm a sucker for making the system as fast it can be.

        What I don't do anymore is mess with heavily experimental out-of-tree stuff like schedulers (CFS by Ingo Molnar is still the king imo) or things like uksm which manipulate the memory and possibly introduce instability.
        Yeah, I was aware of that, but those are just their kernel patches. Just briefly checking out their repository , I do see a bunch of patches they made for other various packages. So it looks like my old impression is no longer true. So maybe it is possible to duplicate large portions of Clear Linux now?

        Comment


        • #14
          A bit off topic but will Valve eventually merge fsync and futex2 upstream?

          Comment


          • #15
            It's astounding how much performance current software is leaving on the table. I really think something needs to give on how old of hardware modern linux distributions and software try to support. We have people complaining over and over that there is no way x86 can compete with ARM when we are severely handicapping modern software for the sake of keeping decrepit hardware alive. We have an optimized distro that's offering a ~15% average performance increased compared to nearly ever other distro (and some specific applications being 100% faster!). We have compiler flags like -znver3 or -skylake that offer great performance gains. We even have individual compilers like the AOCC or even just llvm/clang that offer really tangible performance increases over GCC

            All-in, current Linux (and probably Windows, too) software is leaving ~25-30% extra performance on the table just so we can still install Ubuntu on Grandma's Core 2 Solo from 2006. It's a 14 year old computer. Let it die. That'd be like trying to have an OS still support an i286 in 2000. Just insanity.
            Last edited by AmericanLocomotive; 22 December 2020, 07:30 PM.

            Comment


            • #16
              Originally posted by AmericanLocomotive View Post
              It's astounding how much performance current software is leaving on the table. I really think something needs to give on how old of hardware modern linux distributions and software try to support. We have people complaining over and over that there is no way x86 can compete with ARM when we are severely handicapping modern software for the sake of keeping decrepit hardware alive. We have an optimized distro that's offering a ~15% average performance increased compared to nearly ever other distro (and some specific applications being 100% faster!). We have compiler flags like -znver3 or -skylake that offer great performance gains. We even have individual compilers like the AOCC or even just llvm/clang that offer really tangible performance increases over GCC

              All-in, current Linux (and probably Windows, too) software is leaving ~25-30% extra performance on the table just so we can still install Ubuntu on Grandma's Core 2 Solo from 2006. It's a 14 year old computer. Let it die. That'd be like trying to have an OS still support an i286 in 2000. Just insanity.
              I think it's more complicated than that.
              Old computer have access to the internet and this means that they should run updated software, otherwise we're just preparing botnet material.
              Compiling very optimized software means a lot more effort in packaging and distribution.
              Nonetheless Clear Linux is a very valuable piece of software

              Comment


              • #17
                Originally posted by vladpetric View Post
                Wow, this is amazing, thank you Michael!

                A few months ago someone told me about issues when attempting qemu/KVM on top of Clear Linux / Ryzen (my current setup is a clear linux/kvm/haswell xeon).

                Any more recent thoughts?
                I was the one who told you.

                I don't have a Zen3 to test it with (yet) as I am testing the 6W Zen+ Dali (AMD Athlon Silver 3050e) against the RaspPi 4 for IoT work at the moment.

                https://openbenchmarking.org/result/...FI-2012165FI18

                I am waiting for the 5x00-G release of Zen3.

                Comment


                • #18
                  Originally posted by edwaleni View Post

                  I was the one who told you.

                  I don't have a Zen3 to test it with (yet) as I am testing the 6W Zen+ Dali (AMD Athlon Silver 3050e) against the RaspPi 4 for IoT work at the moment.

                  https://openbenchmarking.org/result/...FI-2012165FI18

                  I am waiting for the 5x00-G release of Zen3.
                  Thank you sir!

                  Comment


                  • #19
                    Originally posted by AmericanLocomotive View Post
                    It's astounding how much performance current software is leaving on the table. I really think something needs to give on how old of hardware modern linux distributions and software try to support. We have people complaining over and over that there is no way x86 can compete with ARM when we are severely handicapping modern software for the sake of keeping decrepit hardware alive. We have an optimized distro that's offering a ~15% average performance increased compared to nearly ever other distro (and some specific applications being 100% faster!). We have compiler flags like -znver3 or -skylake that offer great performance gains. We even have individual compilers like the AOCC or even just llvm/clang that offer really tangible performance increases over GCC

                    All-in, current Linux (and probably Windows, too) software is leaving ~25-30% extra performance on the table just so we can still install Ubuntu on Grandma's Core 2 Solo from 2006. It's a 14 year old computer. Let it die. That'd be like trying to have an OS still support an i286 in 2000. Just insanity.
                    There is no doubt things can always be further optimized and tuned. And a faster distro and using better default compiler options helps everything - including Arm. However it's pointless to exaggerate - you're never going to see a 15% speedup in the real world, let alone that 30% you literally made up.

                    What is actually astounding is how just one bad benchmark causes the geomean to change by 4.5%. Leave out simdjson and the speedup drops to 10%... So all you need to do is to add a few more bad benchmarks. Why stop at 15%? Let's go for a 50% speedup! And I bet the usual crowd will believe it.

                    As we say, lies, damned lies and benchmarks.

                    Comment


                    • #20
                      Is AMD just gonna take this lying down?

                      Comment

                      Working...
                      X