Announcement

Collapse
No announcement yet.

Linux 4.1 Offers Potentially Dazzling Performance

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

  • #11
    I really find the exhilaration quite amusing. Do you really have any idea how unimpressive these benchmark improvements actually are for day-to-day operations via the Kernel?

    Comment


    • #12
      Marc, I think you're reading some other benchmarks. These results are _astounding_, not _unimpressive_. We're looking at results from 5% to 100% faster. And for day-to-day operations via the kernel, ya that's where the 100% improvement came from. That literally is the difference between, am I going to sleep early tonight or do I need to stay up late to baby sit my build.

      This is coming from my perspective of being a kernel maintainer. When you're building for multiple architectures, every second you save is another second you get to do something more important than sit around and read news.

      Comment


      • #13
        Originally posted by damentz View Post
        Marc, I think you're reading some other benchmarks. These results are _astounding_, not _unimpressive_. We're looking at results from 5% to 100% faster. And for day-to-day operations via the kernel, ya that's where the 100% improvement came from. That literally is the difference between, am I going to sleep early tonight or do I need to stay up late to baby sit my build.

        This is coming from my perspective of being a kernel maintainer. When you're building for multiple architectures, every second you save is another second you get to do something more important than sit around and read news.
        Ah I see you've met our main resident apple fanboy, he falls firmly into the Haters gonna Hate box and I wouldn't pay him much mind. He occasionally says something useful but if he's hating on or downplaying something it's usually not worth it to pay him mind.

        Comment


        • #14
          Originally posted by Luke_Wolf View Post
          Haters gonna Hate
          There is not just "Haters gonna hate" clause but "astounders gonna astounding" too . Article like to be belivable should include as much info as possibile, yes even bisect on what exactly made this difference. Otherwise it is just newer kernel is "better", for readers who pretend to blindely believe at anything seems too...

          And I can sign this with my own head, that Michael often has these tramendous halfed compiling kernels improvements but in reallity nothing really there happens

          That said if someone have performance regression and fix regression that is nothing astounding, but just fine
          Last edited by dungeon; 07 June 2015, 08:16 AM.

          Comment


          • #15
            Originally posted by dungeon View Post

            There is not just "Haters gonna hate" clause but "astounders gonna astounding" too . Article like to be belivable should include as much info as possibile, yes even bisect on what exactly made this difference. Otherwise it is just newer kernel is "better", for readers who pretend to blindely believe at anything seems too...

            And I can sign this with my own head, that Michael often has these tramendous halfed compiling kernels improvements but in reallity nothing really there happens

            That said if someone have performance regression and fix regression that is nothing astounding, but just fine
            Bisecting over performance improvements isn't reasonable for a reviewer and you know it. Now the kernel team on the other hand absolutely should be doing that as part of their Continuous Integration system, as the developers themselves should be tracking the performance (among other factors) in order to prevent regressions. Something like so http://arewefastyet.com/ but for the kernel, by the kernel team.

            Also please do point to some of these other times that Michael has halved compilation times, since you think they happen so often, and if you don't believe Michael's numbers well... why are you here in this thread? As if you consider Michael's benchmarking numbers to be untrustworthy then there was no point in you reading the article and thus there's no point in you being in this thread to discuss his untrustworthy numbers. If you want to argue for a different configuration that's one thing, but you seem to be arguing that Michael is lying.

            Comment


            • #16
              Also please do point to some of these other times that Michael has halved compilation times, since you think they happen so often, and if you don't believe Michael's numbers well... why are you here in this thread?
              He, he, i don't said i don't believe Micheal ... i just said that halfed kernel compilation time happens in some of his benchmarks for the reason XYZ aka yet unknown

              This time it happened on Intel hardware, and i clearly remember AMD AM1 launch time last year and his reviews about these procesors becuase i had it too at that time to compare... His kernel compilation time was halfed or even using one core or something like that for him, but never ever for me. And i also tried to reproduce it using Ubuntu images, with same software but it was not reproducable for me, etc...

              And i also remember after few weeks i think his benchmarks sudenly go right on par where it should be, but there was not explination what happened there with his benchmark at first

              Article for you, look at numbers and read a comments:

              http://www.phoronix.com/scan.php?pag...0_ubuntu&num=1
              Last edited by dungeon; 07 June 2015, 10:32 AM.

              Comment


              • #17
                Originally posted by Luke_Wolf View Post
                Bisecting over performance improvements isn't reasonable for a reviewer and you know it. Now the kernel team on the other hand absolutely should be doing that as part of their Continuous Integration system, as the developers themselves should be tracking the performance (among other factors) in order to prevent regressions. Something like so http://arewefastyet.com/ but for the kernel, by the kernel team.
                Well there are daily kernel benchmarks now at http://linuxbenchmarking.com/

                In terms of bisecting this performance regression, PTS can automatically bisect it if I had the time on the machine to do so.... But given I'm taxed already with having to use that machine for other tests and am busy overall, it's not worth it at this time when it often goes unappreciated and hardly been any tips lately, etc.
                Michael Larabel
                https://www.michaellarabel.com/

                Comment


                • #18
                  Hmm sounds familiar... Could be a kernel p-state bug that got fixed, possibly related to certain certain BIOS/kernel combinations. This is an 5960x Engineering Sample chip? Could be an issue.

                  Comment


                  • #19
                    Originally posted by chrisb View Post
                    Only thing i see there that latest kernel at the time was good again

                    That was 3.18 in that article, now in this article 3.18 is bad again, how that can be?... and newest one the best.
                    Last edited by dungeon; 07 June 2015, 07:39 PM.

                    Comment


                    • #20
                      What about that large batch of asm commits you mentioned a few weeks back? Could that be the source of new found speed?

                      Comment

                      Working...
                      X