Announcement

Collapse
No announcement yet.

Updated Fedora 20 KDE Stack: Something's Funky

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

  • Updated Fedora 20 KDE Stack: Something's Funky

    Phoronix: Updated Fedora 20 KDE Stack: Something's Funky

    One month ago we looked at the latest performance of Fedora 20, but with its more liberal update strategy -- especially with the long release time until Fedora 21 -- we're back with some more tests today as since last month the Linux 3.14 kernel has been added and other changes. This article has benchmarks of the Fedora 20 KDE spin out-of-the-box and then with all available updates as of this week to see how the performance has evolved in the half-year since the F20 release.

    http://www.phoronix.com/vr.php?view=20454

  • #2
    I'm just getting frustrated beyond belief with this Intel power governor issue crap.

    Here's a tip for their engineers: pick up the phone, call the Windows division, and ask how they do it.

    Intel definitely comes across as one of those companies where the left hand doesn't have a clue what the right hand is doing.

    Comment


    • #3
      Am I going to buy an Android phone or tablet with an Intel processor if they can't demonstrate competence in handling frequency stepping?

      Comment


      • #4
        Originally posted by johnc View Post
        Am I going to buy an Android phone or tablet with an Intel processor if they can't demonstrate competence in handling frequency stepping?
        To be fair to Intel... Michael's comments AFAIK there's no grounded evidence that it IS the P-State driver having problems. When I read the article anytime i saw a mention of "Well it could be P-State" I took it with a grain of salt and assumed the comment was being made because the last time there were strange performance decreases (and increases) it did infact come down to CPU Frequency stuff.

        note: I'm referring to the Radeon benchmarks on Ubuntu from a few months ago where a change to the ondemand (old style driver) governor was upping performance by like 10%. If there's been new changes / articles since then that showcase a P-State driver problem then I've apparently missed those articles.

        Comment


        • #5
          Originally posted by Ericg View Post
          To be fair to Intel... Michael's comments AFAIK there's no grounded evidence that it IS the P-State driver having problems. When I read the article anytime i saw a mention of "Well it could be P-State" I took it with a grain of salt and assumed the comment was being made because the last time there were strange performance decreases (and increases) it did infact come down to CPU Frequency stuff.

          note: I'm referring to the Radeon benchmarks on Ubuntu from a few months ago where a change to the ondemand (old style driver) governor was upping performance by like 10%. If there's been new changes / articles since then that showcase a P-State driver problem then I've apparently missed those articles.
          It seems to be a well-known, established and ongoing problem. Not just with pstate but even the older cpufreq governor as you mentioned.

          Comment


          • #6
            Could the (somewhat) poor performance we are seeing in the OpenGL benchmarks be explained by (the lack of) async DMA? It was disabled in mesa 10.1.2 for R700 cards [1].

            [1] http://cgit.freedesktop.org/mesa/mes...ba2f0f90ea5aba

            Comment


            • #7
              hyperz?

              Are those games CPU limited? Couldn't it be a combination of pstate and hyperz? Afaik it was turned off by default with mesa 10.1.

              Comment


              • #8
                Originally posted by tomtomme View Post
                Are those games CPU limited? Couldn't it be a combination of pstate and hyperz? Afaik it was turned off by default with mesa 10.1.
                Could be hyperz, could also be MSAA

                Comment


                • #9
                  Originally posted by tomtomme View Post
                  Are those games CPU limited? Couldn't it be a combination of pstate and hyperz? Afaik it was turned off by default with mesa 10.1.
                  openarena probably is
                  xonotic is, but less so on high

                  gputest triangle is pure buffer flipping while fur shader is (probably) pure gpu molesting
                  x264 is pure cpu thing, so..

                  Comment


                  • #10
                    On Arch with kernel 3.14,

                    Using 'performance' over the default 'powersave' (intel P-state) gives a 25/40 % perf boost to Openarena/Xonotic(low) on my radeon card.

                    So I wouldn't be suprised intel P-state is to blame on Michael's benchmarks.

                    Comment

                    Working...
                    X