Announcement

Collapse
No announcement yet.

The Disappointing Direction Of Linux Performance From 4.16 To 5.4 Kernels

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

  • #71
    Yes, I'd love to know how much of this performance was lost from bugs and how much from mitigations. The only way to know for sure is if Michael runs the same benchmark with
    Code:
    mitigations=off
    .

    Comment


    • #72
      Originally posted by road hazard View Post
      nor will I ever buy another Nvidia GPU (because of their hatred for Linux).
      Nvidia has supported Linux for far longer than AMD has. I was running my MX 200 accelerated in Linux back in 1997. Rocking GL Quake! AMD is the cheap seats. Intel invests more in Linux every year than AMD is worth. Intel and Nvidia 4 life!

      Comment


      • #73
        Originally posted by set135 View Post

        the thing sucks over 3amps@125volts idle.
        In the future please express power in Watts units. 3 Amps @ 125 V = 375 Watts. Actually not entirely accurate on AC unless your power factor is 1, but close enough for government work.

        Comment


        • #74
          Originally posted by Paul Frederick View Post

          Nvidia has supported Linux for far longer than AMD has. I was running my MX 200 accelerated in Linux back in 1997. Rocking GL Quake! AMD is the cheap seats. Intel invests more in Linux every year than AMD is worth. Intel and Nvidia 4 life!
          AMD provide amazing Open Source drivers while nvidia does nothing except crappy blobs which don't even run on Wayland afaik. Intel stole a lot of money from people. I wonder how much their bugs cost?

          Comment


          • #75
            Originally posted by Paul Frederick View Post

            Nvidia has supported Linux for far longer than AMD has. I was running my MX 200 accelerated in Linux back in 1997. Rocking GL Quake! AMD is the cheap seats. Intel invests more in Linux every year than AMD is worth. Intel and Nvidia 4 life!
            You were running a Geforce 2MX in Linux in 1997? That seems a bit unlikely.

            Comment


            • #76
              Originally posted by geearf View Post
              You were running a Geforce 2MX in Linux in 1997? That seems a bit unlikely.
              Well it was over 20 years ago now. So I guess I remembered wrong? It was a long time ago though. I ran Nvidia's binary driver when it was in beta. Back when AMD support was non-existent. It was the only accelerated 3D on Linux. Nvidia is still the only accelerated 3D on Linux for me. AMD hardware dissipates too much power. Plus I've seen AMD support Linux then go back on it. So I don't trust them today. Nvidia has never changed their support policy. Nvidia is number 1. I am looking forward to see what Intel puts out though.

              Comment


              • #77
                Originally posted by Danny3 View Post
                I hate when developers care only about security and do the changes no matter the costs.
                I bet none of them used Phoronix Test Suite or other benchmarking tool before pushing these changes into the Linux kernel.
                This is really disappointing!
                I hope this doesn't affect HPC systems, cause they use customized kernels

                Comment


                • #78
                  Originally posted by Volta View Post
                  AMD provide amazing Open Source drivers while nvidia does nothing except crappy blobs which don't even run on Wayland afaik.
                  Not only drivers, but also a lot of tools (rocm, gpuanalizer, CodeXl, HIP, etc) are open and are good

                  Comment


                  • #79
                    Originally posted by bug77 View Post
                    I suspect we're going to need serious silicon redesign to get speculative execution back into safe territory without performance penalties.
                    I believe it is impossible to achieve secure speculative execution and no performance penalties at the same time, because these two terms are mutually contradictory in their nature. Implementing the security measures in silicon has performance implications as well. The size of the window of instructions that the silicon (CPU) can see is currently fairly small (the CPU is unable to determine static invariants which hold for larger pieces of code), which means that the security measures have to be repeatedly/redundantly evaluated for every speculatively executed instruction irrespective of whether the instruction has invariant properties that could be optimized away.

                    Comment


                    • #80
                      Ext4 on my Box is completely borked with 5.4! RootFs gets damaged while trying to mount:
                      Code:
                      EXT4-fs error (device nvme0n1p5): ext4_journal_check_start:61: Detected aborted journal
                      EXT4-fs (nvme0n1p5): Remounting filesystem read-only
                      EXT4-fs (nvme0n1p5): ext4_writepages: jbd2_start: 8191 pages, ino 262253; err -30

                      Comment

                      Working...
                      X