Announcement

Collapse
No announcement yet.

AMD Ryzen Threadripper 2920X & 2970WX Linux Performance Benchmarks

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

  • #11
    Originally posted by hoohoo View Post
    I remember building Linux kernel in circa 1993 on a 486. It took about two hours IIRC. Now it's much bigger and it takes 36 seconds.
    I remember those long waits also. Very frustrating times. Today there is little reason to recompile a kernel. Generally you can rely upon your distro shipping all reasonable system updates.
    Last edited by wizard69; 10-29-2018, 12:08 PM. Reason: Cell phone writing it’s own garbage.

    Comment


    • #12
      Originally posted by Michael View Post

      Whoops thanks, typos fixed.
      Two more but minor ones:

      Originally posted by phoronix View Post
      significantly slower than Intel CPus with this simple
      Originally posted by phoronix View Post
      the 1950X does deliver bettter performance

      Comment


      • #13
        Originally posted by tildearrow View Post

        Two more but minor ones:
        Thanks as always.
        Michael Larabel
        http://www.michaellarabel.com/

        Comment


        • #14
          no games

          Comment


          • #15
            Originally posted by wizard69 View Post

            I remember those long waits also. Very frustrating times. Today there is little reason to recompile a kernel. Generally you can rely upon your distro shipping all reasonable system updates.
            Generally true. Recently I was building the kernel to get the bleeding edge AMD ROCm drivers (can't remember correct driver name). Building on a R5 2400G, I'd run the make, go get a coffee, come back, it was nearly done. So the Linux kernel a few minutes on a 2400G, lol.

            Comment


            • #16
              Originally posted by pal666 View Post
              no games
              I think that on the 2970 you could play all your games simultaneously ;-)

              Comment


              • #17
                Originally posted by hoohoo View Post

                I think that on the 2970 you could play all your games simultaneously ;-)
                Thin clients gaming now at reasonable prices

                Comment


                • #18
                  Originally posted by wizard69 View Post
                  I remember those long waits also. Very frustrating times. Today there is little reason to recompile a kernel. Generally you can rely upon your distro shipping all reasonable system updates.
                  If you want optimal use of memory, it can help to run a stripped-down kernel. This matters in cases such as large numbers of files, where certain kernel flags can lead to significant variances in inode structure size (as does 32-bit vs. 64-bit), and hence the amount of RAM taken up by kernel slabs for the same data. Support for quota and the like is not free.

                  As for the article, I was intrigued by this claim:
                  Originally posted by phoronix
                  The Intel systems with Ubuntu 18.10 / Linux 4.19 were having a small advantage in raw performance with the PostgreSQL database server performance, should you intend to use a Core i9 or Threadripper as a web development workstation. But for real-world hosting, you're certainly better off with a Xeon or EPYC.
                  I can see this for Xeon, simply since it tends to be linked to ECC support, but w.r.t. EPYC it seems like the benefit is "merely" scaling ability (dual-processor, octa-channel, more PCIe lanes). It's true that Postgres scales vastly better than it used to, but still . . . for many cases I'd have thought an equivalent EPYC would perform roughly equivalently, and even a "lesser" CPU might win out (see pgbench line) depending on parallelization vs. single-threaded speed and whether it fits into cache. Assuming you can get or build a server with a Threadripper in it, of course.

                  [I could see EPYC 2 on 7nm leading to a significant improvement, but of course we don't have those yet.]
                  Last edited by GreenReaper; 10-30-2018, 03:12 AM.

                  Comment


                  • #19
                    Michael As an idea for a follow up article: It would be interesting to measure the performance of the IO vs. Compute Dies, ie dies with direct memory access and thos without (see Threadripper 2, Chiplets Aid Core Scaling). Especially for single (or few) threads workloads. Like it is offered by the Ryzen Master software ("Game Mode").

                    Comment


                    • #20
                      Originally posted by untore View Post
                      Thin clients gaming now at reasonable prices
                      Thin clients should be killed with fire.

                      Comment

                      Working...
                      X