Announcement

Collapse
No announcement yet.

AMD EPYC On Ubuntu 18.10 Putting Up A Stronger Fight Against Xeon Gold

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

  • AMD EPYC On Ubuntu 18.10 Putting Up A Stronger Fight Against Xeon Gold

    Phoronix: AMD EPYC On Ubuntu 18.10 Putting Up A Stronger Fight Against Xeon Gold

    With hitting the home stretch to Ubuntu 18.10, I've started with my usual benchmark process for checking out this next Ubuntu Linux release dubbed the Cosmic Cuttlefish. Yesterday were Ubuntu 18.10 benchmarks on seven desktop systems from Intel and AMD while next on my agenda has been checking out the server performance. Here's the first of those server tests on Ubuntu 18.10 with some initial AMD EPYC and Intel Xeon Gold tests.

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    Are you sure EPYC has worse IPC? You state this often, but I'm pretty sure the EPYC has a superior IPC to the Xeon, but lower boost clockspeed.

    Comment


    • #3
      Curious how much power these systems are drawing from the wall. I don't know if their configurations are close enough to provide a good comparison, but performance per watt metrics would be very helpful if they are possible.

      Comment


      • #4
        Originally posted by existensil View Post
        Curious how much power these systems are drawing from the wall. I don't know if their configurations are close enough to provide a good comparison, but performance per watt metrics would be very helpful if they are possible.
        Right I didn't do perf-per-Watt since really not comparable with being 1P vs. 2P.
        Michael Larabel
        https://www.michaellarabel.com/

        Comment


        • #5
          Originally posted by Michael View Post

          Right I didn't do perf-per-Watt since really not comparable with being 1P vs. 2P.
          Surely it's just as comparable as all the other stats?

          Comment


          • #6
            Originally posted by Michael View Post

            Right I didn't do perf-per-Watt since really not comparable with being 1P vs. 2P.
            No offense, but I tend to disagree. Based on performance, these systems are pretty close. If you can get the same performance at lower power consumption due to the single socket configuration, isn't this a selling point? There are no inherent advantages of a dual socket system. Actually, it might even be harder to obtain maximum performance due to the NUMA arrangement.
            The number of sockets is also important when it comes to performance per dollar. The price of the CPU(s) alone is not necessarily the important figure. I assume, the single socket mainboard might be cheaper than the Intel mainboard.

            Comment


            • #7
              Originally posted by GruenSein View Post

              No offense, but I tend to disagree. Based on performance, these systems are pretty close. If you can get the same performance at lower power consumption due to the single socket configuration, isn't this a selling point? There are no inherent advantages of a dual socket system. Actually, it might even be harder to obtain maximum performance due to the NUMA arrangement.
              The number of sockets is also important when it comes to performance per dollar. The price of the CPU(s) alone is not necessarily the important figure. I assume, the single socket mainboard might be cheaper than the Intel mainboard.
              Valid points, will include at least the metrics on OpenBenchmarking.org in my next comparison for those interested in the power consumption / perf per Watt.
              Michael Larabel
              https://www.michaellarabel.com/

              Comment


              • #8
                IMHO, it is also worth noting, that if one is going single-socket, they may get performance very close to 7601 at half the CPU price, with 7551P. 7601's advantage of being able to work dual-socket is not even utilized in the test platform.

                Comment


                • #9
                  Originally posted by phoronix View Post
                  As these two server CPUs carry different pricing, ($2679 for the Xeon Gold 6138 / $5358 for two, or $4619 for the EPYC 7601), I also did some performance-per-dollar tests with Ubuntu 18.10 based on current retail pricing via NewEgg:
                  I think the pricing comparison is even somewhat unfair to AMD. Phoronix uses a dual-socket capable CPU in a single-socket system. An Epyc 7551P is much cheaper yet only slightly slower.

                  Alternatively, Dual Epyc 7451 can be had at roughly the same price as the dual Xeon Gold 6138, and dual 7401 for even less than the single 7601.

                  Originally posted by phoronix View Post
                  Intel CPUs have been hit harder due to Spectre/Meltdown.
                  In addition, L1TF mitigation is only partial in default configuration.
                  If full mitigation is enabled (which should be the default, but I digress), then Intel performs even worse.

                  Comment


                  • #10
                    Originally posted by chithanh View Post
                    I think the pricing comparison is even somewhat unfair to AMD. Phoronix uses a dual-socket capable CPU in a single-socket system. An Epyc 7551P is much cheaper yet only slightly slower.
                    It's all a matter of the hardware I have available... No AMD 2P server. AMD talked of sending out a 2P system, but so far haven't seen it.
                    Michael Larabel
                    https://www.michaellarabel.com/

                    Comment

                    Working...
                    X