Announcement

Collapse
No announcement yet.

Benchmarking A 10-Core Tyan/IBM POWER Server For ~$300 USD

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

  • Benchmarking A 10-Core Tyan/IBM POWER Server For ~$300 USD

    Phoronix: Benchmarking A 10-Core Tyan/IBM POWER Server For ~$300 USD

    If you live in the EU and have been wanting to explore IBM POWER hardware on Linux, a load of Tyan Habanero servers recently became available through a German retailer for 269 EUR (~$306 USD) that comes equipped with a 10-core POWER8 processor. While not POWER9, it's still an interesting Linux-capable beast and the price is unbeatable if you have been wanting to add POWER hardware to your collection. Phoronix reader Lauri Kasanen recently bought one of these IBM POWER servers at the 269 EUR price point and has shared thoughts on this server as well as some benchmarks. Here is Lauri's guest post checking out this low-cost 2U IBM server.

    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
    Did you do the testing with the lid of the chassis on? Without that, I could see how the cooling performance would be drastically hindered.

    Comment


    • #3
      Article title is incorrect. This is an 8-core system, as mentioned by the author on page 1. Not a 10-core system like the title suggests.

      If it shipped to Canada, included the drive caddies, and didn't sound like a jet plane, this would make an interesting little storage server to play with at home.

      Comment


      • #4
        I think some of your benchmark results are not reflective of the hardware. I've grabbed one of these systems myself and got significantly better results for some tests.

        You can find my results here:
        OpenBenchmarking.org, Phoronix Test Suite, Linux benchmarking, automated benchmarking, benchmarking results, benchmarking repository, open source benchmarking, benchmarking test profiles

        OpenBenchmarking.org, Phoronix Test Suite, Linux benchmarking, automated benchmarking, benchmarking results, benchmarking repository, open source benchmarking, benchmarking test profiles

        OpenBenchmarking.org, Phoronix Test Suite, Linux benchmarking, automated benchmarking, benchmarking results, benchmarking repository, open source benchmarking, benchmarking test profiles


        With IBM's optimized HPCG benchmark (a benchmark measuring sparse linear algebra performance that is often used in the HPC space) I get results of around 11 GFlops, which puts it in striking distance to Intel's self-published results for 20-core Broadwell Xeons.

        I do have the 10-core version instead of the 8-core, but it only reaches clockspeeds up to 3.5 GHz, so theoretical peak performance is not much better than the 8-core version. I do however have 16 8GB DIMMs installed, and I urgently suggest you do the same. This system has a quad channel memory interface at all four centaur memory buffer chips, so you need at least 16 DIMMs to maximize memory performance. I get around 100 GB/s sustained triad stream memory performance, which is not bad for a DDR3 based system.

        I also suggest you update the firmware; my unit didn't lower fan speed properly until I upgraded the firmware. I have additionally swapped out the default fans for 3 Supermicro FAN 0074L4 and 2 Supermicro FAN 0126-L4 in front of the CPUs. Together with a custom air duct made from duct tape and packaging material, this keeps the system cool enough to run at turbo boost frequencies permanently while consuming 100W less power and emitting bearable noise, comparable to "normal" servers.

        Comment


        • #5
          This is why I have a "no rack mounted computers at home"-rule.

          Comment


          • #6
            Not sure why you think DDR3 is expensive. It's less than $40 for 16GB sticks... 256GB is under $650 that's dirt cheap.

            Comment


            • #7
              Originally posted by sjukfan View Post
              This is why I have a "no rack mounted computers at home"-rule.
              Depends on the chassis. Large fan can move the same air at lower rpm than a small fan. 1U servers are all unbearably loud. 2U are "ok". 3U and up are generally pretty quiet, or can be made so via BIOS settings. Or go DIY route, rosewill 4U rack chassis with some Noctua fans makes for a nearly silent build.

              Comment


              • #8
                Could someone share the link of the retailer?

                Comment


                • #9
                  Originally posted by toguro123 View Post
                  Could someone share the link of the retailer?
                  On the last page under the last picture -> https://www.piospartslap.de/Tyan-Rac...GHz-2x-750Watt

                  Comment


                  • #10
                    If anyone is considering purchasing one and you have questions regarding this machine, feel free to ask me either in this thread or via DM.

                    Comment

                    Working...
                    X