Times (Elapsed / CPU) for BRP5/6/6-Beta on various CPU/GPU combos - DISCUSSION Thread

Forums › Cruncher's Corner

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3522
Credit: 689088488
RAC: 214619

RE: I have opted in for the

15 Mar 2015 14:10:25 UTC
Message 130778 in response to message 130776
(moderation:
    )

    Quote:
    I have opted in for the Beta and it seems these tasks have a much higher priority over standard units. It's no problem as far as Beta would (besides testing) crunch real data (not dummies). Is that so ?

    It is definitely real data, so 100% of the computing time goes to real science. By flagging the app "beta" we make sure that the second result ("wingman's result") that is used to validate a result is generated by a non-beta application, which gives us the extra-safety to let beta-apps crunch real data.

    Quote:

    I have also noticed that when running a single WU on a Tesla K20, the GPU usage seems to be fluctuating from 0% to >90% ever few seconds. Not sure if that is normal behavior..

    Depends on the sampling interval. There are small time intervals when the CPU is processing stuff and if the sampling interval of the NVIDIA tool is small enough to fit in, this will result in a 0% reading.

    HB

    Mumak
    Joined: 26 Feb 13
    Posts: 325
    Credit: 3316958677
    RAC: 1173029

    RE: It is definitely real

    15 Mar 2015 14:20:14 UTC
    Message 130779 in response to message 130778
    (moderation:
      )

      Quote:
      It is definitely real data, so 100% of the computing time goes to real science. By flagging the app "beta" we make sure that the second result ("wingman's result") that is used to validate a result is generated by a non-beta application, which gives us the extra-safety to let beta-apps crunch real data.

      Thanks for the clarification, this is the perfect combination.

      Quote:

      Depends on the sampling interval. There are small time intervals when the CPU is processing stuff and if the sampling interval of the NVIDIA tool is small enough to fit in, this will result in a 0% reading.

      Sampling interval is set to 2 seconds and almost every other (or a few other) tick the fluctuation occurs. I have checked the standard non-beta units and they show similar behavior, though the load seems somewhat more stable there. If you wish (and think this is not OK and is worth checking), I can create some graphs.

      -----

      ExtraTerrestrial Apes
      ExtraTerrestria...
      Joined: 10 Nov 04
      Posts: 770
      Credit: 539720850
      RAC: 150469

      RE: If you wish (and think

      15 Mar 2015 16:03:32 UTC
      Message 130780 in response to message 130779
      (moderation:
        )

        Quote:
        If you wish (and think this is not OK and is worth checking), I can create some graphs.


        I suppose this is quite normal, my HD4000 showed it as well. It's part of the reason you get higher throughput with 2 concurrent WUs on almost any GPU here.

        MrS

        Scanning for our furry friends since Jan 2002

        ExtraTerrestrial Apes
        ExtraTerrestria...
        Joined: 10 Nov 04
        Posts: 770
        Credit: 539720850
        RAC: 150469

        HBM wrote:Not sure yet. As I

        15 Mar 2015 16:06:15 UTC
        Message 130781 in response to message 130775
        (moderation:
          )

          HBM wrote:
          Not sure yet. As I said the GW search needs most of our attention at the moment. For those participating in the beta test this should make no difference (except you will drop in the stats because everyone else will be catching up ;-) )


          ... except for the Intel GPU guys who stopped receiving betas since the 1st few of them last week. Any particular reason for this? Is it just that the full WUs take so long on those GPUs?

          MrS

          Scanning for our furry friends since Jan 2002

          Mumak
          Joined: 26 Feb 13
          Posts: 325
          Credit: 3316958677
          RAC: 1173029

          Here some graphics FYI. Used

          15 Mar 2015 16:09:41 UTC
          Message 130782
          (moderation:
            )

            Here some graphics FYI.
            Used tool: HWiNFO64, info retrieved via NVML, sampling interval 2 secs
            GPU: Tesla K20c (ECC off, 758 MHz)
            single WU

            v1.39:

            v1.52:

            -----

            Mumak
            Joined: 26 Feb 13
            Posts: 325
            Credit: 3316958677
            RAC: 1173029

            Some preliminary numbers for

            16 Mar 2015 7:08:51 UTC
            Message 130783
            (moderation:
              )

              Some preliminary numbers for the Beta run for comparison:

              GPU: NVIDIA GeForce GTX 660 Ti
              CPU: Intel Core i7-860
              2 x BRP6 v1.39: 17,300 - 18,400
              2 x BRP6 v1.52: 11,200 - 11,400

              GPU: NVIDIA Tesla K20c (ECC off, GPU clock 758 MHz)
              CPU: 2 x Intel Xeon E5-2650 v3
              2 x BRP6 v1.39: 11,200 - 12,700
              2 x BRP6 v1.52: 6,600 - 9,200
              Here you can see the variance with the Beta is much higher. I don't know why yet..

              -----

              Mumak
              Joined: 26 Feb 13
              Posts: 325
              Credit: 3316958677
              RAC: 1173029

              Here in graph - this is on

              16 Mar 2015 8:58:01 UTC
              Message 130784
              (moderation:
                )

                Here in graph - this is on the Tesla K20c.

                -----

                Filipe
                Filipe
                Joined: 10 Mar 05
                Posts: 176
                Credit: 366490343
                RAC: 42451

                HOST: I3 4130 + AMD RADEON R9

                17 Mar 2015 10:06:19 UTC
                Message 130785
                (moderation:
                  )

                  HOST: I3 4130 + AMD RADEON R9 280 (CPU Running 1 GW unit):

                  BRP6 v1.41 21600sec (5 GPU tasks running simultaneously)

                  BRP6 v1.52 13000sec (5 GPU tasks running simultaneously)

                  Improvement: 40%!!!!

                  Mumak
                  Joined: 26 Feb 13
                  Posts: 325
                  Credit: 3316958677
                  RAC: 1173029

                  I have updated the above

                  17 Mar 2015 10:40:11 UTC
                  Message 130786 in response to message 130784
                  (moderation:
                    )

                    I have updated the above graph with more data and other GPUs.

                    -----

                    Chris
                    Chris
                    Joined: 9 Apr 12
                    Posts: 61
                    Credit: 45056670
                    RAC: 0

                    I only ran 2 so far. 15%

                    18 Mar 2015 1:19:23 UTC
                    Message 130787
                    (moderation:
                      )

                      I only ran 2 so far.

                      15% faster on a GTX 640 (about 25k seconds vs 30k seconds), but the CPU usage was reduced to a third (~2k instead of ~6k).

                      Comment viewing options

                      Select your preferred way to display the comments and click "Save settings" to activate your changes.

                      Forums › Cruncher's Corner