Gravitational Wave S6 Directed Search (CasA) v. 1.06 Beta Test

Forums › Technical News

Bernd Machenschalk
Bernd Machenschalk
Moderator
Administrator
Joined: 15 Oct 04
Posts: 4312
Credit: 250534013
RAC: 34143
10 Apr 2014 7:13:39 UTC
Topic 197528
(moderation:
    )

    There's a new app version 1.06 of the S6CasA search. For the first time (in years), we will run the Beta test on Einstein@Home (instead of Albert). To participate in this (and future) Bete test, please set "Run beta/test application versions?" in your Einstein@Home preferences to "yes".

    This App version reserves the last 1% of progress for the last step that we thought to take only minutes (toplist parameter reconstruction), but on some computers apparently leads to tasks sticking at 100% for hours with no apparent progress.

    As you may have noticed from the Applications page, there will also be an OpenCL (GPU) version for this application later, but that will be a separate announcement. In any case this will also require you to set "run Beta test app versions" in your preferences.

    BM

    BM

    Nigel Garvey
    Nigel Garvey
    Joined: 4 Oct 10
    Posts: 51
    Credit: 32722586
    RAC: 87460

    Gravitational Wave S6 Directed Search (CasA) v. 1.06 Beta Test

    10 Apr 2014 8:08:33 UTC
    Message 120928
    (moderation:
      )

      Hi Bernd.

      The (ALTIVEC-Beta) version is listed on the Applications page as running "on Intel".

      NG

      NG

      Bernd Machenschalk
      Bernd Machenschalk
      Moderator
      Administrator
      Joined: 15 Oct 04
      Posts: 4312
      Credit: 250534013
      RAC: 34143

      Well spotted! Fixed. Yep,

      10 Apr 2014 8:27:19 UTC
      Message 120929
      (moderation:
        )

        Well spotted! Fixed. Yep, also a Beta test for our Beta test...

        BM

        BM

        Claggy
        Claggy
        Joined: 29 Dec 06
        Posts: 560
        Credit: 2699403
        RAC: 0

        I tried to get 1.06 Beta

        10 Apr 2014 12:14:20 UTC
        Message 120930
        (moderation:
          )

          I tried to get 1.06 Beta work, But the scheduler sent me the v1.05 (X64) app and work:

          http://einsteinathome.org/host/10224316/tasks

          The host is on the work venue, the work venue only has Use CPU, CasA, Run beta/test application versions, and Run CPU versions of applications for which GPU versions are available selected to 'yes'.

          Quote:
          2014-04-10 12:01:25.6655 [PID=23772] Request: [USER#xxxxx] [HOST#10224316] [IP xxx.xxx.xxx.119] client 7.2.44
          2014-04-10 12:01:25.7916 [PID=23772] [send] effective_ncpus 2 max_jobs_on_host_cpu 999999 max_jobs_on_host 999999
          2014-04-10 12:01:25.7917 [PID=23772] [send] effective_ngpus 1 max_jobs_on_host_gpu 999999
          2014-04-10 12:01:25.7917 [PID=23772] [send] Not using matchmaker scheduling; Not using EDF sim
          2014-04-10 12:01:25.7917 [PID=23772] [send] CPU: req 9812.65 sec, 0.00 instances; est delay 0.00
          2014-04-10 12:01:25.7917 [PID=23772] [send] CUDA: req 0.00 sec, 0.00 instances; est delay 0.00
          2014-04-10 12:01:25.7917 [PID=23772] [send] work_req_seconds: 9812.65 secs
          2014-04-10 12:01:25.7917 [PID=23772] [send] available disk 9.76 GB, work_buf_min 108000
          2014-04-10 12:01:25.7917 [PID=23772] [send] active_frac 0.974359 on_frac 0.939265 DCF 1.057578
          2014-04-10 12:01:25.7927 [PID=23772] [send] [HOST#10224316] is reliable
          2014-04-10 12:01:25.7929 [PID=23772] [send] set_trust: random choice for error rate 0.027737: yes
          2014-04-10 12:01:25.7929 [PID=23772] [mixed] sending non-locality work first (0.4544)
          2014-04-10 12:01:25.8183 [PID=23772] [mixed] sending locality work second
          2014-04-10 12:01:25.9008 [PID=23772] [version] Checking plan class 'X64'
          2014-04-10 12:01:25.9055 [PID=23772] [version] reading plan classes from file '/BOINC/projects/EinsteinAtHome/plan_class_spec.xml'
          2014-04-10 12:01:25.9057 [PID=23772] [version] plan class ok
          2014-04-10 12:01:25.9057 [PID=23772] [version] Checking plan class 'Beta'
          2014-04-10 12:01:25.9057 [PID=23772] [version] plan class ok
          2014-04-10 12:01:25.9057 [PID=23772] [version] Checking plan class 'GWopencl-ati-Beta'
          2014-04-10 12:01:25.9057 [PID=23772] [version] Plan class disabled
          2014-04-10 12:01:25.9057 [PID=23772] [version] Checking plan class 'GWopencl-nvidia-Beta'
          2014-04-10 12:01:25.9057 [PID=23772] [version] Plan class disabled
          2014-04-10 12:01:25.9057 [PID=23772] [version] Checking plan class 'X64-Beta'
          2014-04-10 12:01:25.9057 [PID=23772] [version] plan class ok
          2014-04-10 12:01:25.9057 [PID=23772] [version] Checking plan class 'SSE2'
          2014-04-10 12:01:25.9058 [PID=23772] [version] plan class ok
          2014-04-10 12:01:25.9058 [PID=23772] [version] Checking plan class 'SSE2-Beta'
          2014-04-10 12:01:25.9058 [PID=23772] [version] plan class ok
          2014-04-10 12:01:25.9058 [PID=23772] [version] Best version of app einstein_S6CasA is ID 473 (3.78 GFLOPS)
          2014-04-10 12:01:25.9070 [PID=23772] [debug] Sorted list of URLs follows [host timezone: UTC+3600]
          2014-04-10 12:01:25.9070 [PID=23772] [debug] zone=+03600 url=http://einstein2.aei.uni-hannover.de
          2014-04-10 12:01:25.9070 [PID=23772] [debug] zone=-18900 url=http://einstein2.aei.uni-hannover.de
          2014-04-10 12:01:25.9070 [PID=23772] [debug] zone=-21600 url=http://einstein-dl2.phys.uwm.edu
          2014-04-10 12:01:25.9070 [PID=23772] [debug] zone=-21600 url=http://einstein-dl3.phys.uwm.edu
          2014-04-10 12:01:25.9070 [PID=23772] [debug] zone=-28800 url=http://einstein.ligo.caltech.edu
          2014-04-10 12:01:25.9074 [PID=23772] [send] [HOST#10224316] Sending app_version 473 einstein_S6CasA 7 105 X64; 3.78 GFLOPS
          2014-04-10 12:01:25.9199 [PID=23772] [send] est. duration for WU 187559145: unscaled 26699.62 scaled 30853.92
          2014-04-10 12:01:25.9199 [PID=23772] [HOST#10224316] Sending [RESULT#430559864 h1_0990.50_S6Directed__S6CasAf40a_990.7Hz_1310_0] (est. dur. 30853.92 seconds)
          2014-04-10 12:01:25.9220 [PID=23772] [send] don't need more work
          2014-04-10 12:01:25.9220 [PID=23772] [send] don't need more work
          2014-04-10 12:01:25.9220 [PID=23772] [send] don't need more work
          2014-04-10 12:01:25.9233 [PID=23772] Sending reply to [HOST#10224316]: 1 results, delay req 60.00
          2014-04-10 12:01:25.9236 [PID=23772] Scheduler ran 0.266 seconds

          Claggy

          Bernd Machenschalk
          Bernd Machenschalk
          Moderator
          Administrator
          Joined: 15 Oct 04
          Posts: 4312
          Credit: 250534013
          RAC: 34143

          Thanks for reporting, try

          10 Apr 2014 12:18:57 UTC
          Message 120931
          (moderation:
            )

            Thanks for reporting, try again, please.

            BM

            BM

            Claggy
            Claggy
            Joined: 29 Dec 06
            Posts: 560
            Credit: 2699403
            RAC: 0

            RE: Thanks for reporting,

            10 Apr 2014 13:08:54 UTC
            Message 120932 in response to message 120931
            (moderation:
              )

              Quote:

              Thanks for reporting, try again, please.

              BM


              O.K, i have a v1.06 (X64-Beta) task now, took a couple of scheduler contacts to get it through, But i'm on a relatively poor WiFi connection through a public hotspot so that is probably the reason,

              Claggy

              ahj
              ahj
              Joined: 25 Jul 10
              Posts: 17
              Credit: 4331992
              RAC: 0

              RE: As you may have noticed

              11 Apr 2014 6:45:04 UTC
              Message 120933
              (moderation:
                )

                Quote:
                As you may have noticed from the Applications page, there will also be an OpenCL (GPU) version for this application later, but that will be a separate announcement. In any case this will also require you to set "run Beta test app versions" in your preferences.

                How much of an improvement do you guys think the GPU GW application is over the CPU app?

                Are we talking 2-3 times faster, or an order of magnitude?

                Bernd Machenschalk
                Bernd Machenschalk
                Moderator
                Administrator
                Joined: 15 Oct 04
                Posts: 4312
                Credit: 250534013
                RAC: 34143

                Depends on your card, a bit

                11 Apr 2014 9:16:04 UTC
                Message 120934 in response to message 120933
                (moderation:
                  )

                  Depends on your card, a bit on double precision performance and a lot on GPU memory bandwidth. I saw it doing an ordinary S6CasA task in about 1000s on an ATI 6970.

                  BM

                  BM

                  ahj
                  ahj
                  Joined: 25 Jul 10
                  Posts: 17
                  Credit: 4331992
                  RAC: 0

                  So about 3x faster on average

                  11 Apr 2014 9:51:39 UTC
                  Message 120935 in response to message 120934
                  (moderation:
                    )

                    So about 3x faster on average compared to my i7 4771...

                    Not bad.

                    And I'm assuming these will run on iGPUs?

                    Cheers

                    Bernd Machenschalk
                    Bernd Machenschalk
                    Moderator
                    Administrator
                    Joined: 15 Oct 04
                    Posts: 4312
                    Credit: 250534013
                    RAC: 34143

                    RE: So about 3x faster on

                    11 Apr 2014 9:54:24 UTC
                    Message 120937 in response to message 120935
                    (moderation:
                      )

                      Quote:
                      So about 3x faster on average compared to my i7 4770...

                      You run a S6CasA task in less than an hour (3600s)?

                      BM

                      BM

                      ahj
                      ahj
                      Joined: 25 Jul 10
                      Posts: 17
                      Credit: 4331992
                      RAC: 0

                      Yes. According to you it

                      11 Apr 2014 9:58:32 UTC
                      Message 120938 in response to message 120937
                      (moderation:
                        )

                        Yes.

                        According to you it takes about 1000s to complete one GW unit on a 6970, and on my 4771, it takes 24000s to complete a unit *per thread*, meaning on average I compute one unit every 3000s (24000s/8 cores).

                        Comment viewing options

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

                        Forums › Technical News