Not requesting work for Intel GPU - why? Please help.

Forums › Cruncher's Corner

Raistmer*
Raistmer*
Joined: 20 Feb 05
Posts: 208
Credit: 180991202
RAC: 10781
23 Sep 2013 21:17:20 UTC
Topic 197202
(moderation:
    )

    http://einstein.phys.uwm.edu//host_sched_logs/6660/6660606

    2013-09-23 21:09:37.1758 [PID=11917] Request: [USER#xxxxx] [HOST#6660606] [IP xxx.xxx.xxx.252] client 7.0.64
    2013-09-23 21:09:37.1862 [PID=11917] [send] effective_ncpus 4 max_jobs_on_host_cpu 999999 max_jobs_on_host 999999
    2013-09-23 21:09:37.1862 [PID=11917] [send] effective_ngpus 1 max_jobs_on_host_gpu 999999
    2013-09-23 21:09:37.1862 [PID=11917] [send] Not using matchmaker scheduling; Not using EDF sim
    2013-09-23 21:09:37.1862 [PID=11917] [send] CPU: req 0.00 sec, 0.00 instances; est delay 0.00
    2013-09-23 21:09:37.1863 [PID=11917] [send] Intel GPU: req 1.00 sec, 1.00 instances; est delay 0.00
    2013-09-23 21:09:37.1863 [PID=11917] [send] work_req_seconds: 0.00 secs
    2013-09-23 21:09:37.1863 [PID=11917] [send] available disk 9.74 GB, work_buf_min 432000
    2013-09-23 21:09:37.1863 [PID=11917] [send] active_frac 0.992369 on_frac 0.847424 DCF 1.657400
    2013-09-23 21:09:37.1875 [PID=11917] [send] [HOST#6660606] not reliable; max_result_day 2
    2013-09-23 21:09:37.1876 [PID=11917] [send] set_trust: random choice for error rate 0.000010: yes
    2013-09-23 21:09:37.1876 [PID=11917] [mixed] sending locality work first (0.3807)

    There is capable Intel GPU as part of Ivy Bridge device. It's idle (and log confirms that). But 0 secs of work asked for it it seems. Why so? How to fix ?

    And another attempt:
    9/24/2013 1:19:26 AM | Einstein@Home | update requested by user
    9/24/2013 1:19:30 AM | Einstein@Home | [sched_op] Starting scheduler request
    9/24/2013 1:19:30 AM | Einstein@Home | Sending scheduler request: Requested by user.
    9/24/2013 1:19:30 AM | Einstein@Home | Requesting new tasks for intel_gpu
    9/24/2013 1:19:30 AM | Einstein@Home | [sched_op] CPU work request: 0.00 seconds; 0.00 devices
    9/24/2013 1:19:30 AM | Einstein@Home | [sched_op] intel_gpu work request: 1.00 seconds; 1.00 devices
    9/24/2013 1:19:46 AM | Einstein@Home | Scheduler request completed: got 0 new tasks
    9/24/2013 1:19:46 AM | Einstein@Home | [sched_op] Server version 611
    9/24/2013 1:19:46 AM | Einstein@Home | No work sent
    9/24/2013 1:19:46 AM | Einstein@Home | see scheduler log messages on http://einstein.phys.uwm.edu//host_sched_logs/6660/6660606
    9/24/2013 1:19:46 AM | Einstein@Home | Project requested delay of 60 seconds
    9/24/2013 1:19:46 AM | Einstein@Home | [sched_op] Deferring communication for 1 min 0 sec
    9/24/2013 1:19:46 AM | Einstein@Home | [sched_op] Reason: requested by project

    Holmis
    Joined: 4 Jan 05
    Posts: 1118
    Credit: 1055935564
    RAC: 0

    Not requesting work for Intel GPU - why? Please help.

    23 Sep 2013 21:23:35 UTC
    Message 118798
    (moderation:
      )

      I followed your link to the server logs and found this message:

      2013-09-23 21:09:41.4003 [PID=11917] [version] OpenCL device version required min: 102, supplied: 101

      Could a driver update fix this or is it hardware dependent?

      Raistmer*
      Raistmer*
      Joined: 20 Feb 05
      Posts: 208
      Credit: 180991202
      RAC: 10781

      Thanks. It could be the

      23 Sep 2013 21:36:34 UTC
      Message 118799 in response to message 118798
      (moderation:
        )

        Thanks. It could be the case.

        Device Type: CL_DEVICE_TYPE_GPU
        Name: Intel(R) HD Graphics 2500
        Vendor: Intel(R) Corporation
        Driver version: 8.15.10.2656
        Profile: FULL_PROFILE
        Version: OpenCL 1.1
        Extensions: cl_khr_icd cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_gl_sharing cl_khr_d3d10_sharing cl_intel_dx9_media_sharing cl_khr_3d_image_writes cl_khr_byte_addressable_store

        Will try to update driver/SDK.

        Raistmer*
        Raistmer*
        Joined: 20 Feb 05
        Posts: 208
        Credit: 180991202
        RAC: 10781

        Now: Device Type:

        23 Sep 2013 21:48:18 UTC
        Message 118800
        (moderation:
          )

          Now:

          Device Type: CL_DEVICE_TYPE_GPU
          Name: Intel(R) HD Graphics 2500
          Vendor: Intel(R) Corporation
          Driver version: 9.18.10.3257
          Profile: FULL_PROFILE
          Version: OpenCL 1.2
          Extensions: cl_khr_icd cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_gl_sharing cl_khr_d3d10_sharing cl_intel_dx9_media_sharing cl_khr_3d_image_writes cl_khr_byte_addressable_store cl_khr_gl_event cl_khr_gl_msaa_sharing cl_khr_depth_images cl_khr_gl_depth_images cl_khr_dx9_media_sharing cl_khr_d3d11_sharing cl_khr_image2d_from_buffer

          And GPU got task to crunch. Thanks again. Topic closed.

          Raistmer*
          Raistmer*
          Joined: 20 Feb 05
          Posts: 208
          Credit: 180991202
          RAC: 10781

          After CUDA GPU install no

          26 Oct 2013 16:29:50 UTC
          Message 118801
          (moderation:
            )

            After CUDA GPU install no intel work fetched again.

            Quote:
            2013-10-26 16:23:28.9125 [PID=28934] Request: [USER#xxxxx] [HOST#6660606] [IP xxx.xxx.xxx.252] client 7.0.64
            2013-10-26 16:23:28.9207 [PID=28934] [send] effective_ncpus 4 max_jobs_on_host_cpu 999999 max_jobs_on_host 999999
            2013-10-26 16:23:28.9207 [PID=28934] [send] effective_ngpus 3 max_jobs_on_host_gpu 999999
            2013-10-26 16:23:28.9207 [PID=28934] [send] Not using matchmaker scheduling; Not using EDF sim
            2013-10-26 16:23:28.9207 [PID=28934] [send] CPU: req 0.00 sec, 0.00 instances; est delay 0.00
            2013-10-26 16:23:28.9207 [PID=28934] [send] CUDA: req 0.00 sec, 0.00 instances; est delay 0.00
            2013-10-26 16:23:28.9207 [PID=28934] [send] Intel GPU: req 1.00 sec, 1.00 instances; est delay 0.00
            2013-10-26 16:23:28.9207 [PID=28934] [send] work_req_seconds: 0.00 secs
            2013-10-26 16:23:28.9207 [PID=28934] [send] available disk 9.57 GB, work_buf_min 432000
            2013-10-26 16:23:28.9208 [PID=28934] [send] active_frac 0.999650 on_frac 0.991189 DCF 4.278733
            2013-10-26 16:23:28.9213 [PID=28934] [send] [HOST#6660606] is reliable
            2013-10-26 16:23:28.9214 [PID=28934] [send] set_trust: random choice for error rate 0.000010: yes
            2013-10-26 16:23:28.9214 [PID=28934] [mixed] sending locality work first (0.4547)
            2013-10-26 16:23:28.9838 [PID=28934] [version]: App 'einstein_S6CasA' (24) not selected
            2013-10-26 16:23:28.9840 [PID=28934] [mixed] sending non-locality work second
            2013-10-26 16:23:28.9987 [PID=28934] [version] Checking plan class 'BRP4G-opencl-ati'
            2013-10-26 16:23:28.9996 [PID=28934] [version] reading plan classes from file '/BOINC/projects/EinsteinAtHome/plan_class_spec.xml'
            2013-10-26 16:23:28.9996 [PID=28934] [version] parsed project prefs setting 'gpu_util_brp': 1.000000
            2013-10-26 16:23:28.9996 [PID=28934] [version] No ATI devices found
            2013-10-26 16:23:28.9997 [PID=28934] [version] Checking plan class 'BRP4G-cuda32'
            2013-10-26 16:23:28.9997 [PID=28934] [version] parsed project prefs setting 'gpu_util_brp': 1.000000
            2013-10-26 16:23:28.9997 [PID=28934] [version] Peak flops supplied: 4.48e+10
            2013-10-26 16:23:28.9997 [PID=28934] [version] plan class ok
            2013-10-26 16:23:28.9997 [PID=28934] [version] Don't need CUDA jobs, skipping version 139 for einsteinbinary_BRP4G (BRP4G-cuda32)
            2013-10-26 16:23:28.9997 [PID=28934] [version] Checking plan class 'BRP4G-cuda32-nv301'
            2013-10-26 16:23:28.9997 [PID=28934] [version] parsed project prefs setting 'gpu_util_brp': 1.000000
            2013-10-26 16:23:28.9997 [PID=28934] [version] driver version required min: -30100, supplied: 26306
            2013-10-26 16:23:28.9997 [PID=28934] [version] Checking plan class 'BRP4G-opencl-ati'
            2013-10-26 16:23:28.9997 [PID=28934] [version] parsed project prefs setting 'gpu_util_brp': 1.000000
            2013-10-26 16:23:28.9997 [PID=28934] [version] No ATI devices found
            2013-10-26 16:23:28.9997 [PID=28934] [version] no app version available: APP#25 (einsteinbinary_BRP4G) PLATFORM#9 (windows_x86_64) min_version 0
            2013-10-26 16:23:28.9997 [PID=28934] [version] no app version available: APP#25 (einsteinbinary_BRP4G) PLATFORM#2 (windows_intelx86) min_version 0
            2013-10-26 16:23:29.0062 [PID=28934] [debug] [HOST#6660606] MSG(high) No work sent
            2013-10-26 16:23:29.0062 [PID=28934] [debug] [HOST#6660606] MSG(high) see scheduler log messages on http://einstein.phys.uwm.edu//host_sched_logs/6660/6660606
            2013-10-26 16:23:29.0062 [PID=28934] Sending reply to [HOST#6660606]: 0 results, delay req 60.00
            2013-10-26 16:23:29.0066 [PID=28934] Scheduler ran 0.101 seconds

            My client asks for intel GPU work, but looks like intel_gup plan class not checked by server at all, only CPU, ati and cuda.
            Why so and how to fix?
            Is it client issue or server setup issue?

            http://einstein.phys.uwm.edu//host_sched_logs/6660/6660606

            EDIT: hm... looks like server misconfiguration to me. When I enabled ALL work on ALL devices (including CPU) task was sent.

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

            RE: EDIT: hm... looks like

            26 Oct 2013 18:24:06 UTC
            Message 118802 in response to message 118801
            (moderation:
              )

              Quote:
              EDIT: hm... looks like server misconfiguration to me. When I enabled ALL work on ALL devices (including CPU) task was sent.

              The opencl-intel_gpu apps are thrown in with the CPU apps in the BRP (Arecibo) Search, so probably configured as a CPU app, as opposed to an Intel GPU app.

              http://einstein.phys.uwm.edu/apps.php

              Claggy

              Richard Haselgrove
              Richard Haselgrove
              Joined: 10 Dec 05
              Posts: 2143
              Credit: 2921491296
              RAC: 947573

              These settings work for

              26 Oct 2013 19:13:55 UTC
              Message 118803 in response to message 118801
              (moderation:
                )

                These settings work for me:

                That's with a v7.2 client, of course.

                2013-10-26 19:09:10.9321 [PID=2247 ] [send] CPU: req 0.00 sec, 0.00 instances; est delay 0.00
                2013-10-26 19:09:10.9321 [PID=2247 ] [send] Intel GPU: req 35.72 sec, 0.00 instances; est delay 0.00
                2013-10-26 19:09:10.9321 [PID=2247 ] [send] work_req_seconds: 0.00 secs
                ...
                2013-10-26 19:09:10.9480 [PID=2247 ] [send] [HOST#8864187] Sending app_version 452 einsteinbinary_BRP4 9 134 opencl-intel_gpu; 12.57 GFLOPS
                2013-10-26 19:09:10.9495 [PID=2247 ] [send] est. duration for WU 177343315: unscaled 1392.08 scaled 649.48
                2013-10-26 19:09:10.9496 [PID=2247 ] [HOST#8864187] Sending [RESULT#407656572 p2030.20130203.G176.03+00.43.C.b1s0g0.00000_2767_0] (est. dur. 649.48 seconds)

                Raistmer*
                Raistmer*
                Joined: 20 Feb 05
                Posts: 208
                Credit: 180991202
                RAC: 10781

                And I had GPU enabled (GPU

                26 Oct 2013 21:41:35 UTC
                Message 118804 in response to message 118803
                (moderation:
                  )

                  And I had GPU enabled (GPU app) but CPU (CPU app) disabled.
                  Well, Ok if it's known feature but I would say not obvious feature for project newbie. Maybe worth to make it more clear on properties page.

                  archae86
                  archae86
                  Joined: 6 Dec 05
                  Posts: 3156
                  Credit: 7175814931
                  RAC: 729272

                  RE: Maybe worth to make it

                  26 Oct 2013 22:39:52 UTC
                  Message 118805 in response to message 118804
                  (moderation:
                    )

                    Quote:
                    Maybe worth to make it more clear on properties page.


                    It seems really odd to me. Looking at the valid tasks page for Richard's host with Intel graphics on it, the Binary Radio Pulsar Search (Arecibo, GPU) selection has no entries, while the Binary Radio Pulsar Search (Arecibo) selection has lots of returned work. Just to add to the noise, the Application column on that page for these tasks includes the string "opencl-intel_gpu".

                    So in order to get intel GPU work, you must tick the (apparently) non-GPU box if you are not just allowing everything!

                    I'd venture that will mislead plenty of people.

                    I'm contemplating a build soon using an i3-3225 (2-core Ivy Bridge including HD 4000) with GTX650. This preference behavior would have tripped me up for sure.

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

                    This is what Bernd

                    26 Oct 2013 22:49:24 UTC
                    Message 118806 in response to message 118805
                    (moderation:
                      )

                      This is what Bernd Machenschalk said almost two months ago:

                      http://einsteinathome.org/node/197170&nowrap=true#126858

                      Quote:

                      As you can see on the Applications page the only (active) intel_gpu application version is for BRP4 ("Binary Radio Pulsar Search (Arecibo)"). I agree that this is somewhat confusing as BRP4 is a CPU application. It stems from the fact that the Intel GPUs are only available as part of a CPU and are not as powerful (speed & RAM) as what we usually have for GPU applications.

                      Sending (some) poor Intel GPUs WUs that are designed for more powerful GPUs caused more trouble (in terms of reported errors) than we got computing power from these, so we decided to send these small CPU work units.

                      BM

                      Claggy

                      Duncan
                      Duncan
                      Joined: 26 Oct 13
                      Posts: 1
                      Credit: 740
                      RAC: 0

                      +1 for driver update

                      26 Oct 2013 23:04:48 UTC
                      Message 118807 in response to message 118806
                      (moderation:
                        )

                        +1 for driver update

                        Comment viewing options

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

                        Forums › Cruncher's Corner