Can't get new GPU tasks
25 Apr 2012 19:11:25 UTC
Topic 196294
(moderation:
I have an i7 with 8 CPU cores that run CPDN all the time. The computer's nVidia GPU runs Einstein tasks whenever I'm not using the machine. My results are almost always good. The computer has an additional work buffer of one day so in the Boinc Manager Tasks tab I usually see at least 15 BRP4cuda32 tasks waiting to be processed.
The server status page shows more than 12,000 BRP4cuda32 tasks ready to send.
But I came home from work today to find the computer's CUDA tasks almost finished. Now completely finished. The event log has said repeatedly today:
25/04/2012 19:40:29 | Einstein@Home | Sending scheduler request: To fetch work.
25/04/2012 19:40:29 | Einstein@Home | Requesting new tasks for NVIDIA GPU
25/04/2012 19:40:33 | Einstein@Home | Scheduler request completed: got 0 new tasks
25/04/2012 19:40:33 | Einstein@Home | No work sent
25/04/2012 19:40:33 | Einstein@Home | see scheduler log messages on http://einstein.phys.uwm.edu//host_sched_logs/4235/4235049
These event log messages give no reason for the computer's inability to fetch new Einstein CUDA tasks. I have looked at the web page that the event log directs me to. It says:
2012-04-25 18:21:59.1318 [PID=11442] Request: [USER#xxxxx] [HOST#4235049] [IP xxx.xxx.xxx.221] client 6.12.34
2012-04-25 18:21:59.3366 [PID=11442] [send] effective_ncpus 8 max_jobs_on_host_cpu 999999 max_jobs_on_host 999999
2012-04-25 18:21:59.3366 [PID=11442] [send] effective_ngpus 1 max_jobs_on_host_gpu 999999
2012-04-25 18:21:59.3366 [PID=11442] [send] Not using matchmaker scheduling; Not using EDF sim
2012-04-25 18:21:59.3366 [PID=11442] [send] CPU: req 0.00 sec, 0.00 instances; est delay 0.00
2012-04-25 18:21:59.3367 [PID=11442] [send] CUDA: req 95040.00 sec, 1.00 instances; est delay 0.00
2012-04-25 18:21:59.3367 [PID=11442] [send] work_req_seconds: 0.00 secs
2012-04-25 18:21:59.3367 [PID=11442] [send] available disk 85.93 GB, work_buf_min 8640
2012-04-25 18:21:59.3367 [PID=11442] [send] active_frac 0.999520 on_frac 0.812864 DCF 0.562933
2012-04-25 18:21:59.3372 [PID=11442] [send] [HOST#4235049] is reliable
2012-04-25 18:21:59.3373 [PID=11442] [send] set_trust: error rate 0.055322 > 0.050000, don't trust
2012-04-25 18:21:59.3495 [PID=11442] [version] Checking plan class 'BRP4cuda32'
2012-04-25 18:21:59.3499 [PID=11442] [version] reading plan classes from file '../plan_class_spec.xml'
2012-04-25 18:21:59.3499 [PID=11442] [version] parsed project prefs setting 'gpu_util_brp' : false : 0.000000
2012-04-25 18:21:59.3499 [PID=11442] [version] driver version required max: -29053, supplied: 29573
2012-04-25 18:21:59.3499 [PID=11442] [version] Checking plan class 'BRP4cuda32nv301'
2012-04-25 18:21:59.3499 [PID=11442] [version] parsed project prefs setting 'gpu_util_brp' : false : 0.000000
2012-04-25 18:21:59.3499 [PID=11442] [version] driver version required min: -30100, supplied: 29573
2012-04-25 18:21:59.3499 [PID=11442] [version] Checking plan class 'BRP4SSE'
2012-04-25 18:21:59.3500 [PID=11442] [version] parsed project prefs setting 'also_run_cpu' : true : 1.000000
2012-04-25 18:21:59.3500 [PID=11442] [version] project prefs setting 'also_run_cpu' (1.000000) prevents using plan class.
2012-04-25 18:21:59.3500 [PID=11442] [version] no app version available: APP#19 (einsteinbinary_BRP4) PLATFORM#2 (windows_intelx86) min_version 0
2012-04-25 18:21:59.3500 [PID=11442] [version] Don't need CPU jobs, skipping version 23 for hsgamma_FGRP1 ()
2012-04-25 18:21:59.3500 [PID=11442] [version] no app version available: APP#17 (hsgamma_FGRP1) PLATFORM#2 (windows_intelx86) min_version 0
2012-04-25 18:21:59.3577 [PID=11442] [version] Checking plan class 'SSE2'
2012-04-25 18:21:59.3577 [PID=11442] [version] Don't need CPU jobs, skipping version 113 for einstein_S6LV1 (SSE2)
2012-04-25 18:21:59.3577 [PID=11442] [version] no app version available: APP#20 (einstein_S6LV1) PLATFORM#2 (windows_intelx86) min_version 0
2012-04-25 18:21:59.3594 [PID=11442] [debug] [HOST#4235049] MSG(high) No work sent
2012-04-25 18:21:59.3595 [PID=11442] [debug] [HOST#4235049] MSG(high) see scheduler log messages on http://einstein.phys.uwm.edu//host_sched_logs/4235/4235049
2012-04-25 18:21:59.3595 [PID=11442] Sending reply to [HOST#4235049]: 0 results, delay req 60.00
2012-04-25 18:21:59.3598 [PID=11442] Scheduler ran 0.234 seconds
I cannot see here any reason for my inability to fetch more CUDA work. Perhaps I cannot understand all the information.
The computer's OS is Windows 7. Yesterday I was prompted to download additional Windows updates. There was a very small update for Windows itself and then an optional very large update for the video card. After this update I had to restart the computer.
Could this update for the nVidia card, which came from Windows, be the reason why the computer cannot now fetch new work?
Can't get new GPU tasks
)
Upgrade your driver to 301.x or downgrade to 290.53. Your driver version 295.73 is buggy and can't run our tasks reliably.
BM
BM
Thank you. You are right,
)
Thank you.
You are right, I have driver 295.72. My nVidia Update panel is offering me 296.10. Should I download this?
For Einstein@home, that won't
)
For Einstein@home, that won't work either. You need to proceed to (at least) 301.10 or downgrade to 290.53, as Bernd already stated.
Gruß,
Gundolf
Computer sind nicht alles im Leben. (Kleiner Scherz)
I've installed 301.24. The
)
I've installed 301.24. The reason why this wasn't offered automatically for my GTX 560 Ti is probably because for this type of card it's classed as a beta version (for cards in the 600 series it's the standard upgrade). Einstein likes it! I suddenly have a screenful of new tasks and will just need to check that the first one processes properly.
Vielen Dank
Mo
RE: I've installed 301.24.
)
That would be why. nVidia releases beta drivers much more frequently than official ones because official versions are sent through and pass MS WHQL validation. The latter takes additional time and money, so it's not done for minor updates.