Oops, sorry :-) When you start the BOINC client it'll list all GPUs in the event log (advanced view). For AMD/ATI devices it might talk about CAL and OpenCL - we're interested only in the latter. You should find the list of GPUs more or less at the top of the event log, right before the registered projects are mentioned.
My first OpenCL WU with the v1.22 app validated aigainst a CUDA result. No problem (see this Task). However, I believe that I am using an earlier version of Catalyst (< 12.1). I'll have to check that when I'm home from work.
This happens in 7.0.18:
The scheduler requests new jobs, and 3 seconds later it starts S6LV1. I can not imagine that the download of the task is completed then? Normally we can see something like this
Started download of p2030.20111110.G39.19-00.79.N.b2s0g0.00000_3648.binary
Finished download of p2030.20111110.G39.19-00.79.N.b2s0g0.00000_3648.binary
first.
01-Mar-2012 08:39:39 [Albert@Home] Sending scheduler request: To fetch work.
01-Mar-2012 08:39:39 [Albert@Home] Reporting 4 completed tasks, requesting new tasks for CPU
01-Mar-2012 08:39:48 [Albert@Home] Scheduler request completed: got 1 new tasks
01-Mar-2012 08:39:48 [Albert@Home] Resent lost task h1_0059.95_S6GC1__39_S6LV1A_1
01-Mar-2012 08:39:51 [Albert@Home] Starting task h1_0059.95_S6GC1__39_S6LV1A_1 using einstein_S6LV1 version 110 (SSE2) in slot 10
01-Mar-2012 08:39:52 [Albert@Home] Computation for task h1_0059.95_S6GC1__39_S6LV1A_1 finished
01-Mar-2012 08:39:52 [Albert@Home] Output file h1_0059.95_S6GC1__39_S6LV1A_1_0 for task h1_0059.95_S6GC1__39_S6LV1A_1 absent
01-Mar-2012 08:41:39 [Albert@Home] Sending scheduler request: To fetch work.
01-Mar-2012 08:41:39 [Albert@Home] Reporting 1 completed tasks, requesting new tasks for CPU
01-Mar-2012 08:41:41 [Albert@Home] Scheduler request completed: got 4 new tasks
01-Mar-2012 08:41:43 [Albert@Home] Starting task h1_0059.95_S6GC1__35_S6LV1A_1 using einstein_S6LV1 version 110 (SSE2) in slot 10
01-Mar-2012 08:41:43 [Albert@Home] Starting task h1_0059.95_S6GC1__33_S6LV1A_1 using einstein_S6LV1 version 110 (SSE2) in slot 11
01-Mar-2012 08:41:43 [Albert@Home] Starting task h1_0059.95_S6GC1__34_S6LV1A_1 using einstein_S6LV1 version 110 (SSE2) in slot 12
01-Mar-2012 08:41:44 [Albert@Home] Computation for task h1_0059.95_S6GC1__35_S6LV1A_1 finished
01-Mar-2012 08:41:44 [Albert@Home] Output file h1_0059.95_S6GC1__35_S6LV1A_1_0 for task h1_0059.95_S6GC1__35_S6LV1A_1 absent
01-Mar-2012 08:41:44 [Albert@Home] Starting task h1_0059.95_S6GC1__36_S6LV1A_1 using einstein_S6LV1 version 110 (SSE2) in slot 10
01-Mar-2012 08:41:45 [Albert@Home] Computation for task h1_0059.95_S6GC1__33_S6LV1A_1 finished
01-Mar-2012 08:41:45 [Albert@Home] Output file h1_0059.95_S6GC1__33_S6LV1A_1_0 for task h1_0059.95_S6GC1__33_S6LV1A_1 absent
01-Mar-2012 08:41:46 [Albert@Home] Computation for task h1_0059.95_S6GC1__34_S6LV1A_1 finished
01-Mar-2012 08:41:46 [Albert@Home] Output file h1_0059.95_S6GC1__34_S6LV1A_1_0 for task h1_0059.95_S6GC1__34_S6LV1A_1 absent
01-Mar-2012 08:41:47 [Albert@Home] Computation for task h1_0059.95_S6GC1__36_S6LV1A_1 finished
01-Mar-2012 08:41:47 [Albert@Home] Output file h1_0059.95_S6GC1__36_S6LV1A_1_0 for task h1_0059.95_S6GC1__36_S6LV1A_1 absent
The scheduler requests new jobs, and 3 seconds later it starts S6LV1.
Well, S6LV1 tasks are not the same as BRP tasks. S6LV1 tasks re-use data already present on your host while BRP data is only used once, for a single WU. Looking at the error output of the failed S6LV1 task should tell us what happened.
Please open another thread for that problem if it persists. This thread is meant to discuss BRP v1.22 only.
Oops, sorry :-) When you start the BOINC client it'll list all GPUs in the event log (advanced view). For AMD/ATI devices it might talk about CAL and OpenCL - we're interested only in the latter. You should find the list of GPUs more or less at the top of the event log, right before the registered projects are mentioned.
Oliver
He may need to uninstall the drivers, run driver sweep, then reinstall the 12.1 drivers. Leaving old drivers wreaks havoc on the OpenCL apps at Seti. Probably the same here. His Card is recognized as a 6900 series and he is running the 7.0.18 BOINC so that isn't a problem.
No it doesn't. No GPUs are being used on T4T or the Vboxwrapper test project (the only two projects at this time where VBox is being used), other than for showing graphics of sorts. And then these projects require Vbox 4.1.4 or higher, as far as I know.
Of course, it's because BOINC thinks that the 205 hours it's estimated to go do is really 205h / (39 / 100) = 525h (or almost 22 days). A tad difficult to do in 14 days. So it'll run from start to finish in high priority. And as we can see in here, DCF is no longer really used with Boinc 7. Not that it matters, DCF is 7.5, way too high to use reliably.
So, pretty please, can the fpops estimate be adjusted enough that they don't come in thinking to take 200+ hours?
Since v1.21/v1.22 update, I have got a lot of validation errors.
Please check my stats/failed units, maybe that will help you to find a bug.
My system is BOINC 7.18 + Mac OS X 10.7.3 + AMD HD6750M 1GB
No X windows here, I'm
)
No X windows here, I'm running win7.
Is there somewhere in the logs that will show me if the card is enumerated correctly?
I re-installed the 12.1 driver yesterday and did a custom install instead of express. Everything appeared to be checked.
Its 2am here and I'm sleep typing, so I'll check when I wake up. :P
RE: No X windows here, I'm
)
Oops, sorry :-) When you start the BOINC client it'll list all GPUs in the event log (advanced view). For AMD/ATI devices it might talk about CAL and OpenCL - we're interested only in the latter. You should find the list of GPUs more or less at the top of the event log, right before the registered projects are mentioned.
Oliver
My first OpenCL WU with the
)
My first OpenCL WU with the v1.22 app validated aigainst a CUDA result. No problem (see this Task). However, I believe that I am using an earlier version of Catalyst (< 12.1). I'll have to check that when I'm home from work.
mickydl*
This happens in 7.0.18: The
)
This happens in 7.0.18:
The scheduler requests new jobs, and 3 seconds later it starts S6LV1. I can not imagine that the download of the task is completed then? Normally we can see something like this
first.
RE: The scheduler requests
)
Well, S6LV1 tasks are not the same as BRP tasks. S6LV1 tasks re-use data already present on your host while BRP data is only used once, for a single WU. Looking at the error output of the failed S6LV1 task should tell us what happened.
Please open another thread for that problem if it persists. This thread is meant to discuss BRP v1.22 only.
Cheers,
Oliver
RE: RE: No X windows
)
He may need to uninstall the drivers, run driver sweep, then reinstall the 12.1 drivers. Leaving old drivers wreaks havoc on the OpenCL apps at Seti. Probably the same here. His Card is recognized as a 6900 series and he is running the 7.0.18 BOINC so that isn't a problem.
Could running vbox be a problem?
RE: Could running vbox be
)
Don't know but could be if vbox acquires the GPU somehow...
Oliver
No it doesn't. No GPUs are
)
No it doesn't. No GPUs are being used on T4T or the Vboxwrapper test project (the only two projects at this time where VBox is being used), other than for showing graphics of sorts. And then these projects require Vbox 4.1.4 or higher, as far as I know.
I'll go with driver corruption as well. It certainly never hurts to completely clean out previous drivers and then reinstall any later as new.
Tasks still come in expecting
)
Tasks still come in expecting to run for 205 hours.
So I still have Albert tasks running in a panic.
03/03/2012 02:47:11 | Albert@Home | [rr_sim] Result p2030.20111110.G39.19-00.79.N.b3s0g0.00100_864_3 projected to miss deadline.
03/03/2012 02:47:11 | Albert@Home | [rr_sim] Project has 1 projected ATI deadline misses
03/03/2012 02:47:31 | Albert@Home | [rr_sim] p2030.20111110.G39.19-00.79.N.b3s0g0.00100_864_3 misses deadline by 614511.77
Of course, it's because BOINC thinks that the 205 hours it's estimated to go do is really 205h / (39 / 100) = 525h (or almost 22 days). A tad difficult to do in 14 days. So it'll run from start to finish in high priority. And as we can see in here, DCF is no longer really used with Boinc 7. Not that it matters, DCF is 7.5, way too high to use reliably.
So, pretty please, can the fpops estimate be adjusted enough that they don't come in thinking to take 200+ hours?
Since v1.21/v1.22 update, I
)
Since v1.21/v1.22 update, I have got a lot of validation errors.
Please check my stats/failed units, maybe that will help you to find a bug.
My system is BOINC 7.18 + Mac OS X 10.7.3 + AMD HD6750M 1GB