It works!

[Jason G]

Well I waited an hour, and it never finsihed. I can certainly plug the app in again and watch the state.dat when it freezes.  It seems to be a regular occurance at least with this particular app, and Boinc combination (And I do suspect Boinc's scheduling behaviour is to blame pretty strongly).

Shouldn't take long.

[Raistmer]

I need to know if "refuse to die" mod works or not. If it not work indeed I remove it from CUDA MB.

So I need next info on this topic:

1) When CUDA MB processed task enters in "waiting to run" mode what tasks are in "running" mode ?
2) What access state of GPU_lock file (is it accessible or not)
3) Is any progress inside state.sah of task being in "waiting" state.
4) How many CUDA MB processes running in system at this time
5) temp of GPU (busy or idle).

[Jason G]

I can answer most of those now, will update remaining ones at next lockup

....
So I need next info on this topic:

1) When CUDA MB processed task enters in "waiting to run" mode what tasks are in "running" mode ?
    2 x AP 5.03 on CPU cores
2) What access state of GPU_lock file (is it accessible or not)
     ...
3) Is any progress inside state.sah of task being in "waiting" state.
     ...
4) How many CUDA MB processes running in system at this time
     1 on only GPU, 'Waiting to Run'
5) temp of GPU (busy or idle).
    Busy, Full load temp  ~75C  ( not overheating for this GPU either, no screen jerkyness or other apparent issue, can still run ATI Tool with no Artifacts)

[Raistmer]

4) How many CUDA MB processes running in system at this time
     1 on only GPU, 'Waiting to Run'

I meant OS processes, not BOINC ones.

And it seems you don't use ncpus field? why only 2 AP tasks on dual core + GPU ?

[Jason G]

Read original my post again.  I tried both ways.  If you want me to do this with ncpus=3 .. will do.

[Geek@Play]

1) When CUDA MB processed task enters in "waiting to run" mode what tasks are in "running" mode ?
   At this time I would have 4 AstroPulse work units running on 4 CPU cores.

2) What access state of GPU_lock file (is it accessible or not)
   I do not have this file or where it is at.  Only have "boinc_lockfile" located in slots folders.

3) Is any progress inside state.sah of task being in "waiting" state.
   Included here are 2 state files captured several minutes apart.

4) How many CUDA MB processes running in system at this time
   Only the one (1) that is showing "waiting to run" in Boinc Manager.

5) temp of GPU (busy or idle).
   65C busy, 56C idle.

[attachment deleted by admin]

[Jason G]

Okay,

 switched to ncpus = 3, and whammo! .. stuck at 1.712% ... ( I reckon some kindof oversubscription mechanism is at play here )
[Filling in answers as I gather the data]
[Edit:] Now while collecting info, this one jumped from 1.712% to 100%, looks completed
 - AR is 0.443445
-  Claimed credit says ~1.94 credit 
-  wall time in stderr ~18mins 20 secs. (~Typical full length run)
http://setiathome.berkeley.edu/result.php?resultid=1172313302

1) When CUDA MB processed task enters in "waiting to run" mode what tasks are in "running" mode ?
  3 x AP 5.03 (1 per CPU core +1 )
2) What access state of GPU_lock file (is it accessible or not)
Opens as empty file in notepad without complaint.
3) Is any progress inside state.sah of task being in "waiting" state.
  Not visibly, though timestamp was incrementing every checkpoint period etc
4) How many CUDA MB processes running in system at this time
Only the 1 'Waiting to run"
5) temp of GPU (busy or idle).
Busy 75C (normal), if run ATITool at same time, no artefacts.

[Raistmer]

1) When CUDA MB processed task enters in "waiting to run" mode what tasks are in "running" mode ?
   At this time I would have 4 AstroPulse work units running on 4 CPU cores.

2) What access state of GPU_lock file (is it accessible or not)
   I do not have this file or where it is at.  Only have "boinc_lockfile" located in slots folders.

3) Is any progress inside state.sah of task being in "waiting" state.
   Included here are 2 state files captured several minutes apart.

4) How many CUDA MB processes running in system at this time
   Only the one (1) that is showing "waiting to run" in Boinc Manager.

5) temp of GPU (busy or idle).
   65C busy, 56C idle.

state.sah1
<prog>0.23420485</prog>

state.sah2
<prog>0.29913810</prog>

That is, task in that slot continue to make progress. It seems all as intended to be.
Now it would be interesting to watch until <prog> value reaches to 1 - what will be with BOINC in this case.

[Jason G]

Okay, so why the massive 1.94 credit claim on my full length no progress run ? (See updated run info)

It seems there are no function calls to accumulate the flops while it is running. (compute_fraction_done or whatever Cuda app uses) [Edit: It will be interesting to see what the Wingman Claims).

[Geek@Play]

Ageless posted here............

http://setiathome.berkeley.edu/forum_thread.php?id=52090&nowrap=true#869394

This may be a Boinc problem?  If so my apologies to Raistmer.

[Jason G]

Ageless posted here............

http://setiathome.berkeley.edu/forum_thread.php?id=52090&nowrap=true#869394

This may be a Boinc problem?  If so my apologies to Raistmer.

That was the first suspicion I mentioned (Boinc cpu scheduling foulups) But after experiencing it I'm not so sure.  It may simply be not receiving the updated flops counts for whatever reason.  Hopefully it is their problem not ours 

[Raistmer]

Okay, so why the massive 1.94 credit claim on my full length no progress run ? (See updated run info)

It seems there are no function calls to accumulate the flops while it is running. (compute_fraction_done or whatever Cuda app uses) [Edit: It will be interesting to see what the Wingman Claims).

LoL  You know , I never cared about credits much
But it's interesting indeed why it can't accumulate flops in this case... very interesting.
But except credit question all just work as intended! BOINC tried to leave your GPU IDLE (it started only CPU apps ) but CUDA app resisted and made use GPU at least until it finished its work. All possibilities to keep GPU busy were used!

[Jason G]

LoL... Yeah, something's busted though for this weird behaviour.  Boinc or app doesn't matter.

Running 3 AP's has my system rather oversubscribed it seems, so throughput was quite a bit better running 2 APs rather than 3.

Will be reverting to my 'classic' setup, and if that doesn't work, will consider pawning my PC for beer money    [Damn it's working no beer for me  ]

Jason

[Richard Haselgrove]

[Damn it's working no beer for me  ]

Jason

I'm sure we can sort something out if you join Matt on his tour of Lancashire, Yorkshire, Nottinghamshire, Cambridgeshire....

[Jason G]

I'm sure we can sort something out if you join Matt on his tour of Lancashire, Yorkshire, Nottinghamshire, Cambridgeshire....

If that's the tour where we all travel through townships cross country in a yellow combi-van wearing poncho's and funny hats, stopping at every pub along the way, Count Me In!