Hello. I'm having a problem with BOING MANAGER 6.12.34 (x64) and SETI@home enanched 6.08 (cuda)

I configured it to run when computer is not in use. Configuration as in the images below:

http://img12.imageshack.us/img12/735/preferencesg.gif
http://img341.imageshack.us/img341/8059/configuration.gif

It happens sometimes that I'm working at the keyboard and the BOINC software is working. I'm not sure if it starts when I'm still working or if it does not stop after I'm back at the keyboard after some inactivity. It's annoying as it use a lot of computer resources and I can't work with BOING running. It keep working for long time, sometimes I had to kill it after 30 minutes. I think there's something wrong somewhere....

http://img833.imageshack.us/img833/1084/projects.gif

If upu look at the image above you can see it looks "suspended" but it is working! GPU is at 98% and progress bars (elapsed, estimated) of that project are moving...

Any suggestion ?

i think you should restart your computer and then run boinc again, i have had this problem too(sort of, i suspended al my tasks but still there were tasks computing thinks, i don't know if the percentage was increasing but when i looked at my taskmanager(in windows) it showed some proccesses from my projects running at full speed(like if they weren't suspended))
so i just suggest that you restart your computer, that will make all the running boinc related proccesses stop and restart at the computer restart

reply if it worked

It seems it is enough to stop and restart the BOING task.

Nasty bug anyway. I see on the forum many of us are experiencing the same problem. I'm not even sure if it is a BOINC problem or a specific application problem. Probably the second one, related to GPU computing.

Is there any developer on this board to confirm you're aware of this problem ?

I see on the forum many of us are experiencing the same problem.

Which forum? Care to give a link?

---

Which forum? Care to give a link?

On this forum:

http://boinc.berkeley.edu/dev/forum_thread.php?id=7354
http://boinc.berkeley.edu/dev/forum_thread.php?id=7085

Since 6.12 isn't being worked on anymore, test with BOINC 7, please. The latest download links are in the BOINC 7 Change Log and News thread, version 7.0.20.

---

Installed 7.0.20 (x64) - same problem

The developers are looking into it.

---

Great. In the meantime, I investigated a little bit more about the gpu-non-suspending problem. It appears that the problem is known since 2009:

http://setiathome.berkeley.edu/forum_thread.php?id=52971

The discussion is a bit old but seems to be still up date. If I'm understanding correcly, the problem is related to the algorithm used in the SETI application combined to some "limits" of the gpu computation. In short, SETI application sometimes starts a long analysis that cannot be stopped until it is completed.

It seems to me, to stop the computation, the faster and simpliest solution is to terminate BOINC nd restarting it again. I see that BOINC has a tray menu option that allow to stop gpu work http://img805.imageshack.us/img805/8716/snoozegpudoesnotwork.gif but this option is not working as well. Task will continue to run even selecting that option.

Simple soluction: what about adding an extra tray menu option to terminate gpu work ? It should not be too hard to encode and it could be an acceptable workaround while a better soluction is released...

Great. In the meantime, I investigated a little bit more about the gpu-non-suspending problem. It appears that the problem is known since 2009:

http://setiathome.berkeley.edu/forum_thread.php?id=52971

The discussion is a bit old but seems to be still up date. If I'm understanding correcly, the problem is related to the algorithm used in the SETI application combined to some "limits" of the gpu computation. In short, SETI application sometimes starts a long analysis that cannot be stopped until it is completed.

It seems to me, to stop the computation, the faster and simpliest solution is to terminate BOINC nd restarting it again. I see that BOINC has a tray menu option that allow to stop gpu work http://img805.imageshack.us/img805/8716/snoozegpudoesnotwork.gif but this option is not working as well. Task will continue to run even selecting that option.

Simple soluction: what about adding an extra tray menu option to terminate gpu work ? It should not be too hard to encode and it could be an acceptable workaround while a better soluction is released...

Unfortunately, although that sounds like a simple solution on the surface, in practice it'll cause more problems than it solves.

Especially with modern NVidia drivers, abruptly terminating a CUDA application like that can trigger a protective reaction from the video driver, slowing the card down to a minimal clock rate - from which it can't recover without rebooting the computer. We'll have to be a bit more subtle than that.

Oh, I understood ! Things are never easy as they seem to be !

Another thought... what about dividing CUDA applications in two groups ? CUDA and HEAVY CUDA, where heavy cuda are applications like the SETI one that cannot stop on the fly ?

With a simple setting we could decide which one to run on a given computer (checkbox activate heavy cuda). Right now the option is to keep gpu off. That's a pity. My pc is on 12 hours a day but I'm working at the keyboard for a total of 5 or 6. When I'm at the keyboard I need it all for me, but during the day I eat, go out, talk at the phone, go toilet, stalk secretaries, ecc. Basically it is more the time the computer can work for science than the time I really need it. During the day time it could do light gpu works. Sometimes, when I leave it on during the night or I'm out all the day, I could activate the HEAVY CUDA mode....

Advanced settings could include a schedule where cuda computing is allowed, days and hours (example: a computer that is used 8-17 by humans but it's on 24h a day could work from 17.30 to 7.30) - best of all would be to have the possibility to limit the % of gpu usage. I really know a few about cuda computing, but I guess it is EXTREMELY hard to manage.

Could you please suggest me some projects that use light gpu cuda applications ? I want to check if I can manage them during my working day with the current settings (run after 1 min of inactivity).

Can you post your Boinc startup messages please, at the moment we don't know what sort of PC you have, what GPU you have, or even what Graphics drivers you're running,

Claggy

I'm not sure what you mean for startup message. Please have a look at the pictures below and let me know if you need some more information (example: logs from boinc, in this case please instruct me where to find them or a specific speccy screenshot)

http://img193.imageshack.us/img193/7238/20120316mycomputerbyboi.gif
http://img828.imageshack.us/img828/8858/20120316mycomputerbyspe.gif

O.K, that'll do, what i wanted to see was your host ID, now i can see what GPU you have, how much RAM it has, and what drivers it has:

Computer 6380347

You have a few things against you, you only have a GPU with 256Mb, the Seti app needs about 220Mb free RAM to operate, if there isn't enough it'll fall into CPU fallback mode, and basically use a core, this is what your tasks are doing:

setiathome_CUDA: Found 1 CUDA device(s):
Device 1 : GeForce 8500 GT
totalGlobalMem = 268435456
sharedMemPerBlock = 16384
regsPerBlock = 8192
warpSize = 32
memPitch = 2147483647
maxThreadsPerBlock = 512
clockRate = 918000
totalConstMem = 65536
major = 1
minor = 1
textureAlignment = 256
deviceOverlap = 1
multiProcessorCount = 2
setiathome_CUDA: CUDA Device 1 specified, checking...
Device 1: GeForce 8500 GT is okay
SETI@home using CUDA accelerated device GeForce 8500 GT
setiathome_enhanced 6.03 Visual Studio/Microsoft C++
libboinc: 6.3.22

Work Unit Info:
...............
WU true angle range is : 0.420463
Optimal function choices:
-----------------------------------------------------
name
-----------------------------------------------------
v_BaseLineSmooth (no other)
v_GetPowerSpectrum 0.00019 0.00000
v_ChirpData 0.01611 0.00000
v_Transpose4 0.01692 0.00000
FPU opt folding 0.00341 0.00000
Cuda error 'cudaMalloc((void**) &dev_t_funct_cache' in file 'c:/sw/gpgpu/seti/seti_boinc/client/cuda/cudaAcc_pulsefind.cu' in line 662 : out of memory.
setiathome_CUDA: CUDA runtime ERROR in device memory allocation (Step 3 of 3). Falling back to HOST CPU processing...
setiathome_CUDA: Found 1 CUDA device(s):
Device 1 : GeForce 8500 GT
totalGlobalMem = 268435456
sharedMemPerBlock = 16384
regsPerBlock = 8192
warpSize = 32
memPitch = 2147483647
maxThreadsPerBlock = 512
clockRate = 918000
totalConstMem = 65536
major = 1
minor = 1
textureAlignment = 256
deviceOverlap = 1
multiProcessorCount = 2
setiathome_CUDA: CUDA Device 1 specified, checking...
Device 1: GeForce 8500 GT is okay
SETI@home using CUDA accelerated device GeForce 8500 GT
Restarted at 0.22 percent.
Cuda error 'cudaMalloc((void**) &dev_PoTPrefixSum' in file 'c:/sw/gpgpu/seti/seti_boinc/client/cuda/cudaAcceleration.cu' in line 342 : out of memory.
setiathome_CUDA: CUDA runtime ERROR in device memory allocation (Step 1 of 3). Falling back to HOST CPU processing...setiathome_CUDA: Found 1 CUDA device(s):
Device 1 : GeForce 8500 GT
totalGlobalMem = 268435456
sharedMemPerBlock = 16384
regsPerBlock = 8192
warpSize = 32
memPitch = 2147483647
maxThreadsPerBlock = 512
clockRate = 918000
totalConstMem = 65536
major = 1
minor = 1
textureAlignment = 256
deviceOverlap = 1
multiProcessorCount = 2
setiathome_CUDA: CUDA Device 1 specified, checking...
Device 1: GeForce 8500 GT is okay
SETI@home using CUDA accelerated device GeForce 8500 GT
Restarted at 0.73 percent.
Cuda error 'cudaMalloc((void**) &dev_t_funct_cache' in file 'c:/sw/gpgpu/seti/seti_boinc/client/cuda/cudaAcc_pulsefind.cu' in line 662 : out of memory.
setiathome_CUDA: CUDA runtime ERROR in device memory allocation (Step 3 of 3). Falling back to HOST CPU processing...setiathome_CUDA: Found 1 CUDA device(s):

I suggest you just don't do Cuda computation on your GPU unless you can free some GPU RAM up by disabling some Aero effects,

Note: The 295.73 drivers have a Bug (I don't see any evidence of your host being effected through), when the Monitor connected to the DVI port goes to sleep the Cuda device becomes unavailable,
if you get that output in your tasks, set your monitor not to sleep, and phyiscally turn it off, or downgrade your drivers 290.53 or earlier.

Claggy

Ok. Thank you for your help.

I'll try to free some memory. I'm not that great fan of the windows 7 aero effects.

Never experienced that diver bug. Monitor is connected to the vga port.