cuda-app does not start
4 May 2010 12:11:48 UTC
Topic 194906
(moderation:
I have several Einstein-workunits, one of them is ABP2cuda23. This workunit does not start. Waiting to run.
At this moment, I have no other cuda-wus on my machine, so: ???
CPU-wus are working normal.
I have updatet to the last BOINC-Client, but no change in this behaviour.
Config: i7/WinVista64, Boinc 6.10.51 (Boinc 6.10.48)
cuda-app does not start
)
Since your computers are hidden, I can't tell, but probably your CUDA adapter doesn't have enough available memory. You could do an advanced forum search for cpu_sched_debug over at least 3 months to find other threads about that topic.
Gruß,
Gundolf
Computer sind nicht alles im Leben. (Kleiner Scherz)
RE: You could do an
)
But for that this is different in 6.10.51: he needs and on then.
Ok, its a problem of memory.
)
Ok, its a problem of memory. No entry in the log, but on the status of this workunit, I can read it sometimes: Waitung for GPU memory (Most of the time there is a 'ready to run' text)
But: Where can "I" fix this? Is there no interface between BOINC and the project-servers about GPU-Memory?
My GPU:
04.05.2010 12:28:48 NVIDIA GPU 0: GeForce GT 140 (driver version 19745, CUDA version 3000, compute capability 1.1, 496MB, 208 GFLOPS peak)
Which chance do I have, to crunch this wu?
Perhaps you'll find some
)
Perhaps you'll find some hints in Mac 10.6 CUDA app "Ready to Start" not starting. The problem is not Mac-specific and the thread contains some other links.
Gruß,
Gundolf
Computer sind nicht alles im Leben. (Kleiner Scherz)
RE: My GPU: 04.05.2010
)
The 197.45 drivers do something to the memory detection on the card. It's been reported all over the place that people install 197.45 and see their video card have less reported memory all of a sudden. This isn't a BOINC problem, but a drivers problem. You may want to roll back to an earlier drivers version that reports all memory on the card.
Although it can also be that something is stuck in the memory of your video card. This takes up memory and the only fail-safe method of removing that is a full power-cycle, or a reboot of the machine.
@Jord: Thank you for the
)
@Jord: Thank you for the answer!
I know of the driver-problem with an older level of BOINC. But with 6.10.48 other cuda-apps are working again.
Now, I stayed at 6.10.51, but changed the nvidia-driver:
- 197.57 (beta): waiting for GPU-memory
- 197.13: waiting for GPU-memory
Both constellations do not work. Maybe I have to restart my machine? --> Later!
DNETC-cuda-apps and GPUGRID-cuda-apps are working with 197.45 and Boinc >= 6.10.48, so: Where is the difference?
Ralf.
RE: DNETC-cuda-apps and
)
The difference is probably the amount of memory needed. Einstein needs about 400 to 430 MiB.
Gruß,
Gundolf
Computer sind nicht alles im Leben. (Kleiner Scherz)
Wow, during the last hours,
)
Wow, during the last hours, the cuda-app has done 36% of its WU.
actual settings:
NVIDIA-Driver 197.13 with Boinc 6.10.51
Maybe I will give 197.51 a try?
RE: Wow, during the last
)
Forget this post...
I gave GPUGRID a chance to use the free cuda-time. It is this WU of GPUGRID, which has 36% done.
Einstein is at waiting state all the time.
When I look at my system:
Graphics Adapter with 512 MB Ram, Boinc reports it at 496 MB, I have two monitors running, one of them at 1920*1200 pixel.. I can't believe, that there is enough ram for Einstein's cuda-app.
Maybe I should cancel this and switch off Cuda-apps at einstein@home, if possible.