CUDA failes -- before and after update of BOINC and restart of project
25 Mar 2011 9:22:57 UTC
Topic 195724
(moderation:
Last week, on the 21the or March, I got a lot of errors like:
Quote:
Einstein@Home If this happens repeatedly you may need to reset the project.
Okay. I turned of downloading new WUs and waited until the last E@H, which was yesterday evening, then reported all E@H-WUs, resetted the E@H project as advised. Then stopped boinc, upgraded to version 6.10.58 restarted the whole thing and got again the same error messages.
http://einsteinathome.org/host/1967526/tasks&offset=0&show_names=0&state=5
The only success: It seems to crash faster and seems to waste fewer CPU-cycles.
Fine! I like CUDA!
What experiment is the best to get a better success?
Switch from the newest stable CUDA-Driver 260.19.44 to the older experimental version 270.26?
CUDA failes -- before and after update of BOINC and restart of p
)
That message is not the problem (and it's BS by the way). The problem is:
And, sorry, I can't tell you what to do against that.
Gruß,
Gundolf
Computer sind nicht alles im Leben. (Kleiner Scherz)
6-11% of 512MB memory is
)
6-11% of 512MB memory is used.
$ while `sleep 1`; do nvidia-smi -a| grep Memory; done Memory : 6% Memory : 6% Memory : 6% Memory : 6% Memory : 6% Memory : 9% Memory : 10% Memory : 7% Memory : 6% Memory : 9% Memory : 11% Memory : 8% Memory : 10% Memory : 8% Memory : 6% Memory : 6% Memory : 9% Memory : 9% Memory : 8% Memory : 7% Memory : 11%and I always see these values, I never see much larger ones.
RE: 6-11% of 512MB memory
)
That is not a newer model gpu, what other projects have you tried using it on? If none try Collatz or PrimeGrid and see if it works there. Seti also has a gpu app so you might try that too. I am thinking it is just not up to what Einstein wants or is looking for.
RE: Seti also has a gpu app
)
No, SETI doesn't have a stock CUDA app for Linux.
RE: RE: 6-11% of 512MB
)
This machine is a business machine. I need it for those nasty things like earning money. And it is what I got out of the box. Yes, it is close to two years old and it will still have this configuration for the next three years. Then it might be replaced.
There is only one chance for a new graphics card: When my monitor breaks and I buy a newer model which needs something like two DVI-Channels (with two monitors I would then need a card with 4 channels).
BTW: This machine runs E@H, SETI, Rectilinear Crossing (currently dead) and orbit@home (currently no work).
RE: RE: Last week, on the
)
I believe the real reason becomes obvious if you continue to read the log just few more lines.
It says that there are only 180MB of your 512MB free. If I recall the Einstein requirements correctly this is not enough. The only way I see to make it work to reduce your memory consumption.
Michael
RE: It says that there are
)
==============NVSMI LOG==============
Timestamp : Fri Mar 25 22:30:34 2011
Driver Version : 260.19.44
GPU 0:
Product Name : GeForce 9500 GT
PCI Device/Vendor ID : 64010de
PCI Location ID : 0:4:0
Board Serial : 1266914972570
Display : Connected
Temperature : 43 C
Fan Speed : 50%
Utilization
GPU : 0%
Memory : 6%
Power State : PSTATE 0
Power Capping : Disabled
If 6% of 512 MB would be 332 MB, then I would totally agree.
What could be is a continuos block of only 180 MB. Fragmented Memory ...
But whenever I tried this 'nvidia-smi -a' I have never seen any memory usage higher that 16%. And 16% ist still just 82MB Memory usage with 430 MB free.