Does BRP5 generate more errors than BRP4?
30 Jun 2013 0:15:04 UTC
Topic 197035
(moderation:
I think am getting more errors on BRP5.
Typically the tasks end with
... Error during CUDA device->host time series length transfer (error: 700)
Looking at the tasks that error, I see that almost always, one or more wing-men have also errored out.
for example
http://einsteinathome.org/workunit/167795075
http://einsteinathome.org/workunit/167994884
Is this just co-incidence or is it perhaps a consequence of longer tasks?
Looking at my Pending tasks, which have not errored, it seems less frequent that wing-men error out,
although here
http://einsteinathome.org/workunit/168183349
is an exception.
Does BRP5 generate more errors than BRP4?
)
On reflection, I changed the nVidia drivers around the same time as BRP5 started to 319.17 which meant I changed from Cuda 5.0 to 5.5.
I will reverse back to 319.32 (latest Cuda 5.0 update) that and see if that improves things.
I will need a few days to see if that has made a difference.
I#m not sure, if this is
)
I#m not sure, if this is related to your problem, but check out this thread.
http://einsteinathome.org/node/197031
Thanks - I seem to have found
)
Thanks - I seem to have found at least one common cause for my problem, and i can eventually force it to error a task.
I just noticed the errors seemed to be occurring at similar times, and only on the GPU which was running a monitor. (I run two gtx460s)
If i logout / switch user a few times, eventually - one or two tasks will error.
I guess longer tasks mean that more tasks (as a percentage) will get caught by this.
I´m open to any ideas at this point, but i´m thinking i need to try to find a way to stop X resetting the display.