197 (0x000000C5) EXIT_TIME_LIMIT_EXCEEDED
11 Dec 2016 10:52:21 UTC
Topic 203608
(moderation:
Can you increase the bloody time for task? I mean what the hell is that anyway. Its LONG before task deadline when it exits with this error. I've searched around other projects a bit and this is clearly wrong project setting, most likely in the past you increased the task deadline because of larger tasks but left this setting in place.
The problem is not the time,
)
The problem is not the time, the problem is your (lack of) speed. I mean, why can't your GPU reliably finish a BRP4G task in more than 32 hours? Sure, an AMD Juniper is old, but it shouldn't be that slow. Let's take a look at your tasks:
https://einsteinathome.org/de/host/12390379/tasks
First, you seem to run multiple tasks on that card. Don't do that and you should be fine. Second, your CPU time is very low and the run times vary a lot. That could mean your CPU is overloaded. Don't run more than 7 CPU tasks on that machine, less if there is other load. Then let's see if things improve.
The exit code 197
)
The exit code 197 (EXIT_TIME_LIMIT_EXCEEDED) is as floyd said only triggered if the computation is taking longer than expected. It has nothing to do with the deadline. This limit is there so that wrongly configured tasks don't run indefinitely but get aborted at some point. Your 3 BRP4G tasks running for 32 hours is such a case where the BOINC Client needs to abort. I've seen this for CPU applications where the CPU is busy with non-BOINC work and the BOINC processes running at lowest priority in the background don't get enough CPU cycles. Maybe something similar is happening for you where you were using the GPU for other things of the GPU downclocked because of heat issues. You should check the temperature of the GPU and also the fan and exhaust.
I suspect the problem may
)
I suspect the problem may simply be that the GPU tasks are taking a lot longer than they should because all CPU cores are busy elsewhere - perhaps with Rosetta tasks or some non-BOINC load. If so, setting BOINC to use less than 100% of the CPU cores may give an improvement.
Cheers,
Gary.
From that host's latest
)
From that host's latest scheduler log:
parsed project prefs setting 'gpu_util_brp': 0.200000Wow, I hope you override that.
I wonder how many concurrent
)
I wonder how many concurrent GPU tasks are actually running? With a low end 1GB card maybe they are interfering with each other and causing the long run times. Maybe if set to 0.5, all would work properly.
Cheers,
Gary.
Three tasks because of better
)
Three tasks because of better overall GPU utilisation. Tried five but virtually every task exited with error.
And no, there is no "expected" runtime, that's BS. If it was so than the deadline had no meaning. As long as there is the deadline, no other measure is neccessary, and if so, than set to some reasonable time, not this…
For your information, HD 5750 has higher computational power than 7770 and even R7 350. So I would still consider it lowest-end mainstream card, not low-end. The real progress in performance per dollar over the past 5 years is close to none.