Unusually Fast O1 All Sky Completions on Problem Machine
6 May 2016 18:30:17 UTC
Topic 198599
(moderation:
Hello,
I have a dual E5-2670 machine with an AMD 7970 inside. The 7970 had previously been giving me frequent validate errors and invalid results on BRP6 tasks. I had it fixed by underclocking it, but the problem started again.
At the same time, I noticed that many of my O1 all-sky tasks were taking about 30% longer than normal (normal time, ~55k seconds). And a few took only 15k seconds, 350% faster than normal, yet were validated as good results. I thought perhaps these were "small" tasks, but comparing times to my wingmen, their times appear normal for their CPUs. I am posting here so these can be examined to be sure the validator is working correctly.
Sample works units completed 350% faster than normal, yet validated:
244989928, 244286371, 244989703, 244581088, 244989369, 244989364, 244273986, 244970188, 244970148
The computer is 12232333. O1 tasks are run hyperthreaded. I have removed the AMD card and the O1 result times seem to be normalizing again.
Unusually Fast O1 All Sky Completions on Problem Machine
)
Looking at Task 557462853
We see
2016-05-05 11:26:57.4354 (5688) [debug]: Successfully read checkpoint:24326So it looks like the task times reported are wrong, after restarting from a checkpoint. I looked at a couple of others and they also restart from checkpoint at that time.
I'm not 100% sure but there is something for the devs to look at.
Yes, I saw something similar
)
Yes, I saw something similar too from the dedicated cluster machines. That after a checkpoint the runtime is also zeroed and reported incorrectly. I wonder if this is a problem with a specific Client version.