New data file for the FGRPB1G search
6 Sep 2018 3:21:33 UTC
Topic 216278
(moderation:
The very recent outage of FGRPB1G tasks has been solved with the release of tasks for a new data set - LATeah0104M.dat. There are a couple of noticeable differences in crunching performance when compared to tasks for the previous data - LATeah1019L.dat - which had only started being used itself, about 4 or so days previously.
The first difference is that crunch times are around 25% faster with the new data. I guess that just means there is less parameter space being processed for each task in this new set. I'm sure it doesn't mean that your GPU has suddenly 'improved' by a whopping 25% :-).
The other interesting difference is that there is now an obvious follow-up stage where double precision is used to refine a 'toplist' of the 10 most likely candidate signals. This is just like it used to be quite some time ago. The new tasks once again show a distinct pause at 89.997% estimated completion which lasts perhaps 20 seconds on the couple of tasks I've observed. The previous tasks had virtually no pause at all at 89.997% before jumping straight to 100% and then uploading.
I'm posting this for the benefit of new volunteers who might not know about the follow-up stage. I saw perhaps 20s on a reasonably quick GPU. It could well be longer on less capable hardware. Hopefully the observation might help to allay concern if the crunching seems to stop at ~90% for an otherwise unknown reason.
Cheers,
Gary.
Maybe that could be posted as
)
Maybe that could be posted as a News. Would show it to many more people and also would make the project look more active! Sure it's better to read this on the homepage!
As for myself, with an AMD RX 580 on Windows 10 and the 18.8.2 drivers, a WU now completes in like 7:40 minutes. The 89.997% pause lasts like 16 seconds or so. The older WU took about 9:40 minutes.
Interesting, I am receiving a
)
Interesting, I am receiving a bunch of L wus, mixed with Q ones.
Over the last 2 weeks since
)
Over the last 2 weeks since tasks for "LATeah1019L.dat" finished, we've been processing tasks for LATeah0104?.dat where '?' has changed progressively from 'M' through to 'Q'. The server always issues the two original copies of all tasks in a particular set at pretty much the same time and doesn't seem to 'mix' original tasks for a following set when it still has original tasks for a previous set.
The 'mixing' is simply a function of the need to issue further copies of a previous task to replace the ones that fail in some way or miss the deadline. Right now, we are in 'deadline miss' territory so can expect to see replacements being issued for all the five 0104? tasks over the next two weeks with a further much smaller group from two weeks to four weeks (resends of the resends :-) ).
The new 1021L tasks would seem to be pretty much the same as the 1019L tasks - much longer running than the recent tasks with very little, if any, 'follow-up stage' in evidence. Interestingly, we seem to have missed out on a set labeled '1020L', at least for the moment.
Cheers,
Gary.
I really liked 25 min run
)
I really liked 25 min run times running 2 at a time vs 38 min. My RAC went up by ~150K now I shall watch it go down.
Thank you for the
)
Thank you for the explanation.
Well credits are a secondary thing, as long as WU are useful!
And RAC is starting to drop.
)
And RAC is starting to drop.
RAC is continuing to drop as
)
RAC is continuing to drop as expected, I look foreword to a RAC of less than 400K. What's not to like? Life is good.
I just broke 500K RAC 400 is
)
I just broke 500K RAC, 400 is on the horizon. These things do not pay well.
RAC is falling quite nicely.
)
RAC is falling quite nicely. It will continue for at least 3 weeks unless the data changes.
Look on the bright side. If
)
Look on the bright side. If they hadn't had that data issue some weeks ago, you would have just had the same old stuff you're doing now, all the way through.
At least the plate tectonics has created the Himalayas plus a couple of Mt Everests and the landscape isn't quite so boring as it otherwise would have been. :-).
Cheers,
Gary.