For your information: The range of WUs we are working with are starting to partially converge to a solution. That is the reason they are taking more time. Remember that the Neurona modeller stops the computations as soon as it detects that the model is not feasible. The better the model generated by a WU fits with the ideal solution, the longer it takes to be computed. Most of the WUs end in just a few minutes just because they are not valid, but now are appearing a significant number of "not so bad" WUs. While I have not seen a "bingo" yet in the results, many WUs have been able to be half the way.
/ 
this is really good news.
maybe there is a "bingo" workunit on its way ;)
/ ok. these two units are real monsters.
~15.000secs
http://falua.cesfelipesegundo.com/Neurona/workunit.php?wuid=1676919
http://falua.cesfelipesegundo.com/Neurona/workunit.php?wuid=1676627
edit:
and there are a lot more of these really long units.
currently my machine runs a unit for 2:26 hours.
/ The best solution found so far, but not yet 100% valid, is Prueba20519908 with 12707 seconds of computation time, we are closing in. This is very exciting. After millions of WU I hope we can find something soon.
/ http://falua.cesfelipesegundo.com/Neurona/result.php?resultid=55250333
15549 seconds :)
/ This morning I had a quick look to the results folder and there are already more than 20 solutions for the neuronal network. That means that the model defined in the WU is at least able to sustain activity for the complete 1500 rounds the modeller simulates. I have not yet analysed if the activity not only lasts long enough but if also follows a valid pattern (that's another question), but looks quite promising.
More information will come as it becomes available. Keep going with the hard work.
/ wu are getting longer, maybe it's time to think about the checkpoint
/ wu are getting longer, maybe it's time to think about the checkpoint
I was thinking about that but I have some concerns. To save the results and allow the modeller continue from a checkpoint I need to store the complete network, and that means a lot of space (at least as much as the RAM requirement, 2.5 GB per WU as a minimum) written and read every time the checkpoint is stored and restored.
/ 
I know that this is an old thread, and I'm not doing this project for the credit, but I've had workunits this month that have gone over 6 hours on my 3.4 ghz quad-core, that I've received 8 credits for. That's a lot of crunch time on a decent machine. 12 gigs of RAM, so that's not the problem.
____________
/ ...I've had workunits this month that have gone over 6 hours...
I've recently had a couple of wildly long results as well:
55361275
55361266
____________
/ I can increase the credit limit again, but will have side effects. In average I think it is OK. The WUs with so long CPU time are not so frequent and the 8 credits are also given from time to time to shorter WUs.
/ Hi Javier,
did you increase the credit limit? While looking up my output it seems that the overall outcome is decreasing.
And by the way, how is it possible, that some WU with less CPU-time than others get more credits?
Regards,
sightus
/ Hi sightus!
I changed nothing since the limit was set to 8. The credits are calculated by BOINC automatically but for some reason this has not worked very well since the beginning of the project. Couldn't find a solution for that. My guess is that the system does not work well with WUs with so huge changing CPU times, but this is only a guessing.
/ Ok. Is there gonna be another adjustment of credits? How many WU are left?
Regards
/ No changes are foreseen, and there are still 180000 WUs left.
/ It seems that there are some problems getting some new WU ;)
/