invalid results
8 Oct 2015 16:25:57 UTC
Topic 198265
(moderation:
My W8 machine has been crunching happily for over 2 years with virtually all work validating. I now have 3 Gamma-ray pulsar search #4 v1.15 (FGRP4-SSE2 tasks that are invalid. The CPU temps are normal. Does any one have a clue on what is wrong?
invalid results
)
You just got caught up in the GRP4 version 1.14 to 1.15 conversion problem. There is considerable discussion of this matter on this forum in other threads.
In two cases you were issued the work as 1.14. Your initial quorum partner failed to deliver, and eventually two 1.15 issues agreed with each other, so as yours did not agree with them you lost out.
In the third case you were the recipient of a 1.15 re-issue of work initially issued as 1.14. One of the two initial quorum partners missed deadline, and you received a 1.15 re-issue. However, the late initial recipient did reply before you did, and their result, while late, validated against the other initial release. That closed the polls as to what the "right answer" was deemed to be. As both you and the other 1.15 recipient (who had bigger cause than you for gripe, as they returned quickly, before the late 1.14 partner) did not match the "right answer" you both got invalidated.
This is all past the peak, and the powers that be are considering belated issue of credit in some of the crossed-up cases. You might get one or so more before the issue is over.
Thanx for the reply, I shall
)
Thanx for the reply, I shall not worry any more about it.
Just in case someone else
)
Just in case someone else stops by on this topic: this is definitely not over yet. I, personally, have just one 1.14 invalid form this cause, but I have three remaining pending BRP4 1.14 tasks. All three have suffered a quorum partner failure in 1.14: in one case timed out, and in two cases aborted late enough in the game that the re-issue went out as 1.15. I think this means it is very likely a comparison of our two will not result in validation, another reissue will go out as 1.15, the two 1.15s will match, and mine won't.
If the project comes through on their possible scheme to grant credit in such cases, I may get credit, but the loss from extra computation is incurred regardless. Spilt milk now--let's move on.
But I don't fault the project--whose sole error in my view was failing to let us know as soon as they reasonably could. That would have been comforting, but would not actually have changed my actions, nor my experience.
By contrast Betreger has just two pending 1.14 tasks, both awaiting a 1.14 partner--so there remains hope. One is likely to be lost as the quorum partner has a poor recent record of actually returning GRP4 work, often aborting after long delay. The other has a quorum partner with a track record of very long return time delay, but with success.
So my guess is that both Betreger and I will wind up with four invalids to this affair. It should be nearly over by late October (time enough for last day issues on 1.14 to have timed out, and subsequent 1.15 re-issues to have timed out).
http://einstein.phys.uwm.edu/
)
http://einsteinathome.org/host/10659288
(YOW)
RE: http://einstein.phys.uw
)
That is my current most productive host. Mind you, BOINC, as is the habit, does not give the correct GPU details. Instead of having two GTX970 cards, that host really has one GTX970 card and one GTX750Ti.
The 970 is running a considerable memory clock overclock, and a moderate core clock overclock, pursuant to the opportunity discussed in this thread in Cruncher's Corner.
The 750Ti is a SC model, but has no user modification beyond the factory "superclock".
The host is a mere two-core Haswell of lowly i3 designation, but does seem able to support the GPU work pretty well. If power and purchase cost are of interest, I think for that generation the two-core Haswell has served nicely. I only run a single CPU task on it. Current calculated daily credit production is about 195,000, not allowing for downtime and wife usage impact. The major elements of the configuration have only been stable since September 21, so the RAC does not yet fully reflect the current production, though it is getting pretty close (183,874 at the moment).
Oh, by the way, it is my wife's daily driver, though in all honesty these days she handles most email on her tablet, so on many days it is little used.
The biggest surprise of cheap re-use on that machine is that is uses a Nexus 430-watt supply, which most references would deem considerably inadequate for the graphics cards mounted on it, which I pulled off an earlier build, and only placed on this one when it was born as an inexpensive lower power build (things changed...). Possibly the Einstein Parkes PMPS load imposes less severe power transients than do some games. The only game routinely played on the system is Solitaire, so no trouble there.
Well that explains the
)
Well that explains the 183,000+ RAC Archae!
Since my fastest here is a 660Ti SC on my older quad-core and Rac is 77,000+ I had to of course look up that GTX 970 4GB card.
And when I saw the Cuda cores 1664 I had to come right back here to see if you would explain that.
I even looked up that i3-4130 CPU
The main thing is that Rac sure looks nice
I usually have 6 or 7 hosts here running 24/7 and the only one that ever gets used for the usual pc stuff along with crunching is the laptop I am on right now (also has an old 610 GeForce so it also runs BRP6 X2 daily on this 8-core)
I will never catch your total Rac of 426,000+ that's for sure.
(oh and I am from the *Pong* days so I never play video games)
Thanks
As the project has posted
)
As the project has posted elsewhere, they have undertaken to scan the data for WUs which lost credit to the application change issue discussed here with the intention of bestowing credit, and have so far carried out one pass. The single 1.14 task of mine so far declare invalid by this mechanism now continues to show validate state as invalid, but now shows granted credit of 693.