Triple validated

Message boards : Number crunching : Triple validated
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile Bill G Special Project $75 donor
Avatar

Send message
Joined: 1 Jun 01
Posts: 1282
Credit: 187,688,550
RAC: 182
United States
Message 1741331 - Posted: 11 Nov 2015, 14:48:25 UTC
Last modified: 11 Nov 2015, 14:49:15 UTC

http://setiathome.berkeley.edu/workunit.php?wuid=1960925997

This is not the only WU that has been validated 3 times (or twice, depending on how you look at it) on this computer. Is this happening because of the reworking on the servers?

SETI@home classic workunits 4,019
SETI@home classic CPU time 34,348 hours
ID: 1741331 · Report as offensive
Cosmic_Ocean
Avatar

Send message
Joined: 23 Dec 00
Posts: 3027
Credit: 13,516,867
RAC: 13
United States
Message 1741399 - Posted: 11 Nov 2015, 19:58:22 UTC

I think that just happens naturally.

If there are, for example, up to 30 results reported by the first two iterations, and somewhere a few places beyond the decimal in enough of the values doesn't quite agree, it will be "inconclusive" and need a third iteration. Third iteration returns back roughly the same result, then all three are deemed "weakly similar" and "close enough," so all three get credit.

Richard and/or Josef have explained this one really well multiple times in the past, but that's my rough explanation for it.
Linux laptop:
record uptime: 1511d 20h 19m (ended due to the power brick giving-up)
ID: 1741399 · Report as offensive
Profile Bill G Special Project $75 donor
Avatar

Send message
Joined: 1 Jun 01
Posts: 1282
Credit: 187,688,550
RAC: 182
United States
Message 1741425 - Posted: 11 Nov 2015, 22:12:49 UTC - in response to Message 1741399.  

Thanks, I think you are correct. I just forget a lot of what is posted here.

As I like to say, I have forgotten more than most people will ever know.

SETI@home classic workunits 4,019
SETI@home classic CPU time 34,348 hours
ID: 1741425 · Report as offensive
Cosmic_Ocean
Avatar

Send message
Joined: 23 Dec 00
Posts: 3027
Credit: 13,516,867
RAC: 13
United States
Message 1741462 - Posted: 12 Nov 2015, 1:22:45 UTC

It seems to mostly happen in situations where there are different crunching platforms involved.. CUDA/nvidia vs. OpenCL/ATI vs. CPU. The two GPU types have different limitations on how precise the digits beyond the decimal go when they do their floating point math.

Purely mathematically, if you take pi for example, 3.141593 is not the same as 3.1415926. The validator sees that those two reported results do not exactly match, so a third iteration goes out. That one returns 3.1415927, and the validator sees that between the three of them, there's only a difference of 0.000001 or less, so it decides they are close enough and all three must be right.

Beyond that, I don't know the exact specific thresholds that make the decision on the rounding errors, but if a _2 task gets validated against _0 and _1, then that almost always means there was a difference in rounding between the first two and the third one agreed with the consensus.
Linux laptop:
record uptime: 1511d 20h 19m (ended due to the power brick giving-up)
ID: 1741462 · Report as offensive
Profile Bill G Special Project $75 donor
Avatar

Send message
Joined: 1 Jun 01
Posts: 1282
Credit: 187,688,550
RAC: 182
United States
Message 1741539 - Posted: 12 Nov 2015, 14:07:43 UTC - in response to Message 1741462.  

Beyond that, I don't know the exact specific thresholds that make the decision on the rounding errors, but if a _2 task gets validated against _0 and _1, then that almost always means there was a difference in rounding between the first two and the third one agreed with the consensus.

Certainly seems to be what happened here. Thanks again.

SETI@home classic workunits 4,019
SETI@home classic CPU time 34,348 hours
ID: 1741539 · Report as offensive

Message boards : Number crunching : Triple validated


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.