Which lends further support to the request that both unsuccessful and
invalid results should be punished.
jm7
Richard
Haselgrove
<r.haselgr...@bti To
nternet.com> "Raistmer" <[email protected]>,
Sent by: "Jorden van der Elst"
<boinc_dev-bounce <[email protected]>
[email protected] cc
u> BOINC Developers Mailing List
<[email protected]>
Subject
05/26/2010 11:05 Re: [boinc_dev] host punishment
AM mechanism revisited
> Actually, when GPU comes into corrupted state, other projects could
> experience same problems.
> I.e., device "works" (not report errors), but makes invalid computations
> (or provides invalid data).
> Invalid data for SETI most probably leads to pulses overflow.
> Invalid data for some another project will lead to something different of
> course, but invalid data still invalid and not usable.
In my personal experience of a card in this state (twice in the last four
card-years), each time the card returned 'success' (later 'invalid') for
SETI, but interleaved with 'computation error' for GPUGrid, as the two
projects scrambled alternately to keep up with the demanded download rate.
_______________________________________________
boinc_dev mailing list
[email protected]
http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev
To unsubscribe, visit the above URL and
(near bottom of page) enter your email address.
_______________________________________________
boinc_dev mailing list
[email protected]
http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev
To unsubscribe, visit the above URL and
(near bottom of page) enter your email address.
