More than 3 times faster with Albert?
12 Jan 2006 18:11:21 UTC
Topic 190592
(moderation:
Hi,
my computer became more than 3 times faster with the Albert WUs (9000 secs vs. 30000 secs): http://einsteinathome.org/host/393405/tasks
How can that be? Most other hosts I checked are only twice as fast. The 30,000 seconds were the normal range for the old WUs.
Any ideas?
More than 3 times faster with Albert?
)
I suppose the fact that they're much smaller might make them go more quickly. Where a large Einstein datafile might have at most 16 WUs sliced from it, the albert datafiles have scores of them.
microcraft
"The arc of history is long, but it bends toward justice" - MLK
RE: Hi, my computer became
)
The WUs are no longer equal in length. For further information on
Albert you may want to read this.
Michael
Team Linux Users Everywhere
You may also see an
)
You may also see an improvement with regard to your PowerMac as the Albert seems to be slightly better coded for Altivec, I have had run times as low as 51 minutes ... :)
More typically the G5 does them in 2.5 to 4 hours ... most about 3 hours ...
RE: You may also see an
)
Paul,
that's exactly what puzzles me! My Macs (a 2x2.5 and a 2x2.7 GHz machines) take appr. 9000 seconds for each WU. My Windows machine is a 1.7 GHz machine and also takes 9000 seconds now.
Before these Albert WUs started the relation was roughly 15000 seconds for the Mac and 30000 seconds for the Windows machine, which is logical due to the difference in MHz and better Altivec coding. However, for the Albert WUs all 3 machines take appr. 9000 seconds.
RE: RE: Hi, my computer
)
Michael,
if you look at the results you will see, that the WUs are almost identical:
14286024 3474827 10 Jan 2006 10:06:44 UTC 12 Jan 2006 15:17:07 UTC Over Success Done 9,355.45 22.17 pending
14286015 3474824 10 Jan 2006 10:06:42 UTC 12 Jan 2006 8:47:42 UTC Over Success Done 9,313.55 22.08 pending
14286009 3474822 10 Jan 2006 10:06:40 UTC 12 Jan 2006 4:39:27 UTC Over Success Done 9,346.66 22.15 pending
14286000 3474819 10 Jan 2006 10:06:38 UTC 12 Jan 2006 0:38:17 UTC Over Success Done 9,333.94 22.12 pending
14285994 3474817 10 Jan 2006 10:06:36 UTC 11 Jan 2006 20:30:13 UTC Over Success Done 9,365.31 22.20 18.90
14285982 3474813 10 Jan 2006 10:06:34 UTC 11 Jan 2006 12:50:51 UTC Over Success Done 9,355.38 22.17 19.04
13653054 3269200 4 Jan 2006 13:05:52 UTC 10 Jan 2006 8:36:30 UTC Over Success Done 30,557.53 72.43 47.41
13653048 3269198 4 Jan 2006 13:05:50 UTC 9 Jan 2006 12:52:29 UTC Over Success Done 30,653.36 72.66 57.40
RE: RE: RE: Hi, my
)
That is because he has only finished a handfull Albert workunits.
As someone who came over here
)
As someone who came over here from SETI, an aspect of Einstein, particularly in the Albert 4.37 era, that is much different is the short-term non-randomness of credit variation.
The policy of trying not to require the download of a new large data file means one is in a dramatically smaller community of quorum participants. I've seen half a dozen or more sequential results have their credit determined by the exact same computer belonging to the exact same owner. If that computer claims high, you get high credit, if low, then low. Furthermore, if that computer runs a long queue, your results will stay pending for days, while if it runs a shorter queue than yours, you'll get instant credit.
For folks watching credit, if only to monitor the health of their installations and computers, I'd imagine this can give some real head jerks for a while.
I suppose over the long run this well may average out, but it means the "long run" is much longer than for BOINC communities with larger randomization of quorum members from result to result. It is of interest to me because I have four rather dissimilar machines (from a Pentium II 700 MHz runing Win98SE to a Pentium 4 EE 3.2 GHz HT running Win XP), and would like to tune the SETI to Einstein ratio per machine for higher overall output. I've realized I'll need to wait a couple of weeks at least, probably more, for comparisons to be even a little indicative.
RE: short-term
)
Which is one of the reasons we can hope that Bruce and company will quickly embrace the new credit calculation method ...
i tend to double check my
)
i tend to double check my computers output through boincstats and made a link on the desktop fer it. it is a day behind, but seems to faithfully reports the computers output by all reported credits not just credits received ( i assume so i better get the firejacket)
RE: RE: short-term
)
Fair enough. But I need some help to figure out what to claim. I'll try and talk to the Guru (aka David Anderson) on the phone this coming weekend and sort it out.
Director, Einstein@Home