Join PrimeGrid
Returning Participants
Community
Leader Boards
Results
Other
drummers-lowrise
|
Message boards :
Sieving :
BOINC sieving a reality?
| Author |
Message |
|---|
|
|
|
Looks like it to me! I just noticed Rytis's post in the Woodall Sieving Reservation Thread reserving work for a BOINC sieving test--and when I checked the home page and my project preferences page, this is what I saw:
Woo hoo!  Now all the AMD users here won't have to worry about their CPUs being underutilized. ;-) (I don't have any AMDs attached to the project at the moment, but I'm sure those who do will be happy.)
Is this intended to replace the manual sieve, or supplement it? If you're looking for opinions on that, I'd vote for the manual sieving to be left open as an option, since in some cases it's better to do sieving manually (i.e. with older computers).
____________
 | |
|
RytisVolunteer moderator
Project administrator
Send message
Joined: 22 Jun 05
Posts: 2653
ID: 1
Credit: 102,368,713
RAC: 82,384
                     
|
|
Yes, it indeed is reality :)
We'll leave manual sieve open for now (maybe reducing open ranges). Manual reservation will be available at least until we have Linux sieve.
____________
| |
|
|
|
Yes, it indeed is reality :)
We'll leave manual sieve open for now (maybe reducing open ranges). Manual reservation will be available at least until we have Linux sieve.
Ah, I see. I was thinking, though, it would be nice if it were possible to continue with the manual sieving for the users who would prefer it--sort of like how Riesel Sieve does it, by having a manual sieving operation, but with the admins reserving big ranges for BOINC--and even though BOINC sieving there is outputting a whole lot more than the manual sieving, they still keep it open because some users prefer it (and some users, like me, prefer to use BOINC in some cases and manual sieving in others).
____________
| |
|
|
|
|
So at what point it credit cranted for the sieving work units? There are a few now that have met the 2 completion requirement that have not granted credit. Is this on purpose, or something not yet caught?
http://www.primegrid.com/workunit.php?wuid=14895763
Like that one.
____________
 | |
|
|
|
|
any est. time to finish one?
EDIT:And how about the range of sieve? 1G or 1M or other amount? | |
|
|
|
|
takes about 15 min on my core 2
range is 2e7 or 20M if im not mistaken...
BoB
____________
 | |
|
|
|
|
I took about 25mins on my 4600+
@bob:you are right. It is 20M. I find that in the directory of BOINC. | |
|
|
|
|
They seem to be taking about an hour and 45 minutes apiece on my P4 3.2Ghz with hyperthreading. I guess the big differences in times compared to the others listed here could be attributed to two things: a)It's an Intel, and Intel's don't exactly excel at sieving b)The hyperthreading means that a workunit takes almost twice as long to complete as it would on the same system with hyperthreading turned off, but it can run two tasks at a time.
Even with those two factors accounted for, though, aren't my runtimes still a little long? I know I don't have any other CPU-intensive applications running on the same hyperthread, so that can't be the reason. (Even if I did, then that wouldn't affect the CPU time that BOINC shows: Instead, the CPU time would increase slower, but it would take the same amount of CPU time in the end.)
____________
| |
|
|
|
They seem to be taking about an hour and 45 minutes apiece on my P4 3.2Ghz with hyperthreading. I guess the big differences in times compared to the others listed here could be attributed to two things: a)It's an Intel, and Intel's don't exactly excel at sieving b)The hyperthreading means that a workunit takes almost twice as long to complete as it would on the same system with hyperthreading turned off, but it can run two tasks at a time.
Even with those two factors accounted for, though, aren't my runtimes still a little long? I know I don't have any other CPU-intensive applications running on the same hyperthread, so that can't be the reason. (Even if I did, then that wouldn't affect the CPU time that BOINC shows: Instead, the CPU time would increase slower, but it would take the same amount of CPU time in the end.)
Hello
My Amd 3800+ runs about 30 Min. / work unit, My Intel 3.2 runs one in about 21 min. / work unit.
Mike | |
|
John Honorary cruncher
Send message
Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0
               
|
|
Some recent observations for completion times:
A 2e7 sieve range in the 5M<n<10M dat file takes:
6,220.16 sec - Intel(R) Pentium(R) 4 CPU 3.20GHz
4,002.90 sec - AMD Athlon(TM) XP1800+
1,713.61 sec - AMD Athlon(tm) 64 X2 Dual Core Processor 3800+
1,334.28 sec - Intel(R) Xeon(R) CPU E5310 @ 1.60GHz
1,161.05 sec - Intel(R) Core(TM)2 CPU 6400 @ 2.13GHz
963.83 sec - Intel(R) Core(TM)2 CPU 6300 @ 1.86GHz
____________
| |
|
|
|
Some recent observations for completion times:
A 2e7 sieve range in the 5M<n<10M dat file takes:
Intel(R) Pentium(R) 4 CPU 3.20GHz 6,220.16 sec
AMD Athlon(TM) XP1800+ 4,002.90 sec
AMD Athlon(tm) 64 X2 Dual Core Processor 3800+ 1,713.61 sec
Intel(R) Xeon(R) CPU E5310 @ 1.60GHz 1,334.28
Intel(R) Core(TM)2 CPU 6400 @ 2.13GHz 1,161.05 sec
Intel(R) Core(TM)2 CPU 6300 @ 1.86GHz 963.83 sec
Okay, I guess that means my times are about normal for my CPU, then. Unless, of course, the time listed for a P4 3.2Ghz/HT is based solely on my computer's results. 
P.S.: Would it be possible to include on your account page the number of factors you've found, sort of like how the LLR subprojects will tell you how many primes you've found? Because of the inequality between different ranges of sieving, it probably wouldn't work to show on there, say, how many G a user has done, but it still would be cool to have it show how many factors you've found. 
____________
| |
|
JohnHonorary cruncher
Send message
Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0
|
P.S.: Would it be possible to include on your account page the number of factors you've found, sort of like how the LLR subprojects will tell you how many primes you've found? Because of the inequality between different ranges of sieving, it probably wouldn't work to show on there, say, how many G a user has done, but it still would be cool to have it show how many factors you've found.
See this thread:
http://www.primegrid.com/forum_thread.php?id=719
____________
| |
|
|
|
P.S.: Would it be possible to include on your account page the number of factors you've found, sort of like how the LLR subprojects will tell you how many primes you've found? Because of the inequality between different ranges of sieving, it probably wouldn't work to show on there, say, how many G a user has done, but it still would be cool to have it show how many factors you've found.
See this thread:
http://www.primegrid.com/forum_thread.php?id=719
Ah, looks like zombie67 beat me to the chase. I missed that post somehow. :-)
____________
| |
|
|
|
|
Actually, you did post first. :P
____________
| |
|
|
|
|
So how quickly (range per week?) is PG BOINC sieving? Versus manual sieving?
____________
| |
|
JohnHonorary cruncher
Send message
Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0
|
So how quickly (range per week?) is PG BOINC sieving? Versus manual sieving?
It's an apples to oranges comparison. We're not able to compare them side by side. As PG BOINC sieving is still in development and testing, it's not working on the front line of sieving.
PG BOINC is working on the Woodall 5M<n<10M range and completing about 330G per week. As there is no current manual sieving in that range we're not able to directly compare it.
Manual sieving is working on the 3.25M<n<4M range. This range is much smaller than the 5M<n<10M so we're moving substantially faster. We're completing about 1000G per week with just one siever (using 64 bit Linux gcwsieve).
One factor in the speed of the manual sieve is that we can use the 64 bit Linux client of gcwsieve. This is about 2X as fast as the 32 bit Windows client being used in PG BOINC sieving.
The goal is to eventually offer the 32 bit Windows/Linux and 64 bit Windows/Linux in PG BOINC sieving.
Finally, the 4M<n<5M Woodall range has been halted until the same range in the Cullen sieve catches up. Then we're going to test sieving both Cullen's and Woodall's together. It's a work-around for eventual PG BOINC sieving.
If I had to make a rough guess based on previous manual sieving in the 5M<n<10M range, PG BOINC sieving is like adding around 60 full time cores (using 32 bit gcwsieve) to the effort.
____________
| |
|
|
|
Some recent observations for completion times:
A 2e7 sieve range in the 5M<n<10M dat file takes:
6,220.16 sec - Intel(R) Pentium(R) 4 CPU 3.20GHz
4,002.90 sec - AMD Athlon(TM) XP1800+
1,713.61 sec - AMD Athlon(tm) 64 X2 Dual Core Processor 3800+
1,334.28 sec - Intel(R) Xeon(R) CPU E5310 @ 1.60GHz
1,161.05 sec - Intel(R) Core(TM)2 CPU 6400 @ 2.13GHz
963.83 sec - Intel(R) Core(TM)2 CPU 6300 @ 1.86GHz
From what it looks like, the figure above for a P4 3.2Ghz was based on my computer's results--so I may as well give you an update on how long it takes my computer to do a gcwsieve workunit. I just cleaned out my computer's case today--the heat sink was clogged with dust and it was giving me problems on other projects that use LLR--and now that I've cleaned it out, my computer's speed has increased drastically. A gcwsieve workunit that my computer just did completed in about an hour, exactly--which would be more like what I would expect. (I thought my 1hr. 45 min. runtimes were a bit long, considering what others were reporting!)
____________
| |
|
|
|
|
Update: I just had a gcwsieve workunit complete that took almost exactly 41 minutes. Considering that others with an Intel P4 3.2Ghz reported 20 minute runtimes (presumably without hyperthreading) then 40 minutes makes sense for my system with hyperthreading. (Probably the reason I got an hour at first was because the workunit was partially done before I cleaned out my system.)
____________
| |
|
Message boards :
Sieving :
BOINC sieving a reality? |
