Author |
Message |
|
What's the current Proth Prime Search Sieve speed per day? Or, how much T/day is PrimeGrid sieving?
Thank you in advance,
Carlos
____________
|
|
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 13579 ID: 53948 Credit: 250,650,236 RAC: 175,447
                           
|
What's the current Proth Prime Search Sieve speed per day? Or, how much T/day is PrimeGrid sieving?
Thank you in advance,
Carlos
It varies, of course, but over the last three days we've averaged about 550T per day. Each task sieves 9G and we did 62 thousand tasks per day.
____________
My lucky number is 75898524288+1 |
|
|
|
Oh my, that is significantly faster than the manual rate on the 6M-9M range. All of the work that has been done so far there could have been done by BOINC in about 16 days.
What exactly is the reason for getting the 6M-9M ranges started manually if they progress so much slower than BOINC?
____________
|
|
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 13579 ID: 53948 Credit: 250,650,236 RAC: 175,447
                           
|
Oh my, that is significantly faster than the manual rate on the 6M-9M range. All of the work that has been done so far there could have been done by BOINC in about 16 days.
What exactly is the reason for getting the 6M-9M ranges started manually if they progress so much slower than BOINC?
I suspect that it has to do with how things were done in the past. Like many projects here, the sieving likely started out as manual sieving and was then moved to BOINC to increase participation.
An argument could certainly be made that we should move all of the manual sieves to BOINC, especially those that use the same sieve software we're already using on BOINC. The same could be said of some of the PRP projects.
____________
My lucky number is 75898524288+1 |
|
|
|
An argument could certainly be made that we should move all of the manual sieves to BOINC, especially those that use the same sieve software we're already using on BOINC. The same could be said of some of the PRP projects.
What's the argument against this? |
|
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 13579 ID: 53948 Credit: 250,650,236 RAC: 175,447
                           
|
An argument could certainly be made that we should move all of the manual sieves to BOINC, especially those that use the same sieve software we're already using on BOINC. The same could be said of some of the PRP projects.
What's the argument against this?
For some of the projects we would need to make serious changes to the software to run under BOINC. Hardly impossible but also a lot more complicated than a configuration change.
____________
My lucky number is 75898524288+1 |
|
|
|
The optimal sieve work being handled is about what ~700P, ~750P, ~800P?
How can PrimeGrid grab more GPU'S? Give more points, make challenges (boincstats.com), etc?
____________
|
|
|
|
Another question, why quorum of 2 for sieve tasks?
____________
|
|
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 13579 ID: 53948 Credit: 250,650,236 RAC: 175,447
                           
|
Another question, why quorum of 2 for sieve tasks?
Two reasons: cheating, and errors.
The more and more we look at the data, the worse and worse it looks. Simply put, although you can trust most people's computers to produce the correct result most of the time, a significant amount of the time the data is bad either because of faulty hardware or software, or because of intentional cheating. The closer we look, the worse it seems.
With sieving, while an occiasional random error isn't that big a problem (missing a factor isn't so bad, and false factors are quickly detected), we have seen massive numbers of missing factors due to cheating in the past (months worth of sieving needed to be done over), and we recently discovered that there may be a systemic problem with the OpenCL sieve that could potentially caused a large number of factors to be missed.
It's probably fair to say that everything at PrimeGrid may eventually require doublechecking.
____________
My lucky number is 75898524288+1 |
|
|
Dirk Send message
Joined: 10 Mar 10 Posts: 512 ID: 56675 Credit: 723,935,401 RAC: 120,585
                      
|
That's quite disconcerting news. |
|
|
|
and we recently discovered that there may be a systemic problem with the OpenCL sieve that could potentially caused a large number of factors to be missed.
I though that was already fixed. Wasn't that the issue when you guys first started to sieve the k's up to 3M?
____________
|
|
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 13579 ID: 53948 Credit: 250,650,236 RAC: 175,447
                           
|
and we recently discovered that there may be a systemic problem with the OpenCL sieve that could potentially caused a large number of factors to be missed.
I though that was already fixed. Wasn't that the issue when you guys first started to sieve the k's up to 3M?
It's most definitely not fixed.
____________
My lucky number is 75898524288+1 |
|
|
|
we recently discovered that there may be a systemic problem with the OpenCL sieve that could potentially caused a large number of factors to be missed.
.
Are primes missed? I mean, do you think the client could be finding primes as composites removed?
____________
|
|
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 13579 ID: 53948 Credit: 250,650,236 RAC: 175,447
                           
|
we recently discovered that there may be a systemic problem with the OpenCL sieve that could potentially caused a large number of factors to be missed.
.
Are primes missed? I mean, do you think the client could be finding primes as composites removed?
There's absolutely no chance of that happening. The server verifies that every factor returned by a sieve task is a valid factor.
____________
My lucky number is 75898524288+1 |
|
|
|
I was wondering why the oldest unfinished units for these when you check subproject status seems to be decenber 2015, and the same host has 8 tasks that finish here. am I reading this right as it seems way excessive time limit, or is there something I dont know about? If so I apologise. |
|
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 13579 ID: 53948 Credit: 250,650,236 RAC: 175,447
                           
|
I was wondering why the oldest unfinished units for these when you check subproject status seems to be decenber 2015, and the same host has 8 tasks that finish here. am I reading this right as it seems way excessive time limit, or is there something I dont know about? If so I apologise.
There's a known bug in BOINC that's been around a long time where sometimes the expiration date gets corrupted. Usually we notice this because the corrupted deadlines are too short, but extra long deadlines are possible too.
The cause of the problem is unknown, but I know it's not specific to PrimeGrid.
____________
My lucky number is 75898524288+1 |
|
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 13579 ID: 53948 Credit: 250,650,236 RAC: 175,447
                           
|
Upon review, the replay officials have overturned the call on the field. This is a different problem, but one we've also seen before. Ignore my previous message.
____________
My lucky number is 75898524288+1 |
|
|
|
Ah ok, thanks cobber:) |
|
|