Join PrimeGrid
Returning Participants
Community
Leader Boards
Results
Other
drummers-lowrise
|
Message boards :
Seventeen or Bust :
Balls-up
| Author |
Message |
|---|
 Dave 
Send message
Joined: 13 Feb 12
Posts: 3258
ID: 130544
Credit: 2,477,120,207
RAC: 4,460,000
                           
|
|
Just relaunched BOINC after a game, pasued the PSPs it had jumped on to for it to carry on with the SoBs in progress that were 3 to 4 days in out of 5, & all 4 errored.
Tried rolling back to yesterday's BOINC folder backup (as I make regular backups now),. checked online & it said all tasks were abandoned, even the ones I'd got in my cache & not started.
So all the 11 SoBs I've had in cache, 4 errorded & 7 not-started, are now aborted.
Fresh cache.
Effing computers sometimes.................... | |
|
|
|
|
I wonder if running only 3 SOB at a time as opposed to 6 would take less time to complete? And be worth it ? It sure is faster with 6 running as opposed to 12 at a time. Hope I also do not get errors after several days of computer time.
Merry Christmas All. | |
|
|
|
|
3 would be faster than 6, but depending on the system would probably not be enough to change. The biggest issue with running more is memory bandwidth, contention and cache hits. Faster memory, if your system supports it, would speed up the work most likely more than running less units. I am sure someone else can expand on this more.
| |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 14045
ID: 53948
Credit: 485,871,339
RAC: 676,071
                          
|
I wonder if running only 3 SOB at a time as opposed to 6 would take less time to complete? And be worth it ? It sure is faster with 6 running as opposed to 12 at a time. Hope I also do not get errors after several days of computer time.
Merry Christmas All.
Faster? Yes, somewhat, but nothing like going from 12 to 6. You only really have 6 cores, so running 12 tasks is similar to running each task 50% of the time. (Hyperthreading IS beneficial for other tasks, but not LLR).
Going from 12 to 6 you cut the time in half (or maybe even slightly better).
Going from 6 to 3, however, I expect you'll see something between a 5% and 20% improvement. Since you'll be reducing the number of tasks running by 50%, overall, it's not worth it.
____________
My lucky number is 75898524288+1 | |
|
Dave
Send message
Joined: 13 Feb 12
Posts: 3258
ID: 130544
Credit: 2,477,120,207
RAC: 4,460,000
|
|
You don't even want to think about having HT on for SoBs.
+ 50% utilisation is a waste of a core on a 4-core system (or possibly 2 cores on a hex-core system). | |
|
|
|
|
Thank you Michael and Dave. | |
|
|
|
3 would be faster than 6, but depending on the system would probably not be enough to change. The biggest issue with running more is memory bandwidth, contention and cache hits. Faster memory, if your system supports it, would speed up the work most likely more than running less units. I am sure someone else can expand on this more.
I very recently upgraded my memory on my 3570K from 1866 to 2400 and my BOINC benchmarks went from 47xx and 13xxx to 53xx and 15xxx @4.4 OC so memory speed DOES make a big difference.
____________
Largest Primes to Date:
As Double Checker: SR5 109208*5^1816285+1 Dgts-1,269,534
As Initial Finder: SR5 243944*5^1258576-1 Dgts-879,713
 | |
|
|
|
|
I wonder what is the normal credit for SOB? It is so rare to fid anyone who completed and got credit for one. | |
|
HonzaVolunteer moderator
Volunteer tester
Project scientist
Send message
Joined: 15 Aug 05
Posts: 1963
ID: 352
Credit: 6,427,304,482
RAC: 2,416,882
                               
|
I wonder what is the normal credit for SOB? It is so rare to fid anyone who completed and got credit for one.
Around 44k - 50k credit per task, depending on k and n values.
____________
My stats | |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 14045
ID: 53948
Credit: 485,871,339
RAC: 676,071
|
I wonder what is the normal credit for SOB? It is so rare to fid anyone who completed and got credit for one.
The credits are based upon the size of the task, and since each number tested is different, no two SoB tasks have the same credit. They're fairly close in value, however, and are increasing over time as the numbers we're testing get larger and larger.
So far, the most credit awarded to a single SoB task is 50032.94 credits.
____________
My lucky number is 75898524288+1 | |
|
|
|
|
Thanks again. | |
|
|
|
|
Mike, so what do you think of the BOINC benchmark increase I posted above when upgrading from 1866 9-10-9-24 to 2400 10-12-12-31 (both G.Skill). I ran multiple times and are rounded averages.
____________
Largest Primes to Date:
As Double Checker: SR5 109208*5^1816285+1 Dgts-1,269,534
As Initial Finder: SR5 243944*5^1258576-1 Dgts-879,713
| |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 14045
ID: 53948
Credit: 485,871,339
RAC: 676,071
|
Mike, so what do you think of the BOINC benchmark increase I posted above when upgrading from 1866 9-10-9-24 to 2400 10-12-12-31 (both G.Skill). I ran multiple times and are rounded averages.
Since the BOINC benchmarks don't use AVX (or any other SIMD instructions suchs as SSE), they're not as memory intensive as LLR -- and, in fact, the benchmarks, being designed primarily as arithmetic benchmarks are likely written so as to avoid memory bandwidth problems. As such, they're a poor representation of real world apps similar to ours.
Fast memory with an AVX processor provides a significant performance boost, and even more so with FMA.
Since fast memory is NOT expensive (I found that equivalent 2400 MHz DDR3 memory was the same price as 1600 MHz), it's a really good 'bang for the buck' purchase. Switching the memory speed between 1600 and 2400 MHz (via the BIOS) significantly changes the LLR performance.
____________
My lucky number is 75898524288+1 | |
|
Message boards :
Seventeen or Bust :
Balls-up |
