| Author |
Message |
|---|
|
|
|
Why is not recommended to run SGS with multi-tasking?
____________
Thanks,
Jim
 |
|
|
|
|
Why is not recommended to run SGS with multi-tasking?
SGS are very short tasks, and multithreading typically lowers real-life performance.
____________
My lucky number is 6219*2^3374198+1
 |
|
|
|
|
|
You can try and measure it. Multithreading will maybe make each task a bit faster, but also make the total number of completed tasks per day ("throughput") smaller, for that computer. If that is the case, you can decide for yourself if you want to go down in throughput while making the processing of each task (from it is received to it is returned) faster. /JeppeSN |
|
|
|
|
|
That's what I plan to do since empirically, for my two computers, there seems to benefit in selecting multithreading for SGS. I need to run a more controlled test which I will do.
I just wanted to make sure that there was no harm (besides timeliness) to running the task.
____________
Thanks,
Jim
|
|
|
 mikey
Send message
Joined: 17 Mar 09
Posts: 1657
ID: 37043
Credit: 733,674,269
RAC: 73,187
|
|
I just went into my preferences to enable Sophie Germain LLR cpu tasks and after saving my settings it says (disabled), is there a problem with the SGS LLR tasks and are they done being sent out?
Sophie Germain Prime Search LLR (SGS) CPU (disabled)
And of course I realized the problem...I was running a gpu project in that venue in the past and had disabled the cpu, changing that around it now works just fine.
Sophie Germain Prime Search LLR (SGS) CPU |
|
|
|
|
|
I ran SGS on two different iMacs both with multitasking on and off. SGS multitasking resulted in approximately running half the time than SGS multitasking turned off. Surprised me some, but I'm going with multitasking on for SGS.
____________
Thanks,
Jim
|
|
|
Bur Volunteer tester
Send message
Joined: 25 Feb 20
Posts: 515
ID: 1241833
Credit: 414,276,308
RAC: 40,770
|
|
Quite important, did you try assigning cores to the processes? That decreases run times by about 10-15% for single-threading.
Also, approximately half the time when running with 2 threads? That's a close call then, on the other hand, a slightly smaller throughput might be worth it due to increased 1st rate. Still, I would really test this for a few days and compare throughputs. I ran SGS a long time on an i3-2120 (2 cores) and throughput was definitely higher for single-thread.
Just out of curiosity, what does Prime95 benchmark say about it? FFT is 128k which really sounds like single-threading should be the fastest option. But of course, in the end what matters is what PG says you're computer delivered in the last 24 hours.
____________
1281979 * 2^485014 + 1 is prime ... no further hits up to: n = 5,700,000 |
|
|
|
|
|
I must admit, I'm not very good about computer architectures.
I have a one processor, 2-core iMac. I have this computer solely dedicated to SGS so I use 100% CPU and 100% usage for each CPU.
In my preferences, I selected "no-limit" the number of tasks I can run simultaneously and "no-limit" on max # of threads for each task. What resulted is 1 task running with 4 threads.
Frankly, I'm not sure why 4 threads are running instead of 2 threads.
Now that I look at it, it seems odd to me.
Any suggestions?
____________
Thanks,
Jim
|
|
|
|
|
|
You have hyperthreading enabled so it looks to boinc like you have 4 CPUs not 2. You should set boinc to 50% cpu and 2 threads although 2 x single threaded tasks is likely to be faster for SGS. |
|
|
|
|
|
Quick test gives the following time per WU:
1c = 20 mins
2c = 14
3c = 10.3
4c = 9.30
5c = 8
6c = 7.30
8c = 9
So, on an 8cores system
+/-6 WU/hour with 8c/WU (60/9)
24WU/hour with 1c/WU (60/20)*8 |
|
|
|
|
|
The question is do you want to be the finder or the dc'er? SGS does not take advantage of the new LLR I don't think (please correct me if I am wrong) so by running them faster your chance of being the finder seems to me to improve? Probably wrong on this as well but finding primes is a luck of the draw anyway.
Cheers |
|
|
|
|
The question is do you want to be the finder or the dc'er? SGS does not take advantage of the new LLR I don't think (please correct me if I am wrong) so by running them faster your chance of being the finder seems to me to improve? Probably wrong on this as well but finding primes is a luck of the draw anyway.
Cheers
Correct, SGS does not (yet?) take advantage of the fast double checking LLR2. There are two ways to see the list of projects using LLR2: first the home page sub-project list with the superscript "F", and second this thread https://www.primegrid.com/forum_thread.php?id=9303&nowrap=true#143209 |
|
|
|
|
The question is do you want to be the finder or the dc'er? SGS does not take advantage of the new LLR I don't think (please correct me if I am wrong) so by running them faster your chance of being the finder seems to me to improve? Probably wrong on this as well but finding primes is a luck of the draw anyway.
Cheers
I agree :)
Ideally, I believe that to have the best chance to find a prime the settings should be:
- no cache (to avoid crunching "stale" WU)
- max core/WU to reduce time to complete
But for project credit (thus helping the project more)
- maximizing WU is the best (cache depending on your internet access, 1core/WU) |
|
|
