| Author |
Message |
|---|
|
|
|
Since the 24 of nov i have been getting 1.36 pps sieve jobs on all my computers but my amd's suck at seiving these tasks so would like to just have them go to my nvidia gpu's but if i turn off use cpu then i cant get and jobs for pps llr... it there a way to block sieve jobs from going to my cpu ?? |
|
|
|
|
|
The only thing that comes to my mind is using some computers for sieving and the others for llr. That, of course, can be done with home,school,work profiles. |
|
|
 Michael Goetz Volunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13513
ID: 53948
Credit: 237,712,514
RAC: 0
                           
|
Since the 24 of nov i have been getting 1.36 pps sieve jobs on all my computers but my amd's suck at seiving these tasks so would like to just have them go to my nvidia gpu's but if i turn off use cpu then i cant get and jobs for pps llr... it there a way to block sieve jobs from going to my cpu ??
Turn OFF the checkbox for PPS sieve and turn ON the checkbox for using other tasks when nothing is available (which will always be the case with the GPU since you deselected the only GPU task.)
____________
My lucky number is 75898524288+1 |
|
|
|
|
Since the 24 of nov i have been getting 1.36 pps sieve jobs on all my computers but my amd's suck at seiving these tasks so would like to just have them go to my nvidia gpu's but if i turn off use cpu then i cant get and jobs for pps llr... it there a way to block sieve jobs from going to my cpu ??
Turn OFF the checkbox for PPS sieve and turn ON the checkbox for using other tasks when nothing is available (which will always be the case with the GPU since you deselected the only GPU task.)
I think what he's saying is he wants to continue crunching cuda pps(sieve), but only on the GPU. He can't block CPU because then he can't get CPU llr tasks. So, he wants sieve on GPU, llr on CPU without receiving CPU sieve work.
Something just occurred to me. You can do it by using an app_info.xml file. Don't know why I didn't think of that in the first place. |
|
|
|
|
|
Thank you so much.. thats exactly what i was wanting to do so for i have been baby sitting my 2 boxes with nvidia gpus's aborting the llr 1.36 task since my gpu can do them in 15-20 min versus my cpu taking 20 hours... when they could be doing pps llr in 15 minutes ...amirite? |
|
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13513
ID: 53948
Credit: 237,712,514
RAC: 0
|
Since the 24 of nov i have been getting 1.36 pps sieve jobs on all my computers but my amd's suck at seiving these tasks so would like to just have them go to my nvidia gpu's but if i turn off use cpu then i cant get and jobs for pps llr... it there a way to block sieve jobs from going to my cpu ??
Turn OFF the checkbox for PPS sieve and turn ON the checkbox for using other tasks when nothing is available (which will always be the case with the GPU since you deselected the only GPU task.)
I think what he's saying is he wants to continue crunching cuda pps(sieve), but only on the GPU. He can't block CPU because then he can't get CPU llr tasks. So, he wants sieve on GPU, llr on CPU without receiving CPU sieve work.
That's exactly what my procedure does. :)
It's how I normally have my computer set up. No app_info needed, at least for now as long as there's only one GPU app.
____________
My lucky number is 75898524288+1 |
|
|
|
|
|
I think I am running the way you wish to be -
My preferences are set to:
Use CPU Enforced by version 6.10+ yes
Use ATI GPU Enforced by version 6.10+ no
Use NVIDIA GPU Enforced by version 6.10+ yes
Projects
321 Prime Search (LLR) no
Cullen Prime Search (LLR) no
Prime Sierpinski Problem (LLR) no
Proth Prime Search (LLR) yes
Seventeen or Bust (LLR) no
Sophie Germain Prime Search (LLR) no
The Riesel Problem (LLR) no
Woodall Prime Search (LLR) no
321 Prime Search (Sieve) no
Cullen/Woodall Prime Search (Sieve) no
The Riesel Problem (Sieve) no
Proth Prime Search (Sieve) no
Send work from any subproject if selected projects have no work yes
This feeds my CPU with LLR's and my GPU with sieves.
This will work only because there is no shortage of PPS LLRs and the work cache is almost always >0.
Your BOINC time to completion will be strange. On this system Sieves show a initial run time of 19 min and the LLRs just over 1 hour even though they complete in <5 min. a well known problem with BOINC.
I run with an additional work cache of .5 days in order to maintain enough PPS LLRs to cover server outages.
Hope this is what you were looking for and if so hope it helps.
-John
____________
There's someone in our head but it's not us. |
|
|
|
|
|
Yeah - my kind of speech!
Either kill the ppsSieve for CPUs completely - it does not make sense with 1680 minutes per core on a 2 GHz Nehalem versus 25 minutes on a GTX 260 or 13 minutes on a GTX 460. 8 to 16 times slower. Not to speak of GTX 470/480/580... Or split it into two subprojects with shorter WUs for the CPU. With boinc the management would be less troublesome. |
|
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13513
ID: 53948
Credit: 237,712,514
RAC: 0
|
|
UG, sorry for messing up this thread and making it 8 feet wide because I left out a "/" in the BB tag. :( (It's too late to edit it now, but if one of the mods want to delete it, please do.)
Here's the way my post was supposed to look:
Your BOINC time to completion will be strange. On this system Sieves show a initial run time of 19 min and the LLRs just over 1 hour even though they complete in <5 min. a well known problem with BOINC.
Yes, I should have mentioned that.
The GPU and CPU apps have wildely different DCF's. amd every time a GPU app finishes (which happens a lot more frequently than a CPU app finishes), it raises the DCF dramatically. When a CPU app finishes, it lowers the DCF only a little.
As a result, the time duration estimates for the GPU tasks will be pretty accurate, but the CPU task estimates will be several times higher than they should be. These will correct themselves as the tasks run, but the "waiting to run" CPU tasks will also have estimated run times that are too high.
This has two consequences:
1) BOINC will download too little work for your CPU because it thinks the tasks will take a lot longer than it actually will. You may set your preferences to keep a cache of one day's work, but it may only download enough tasks for 6 hours.
2) When running long tasks (Woodahll, Cullen, PSP, and SoB LLRs), BOINC will likely decide that it can't finish the task it just downloaded by its deadline. It will therefore put it into high priority mode and start running it immediately while preempting a task that's already running. As this task gets more complete, the estimate will correct itself. Upon realizing that it doesn't have that much work to do, BOINC will download more. The newly downloaded task will then also go into high-priority mode and preempt th previous task. The end result can be a whole bunch of almost-completed tasks waiting for a never-ending supply of new high-priority tasks. (Eventually the preempted tasks will run, once they're close to missing their deadlines for real.)
____________
My lucky number is 75898524288+1 |
|
|
