Welcome to MilkyWay@home

Project task limit?


Advanced search

Message boards : Number crunching : Project task limit?
Message board moderation

To post messages, you must log in.

AuthorMessage
Blake

Send message
Joined: 1 Jul 12
Posts: 8
Credit: 221,140,373
RAC: 902,954
200 million credit badge10 year member badge
Message 73820 - Posted: 12 Jun 2022, 18:06:54 UTC

Hi All, I'm seeing
[Milkyway@Home] This computer has reached a limit on tasks in progress

in the Event Log. Normally I wouldn't care, but it's happening at a time when the GPU is idle. Are all the CPU tasks I've collected keeping me from getting new GPU ones?

P.S. Most people on the leaderboard seem to have opted out of Milky's CPU-only applications somehow, which I couldn't figure out how to do.
ID: 73820 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profilemikey
Avatar

Send message
Joined: 8 May 09
Posts: 3038
Credit: 513,180,741
RAC: 251,499
500 million credit badge13 year member badgeextraordinary contributions badge
Message 73822 - Posted: 12 Jun 2022, 18:44:17 UTC - in response to Message 73820.  
Last modified: 12 Jun 2022, 18:50:39 UTC

Hi All, I'm seeing
[Milkyway@Home] This computer has reached a limit on tasks in progress

in the Event Log. Normally I wouldn't care, but it's happening at a time when the GPU is idle. Are all the CPU tasks I've collected keeping me from getting new GPU ones?


There is a 10 minute backoff time out enforced by the Project where NO communications can be sent regarding gpu stuff, then you will get your full allotment again. What alot of people do is have a zero resource share project that behaves and will send you a task that lasts a bit over over 10 minutes so MilkyWay can then refill your cache again without you needing to ask for tasks.

P.S. Most people on the leaderboard seem to have opted out of Milky's CPU-only applications somehow, which I couldn't figure out how to do.


You can change your preferences here https://milkyway.cs.rpi.edu/milkyway/prefs.php?subset=project

you can see in the top line wher3 you can choose to get or not get cpu tasks and the next 2 lines help people with multiple gpu's decide if they want both kinds of gpu's to get tasks or not.
ID: 73822 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Blake

Send message
Joined: 1 Jul 12
Posts: 8
Credit: 221,140,373
RAC: 902,954
200 million credit badge10 year member badge
Message 73824 - Posted: 12 Jun 2022, 23:20:06 UTC

Thanks for the helpful reply. So are we saying that the CPU-only tasks do, indeed, contribute to the overall per-host limit? That would seem to disincentivize CPU participation (which, I imagine, is equally valuable to science).
ID: 73824 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profilemikey
Avatar

Send message
Joined: 8 May 09
Posts: 3038
Credit: 513,180,741
RAC: 251,499
500 million credit badge13 year member badgeextraordinary contributions badge
Message 73826 - Posted: 13 Jun 2022, 9:55:46 UTC - in response to Message 73824.  

Thanks for the helpful reply. So are we saying that the CPU-only tasks do, indeed, contribute to the overall per-host limit? That would seem to disincentivize CPU participation (which, I imagine, is equally valuable to science).


No it's only gpu tasks that have the 10 minute back-off but yes both kinds of processors, cpu's and gpu's, have a task limit to prevent just a few people from being the sole crunchers. This does a couple of things 1) it prevents the data from being 'tainted' if there is a problem and 2) it lets alot more people contribute to the Science. ie with the Nvidia 3080Ti and the 3090Ti series gpu's one person could easily go thru half the database of available tasks in just a few days if they ran multiple tasks at once. My own AMD 580 series gpu's were blowing thru the gpu tasks in about 90+ seconds for each task. while my Nvidia 1060's were going thru them in about 200 seconds each, I only run one task at a time.
ID: 73826 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Blake

Send message
Joined: 1 Jul 12
Posts: 8
Credit: 221,140,373
RAC: 902,954
200 million credit badge10 year member badge
Message 73888 - Posted: 22 Jun 2022, 7:03:58 UTC

Okay I give up. Which other Boinc project am I supposed to join, with the resource share set to zero, that will send me perfectly sized 10-minute work units for the GPU? Thanks & happy crunching.
ID: 73888 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profilemikey
Avatar

Send message
Joined: 8 May 09
Posts: 3038
Credit: 513,180,741
RAC: 251,499
500 million credit badge13 year member badgeextraordinary contributions badge
Message 73889 - Posted: 22 Jun 2022, 10:17:21 UTC - in response to Message 73888.  

Okay I give up. Which other Boinc project am I supposed to join, with the resource share set to zero, that will send me perfectly sized 10-minute work units for the GPU? Thanks & happy crunching.


Try the PrimeGrid PPSsieve gpu tasks, the benefit is they also pay alot of credits for each one, if you get GC's for crunching try Einstein though you may end up doing a couple of them before MilkyWay kicks back in again, the SRBase TF tasks also give a nice chunk of credits for each gpu task and are also on the GC list but there also you will end up doing more than one due to their short duration, Collatz used to be another one but they are moving to a non flooding location so are down right now.
ID: 73889 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Blake

Send message
Joined: 1 Jul 12
Posts: 8
Credit: 221,140,373
RAC: 902,954
200 million credit badge10 year member badge
Message 73932 - Posted: 29 Jun 2022, 5:49:11 UTC

Thanks for the suggestions, which have been helping. But is anyone else wondering how the current leaderboard leader is getting 22K work units per day from the server?

https://milkyway.cs.rpi.edu/milkyway/host_app_versions.php?hostid=931264

It's going to be hard to have a friendly competition if I can't even keep 1 GPU continuously occupied because of the back-off period, etc. There must be a missing piece in this puzzle?
ID: 73932 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ProfileHRFMguy

Send message
Joined: 12 Nov 21
Posts: 221
Credit: 280,797,876
RAC: 1,500,040
200 million credit badge1 year member badge
Message 73933 - Posted: 29 Jun 2022, 14:22:37 UTC - in response to Message 73932.  

Thanks for the suggestions, which have been helping. But is anyone else wondering how the current leaderboard leader is getting 22K work units per day from the server?

https://milkyway.cs.rpi.edu/milkyway/host_app_versions.php?hostid=931264

It's going to be hard to have a friendly competition if I can't even keep 1 GPU continuously occupied because of the back-off period, etc. There must be a missing piece in this puzzle?
Hmmm. Something is fishy here. I periodically refresh that link, to find that the number of tasks today increments(expected) but so does the Max tasks per day! Whats up with that? It makes me think the labels are not what they say they are.
ID: 73933 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Blake

Send message
Joined: 1 Jul 12
Posts: 8
Credit: 221,140,373
RAC: 902,954
200 million credit badge10 year member badge
Message 73941 - Posted: 1 Jul 2022, 22:34:28 UTC
Last modified: 1 Jul 2022, 22:36:45 UTC

For those interested, I am now getting enough WUs to keep the GPU continuously occupied. I didn't change anything on the client side to achieve this. So the take-away is (a) we don't really know how Milky Way works; but (b) if you're experiencing idle GPU time during your quest to make a splash on the leaderboard, the problem may go away by itself.

Maybe fewer WUs were being handed out by the server for temporary reasons? Still I saw that someone had gotten 20K WUs in 1 day during that period. Strange.
ID: 73941 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote

Message boards : Number crunching : Project task limit?

©2023 Astroinformatics Group