Advanced search

Message boards : Frequently Asked Questions (FAQ) : No tasks in quite a while.

Author Message
Paul
Send message
Joined: 25 Apr 13
Posts: 26
Credit: 182,663,053
RAC: 143,115
Level
Ile
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 60718 - Posted: 3 Sep 2023 | 19:53:59 UTC

Used to have several running at once. Now, for the past several weeks, nothing. I've even expanded the amount of disk space for GPU.

Keith Myers
Send message
Joined: 13 Dec 17
Posts: 1297
Credit: 5,442,861,959
RAC: 9,970,037
Level
Tyr
Scientific publications
watwatwatwatwat
Message 60722 - Posted: 3 Sep 2023 | 22:13:19 UTC - in response to Message 60718.

Used to have several running at once. Now, for the past several weeks, nothing. I've even expanded the amount of disk space for GPU.

The only work available is the ATMbeta tasks of which there has been plenty for weeks.

Do you have those tasks selected in Preference along with the mandatory accept beta tasks?

Paul
Send message
Joined: 25 Apr 13
Posts: 26
Credit: 182,663,053
RAC: 143,115
Level
Ile
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 60729 - Posted: 6 Sep 2023 | 1:54:31 UTC - in response to Message 60722.

Beta tasks?

Paul
Send message
Joined: 25 Apr 13
Posts: 26
Credit: 182,663,053
RAC: 143,115
Level
Ile
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 60730 - Posted: 6 Sep 2023 | 2:14:59 UTC - in response to Message 60729.

Couldn't find anything saying Beta.

Ian&Steve C.
Avatar
Send message
Joined: 21 Feb 20
Posts: 1038
Credit: 40,148,957,483
RAC: 70,534,721
Level
Trp
Scientific publications
wat
Message 60731 - Posted: 6 Sep 2023 | 2:22:06 UTC - in response to Message 60730.

Couldn't find anything saying Beta.


go here: https://gpugrid.net/prefs.php?subset=project

and edit your preferences to

"Run test applications" = yes
and make sure the checkbox for "ATM (beta)" is selected

"test" = "beta"
____________

Paul
Send message
Joined: 25 Apr 13
Posts: 26
Credit: 182,663,053
RAC: 143,115
Level
Ile
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 60732 - Posted: 6 Sep 2023 | 2:36:32 UTC - in response to Message 60731.

Thank You! It worked, have 2 tasks now. Any other suggestions welcome. Your replies aren't notifying me, only if I happen to logon.

Keith Myers
Send message
Joined: 13 Dec 17
Posts: 1297
Credit: 5,442,861,959
RAC: 9,970,037
Level
Tyr
Scientific publications
watwatwatwatwat
Message 60733 - Posted: 6 Sep 2023 | 3:56:23 UTC - in response to Message 60732.

Read the main thread on these tasks in News >> ATM

https://www.gpugrid.net/forum_forum.php?id=24

They cannot be stopped or they error out.

Tasks with naming convention 2-5 through 5-5 will show 100% completion when started, this is normal, so just let them run until they finish.

The tasks DO NOT checkpoint.

Paul
Send message
Joined: 25 Apr 13
Posts: 26
Credit: 182,663,053
RAC: 143,115
Level
Ile
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 60983 - Posted: 15 Jan 2024 | 2:16:28 UTC - in response to Message 60732.

I have over a TB of space for GPU, but no projects.

Keith Myers
Send message
Joined: 13 Dec 17
Posts: 1297
Credit: 5,442,861,959
RAC: 9,970,037
Level
Tyr
Scientific publications
watwatwatwatwat
Message 60999 - Posted: 15 Jan 2024 | 20:55:23 UTC

Current tasks are still the few acemd and the new Quantum Chemistry on gpu tasks.

There was a big slug of them released this morning. All gone now.

You need test applications selected and ALL applications in project preferences to get them.

https://www.gpugrid.net/forum_thread.php?id=5429

Nuadormrac
Send message
Joined: 21 Jul 12
Posts: 7
Credit: 398,051,758
RAC: 945,582
Level
Asp
Scientific publications
watwatwatwatwatwatwat
Message 61050 - Posted: 24 Jan 2024 | 5:07:46 UTC

There are WUs in progress but for about a month they always go to someone else. I wonder if I need to set BOINC to not run other projects atm, or some kind of nag addon to BOINC that makes it hammer the server for requests every minute or so. We don't seem to be getting work to those who don't have tasks already.
____________

[BAT] Svennemans
Send message
Joined: 27 May 21
Posts: 50
Credit: 332,759,517
RAC: 1,616,110
Level
Asp
Scientific publications
wat
Message 61054 - Posted: 24 Jan 2024 | 13:53:26 UTC - in response to Message 61050.

There are WUs in progress but for about a month they always go to someone else. I wonder if I need to set BOINC to not run other projects atm, or some kind of nag addon to BOINC that makes it hammer the server for requests every minute or so. We don't seem to be getting work to those who don't have tasks already.


I had the same issue, nothing for over a month despite others getting constant tasks and despite creating an aggressive 'request work' script for gpugrid.

But I think I've found the reason - and it may apply to you, or at least it couldn't harm to check.

When I enabled the 'work_fetch_debug' flag in the BoincMgr event log, I noticed messages saying "tasks won't finish on time". When I checked my computer details on gpugrid.net, the 'task duration correction factor' was set to a ridiculously high value.

Solution: removed gpugrid project from my BoincMgr, deleted the computer record in my account on gpugrid.net, reattached and hey presto! Got a workunit!

Worth a try?

Keith Myers
Send message
Joined: 13 Dec 17
Posts: 1297
Credit: 5,442,861,959
RAC: 9,970,037
Level
Tyr
Scientific publications
watwatwatwatwat
Message 61057 - Posted: 24 Jan 2024 | 19:24:44 UTC - in response to Message 61054.

Probably a holdover from previously running the acemd tasks.

You needn't have to resort to such drastic procedures of removing BOINC and the project.

The task_duration_factor is simply a parameter value in the client_state.xml file.

You could have just stopped BOINC and edited the dcf value for GPUGrid in its section.

I just set dcf values to 0.01 for all my projects. For the projects that still use the dcf parameter, any new tasks will simply recalculate from that low value to what the true dcf value for the ATMBeta tasks really are.

But be advised, the dcf value applies to ALL the applications in a project. Changing to a different app than what you were running will and can produce a large variance in the dcf value simply because one app may be more efficient than another app.

Which puts you right back into your original issue of not getting work because BOINC thinks you can't finish in time.

Solution then is to just edit client_state.xml again back to a low dcf value.

[BAT] Svennemans
Send message
Joined: 27 May 21
Posts: 50
Credit: 332,759,517
RAC: 1,616,110
Level
Asp
Scientific publications
wat
Message 61061 - Posted: 24 Jan 2024 | 20:46:17 UTC - in response to Message 61057.

Probably a holdover from previously running the acemd tasks.

You needn't have to resort to such drastic procedures of removing BOINC and the project.

The task_duration_factor is simply a parameter value in the client_state.xml file.

You could have just stopped BOINC and edited the dcf value for GPUGrid in its section.

I just set dcf values to 0.01 for all my projects. For the projects that still use the dcf parameter, any new tasks will simply recalculate from that low value to what the true dcf value for the ATMBeta tasks really are.

But be advised, the dcf value applies to ALL the applications in a project. Changing to a different app than what you were running will and can produce a large variance in the dcf value simply because one app may be more efficient than another app.

Which puts you right back into your original issue of not getting work because BOINC thinks you can't finish in time.

Solution then is to just edit client_state.xml again back to a low dcf value.


It seems to be a bit more complicated than that. I tried the client_state.xml route several times before, without any success.
Now, after that single windows task which predictably failed - see the News section topic for that - I once again get "won't finish in time" messages.
Even though my dcf is still at 1...

It could also be an issue in the expected flops/task settings for windows tasks.

In any case, when I began crunching ATM tasks some months ago, I used to get dozens of tasks without issue, then one day that changed and the "on time" issue started. Strange...

Keith Myers
Send message
Joined: 13 Dec 17
Posts: 1297
Credit: 5,442,861,959
RAC: 9,970,037
Level
Tyr
Scientific publications
watwatwatwatwat
Message 61062 - Posted: 24 Jan 2024 | 23:05:56 UTC - in response to Message 61061.

Make sure you actually have run BOINC benchmarks on your host. If you don't, the client will report the default low 1,000,000 FLOPS value for the host.

That will produce erroneous estimated calculation times.

[BAT] Svennemans
Send message
Joined: 27 May 21
Posts: 50
Credit: 332,759,517
RAC: 1,616,110
Level
Asp
Scientific publications
wat
Message 61066 - Posted: 25 Jan 2024 | 2:03:38 UTC - in response to Message 61062.
Last modified: 25 Jan 2024 | 2:05:30 UTC

Make sure you actually have run BOINC benchmarks on your host. If you don't, the client will report the default low 1,000,000 FLOPS value for the host.

That will produce erroneous estimated calculation times.


I'm at 5000+ million flops and 25000 million iops, so a fair bit higher than your computers, with a dcf of 0.01 and still get "won't finish on time" messages...

https://www.gpugrid.net/show_host_detail.php?hostid=617069

Any other theories?

Keith Myers
Send message
Joined: 13 Dec 17
Posts: 1297
Credit: 5,442,861,959
RAC: 9,970,037
Level
Tyr
Scientific publications
watwatwatwatwat
Message 61067 - Posted: 25 Jan 2024 | 3:22:00 UTC

The last time I ran a 1080 Ti, the project only had the single acemd3 executable which only took a couple of hours max to compute.

The python tasks are a different animal. I don't know offhand how long a 1080 Ti takes to compute the ATMbeta tasks.

I would have to scour the hosts and see if I could find someone else running that card to compare times.

I still run a 2080 Ti on my two hosts along with some 3080's and never have seen any instance where I couldn't return tasks within 24 hours to get the credit bonus. I think my 2080 Ti's have taken as long as 14-18 hours sometimes.

The 1080 Ti is a generation older with slower clocks and memory. I could imagine it might take over a day to compute. But still within the five day expiration limit. So never should see a "won't finish" message.

Granted Linux is always faster than Windows.

Don't have any other theories for your experience now.

Profile ServicEnginIC
Avatar
Send message
Joined: 24 Sep 10
Posts: 568
Credit: 6,848,742,024
RAC: 19,778,607
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 61068 - Posted: 25 Jan 2024 | 6:40:09 UTC - in response to Message 61066.

Any other theories?

Yes.
Try setting "Store at least XX days of work" parameter as low as possible for BOINC Manager to start requesting for GPU work.
A detailed explanation can be found at Message #54300.

[BAT] Svennemans
Send message
Joined: 27 May 21
Posts: 50
Credit: 332,759,517
RAC: 1,616,110
Level
Asp
Scientific publications
wat
Message 61071 - Posted: 25 Jan 2024 | 11:45:46 UTC - in response to Message 61067.

The last time I ran a 1080 Ti, the project only had the single acemd3 executable which only took a couple of hours max to compute.

The python tasks are a different animal. I don't know offhand how long a 1080 Ti takes to compute the ATMbeta tasks.

I would have to scour the hosts and see if I could find someone else running that card to compare times.

I still run a 2080 Ti on my two hosts along with some 3080's and never have seen any instance where I couldn't return tasks within 24 hours to get the credit bonus. I think my 2080 Ti's have taken as long as 14-18 hours sometimes.

The 1080 Ti is a generation older with slower clocks and memory. I could imagine it might take over a day to compute. But still within the five day expiration limit. So never should see a "won't finish" message.

Granted Linux is always faster than Windows.

Don't have any other theories for your experience now.


On average, my 1080Ti took 5hrs to complete an ATM WU, so I completed just shy of 5 WU's per day. This was back in October, when I still received them. So GPU age/performance is definitely not the issue. Unless of course they changed the WU size recently.

[BAT] Svennemans
Send message
Joined: 27 May 21
Posts: 50
Credit: 332,759,517
RAC: 1,616,110
Level
Asp
Scientific publications
wat
Message 61072 - Posted: 25 Jan 2024 | 12:02:28 UTC - in response to Message 61068.

Any other theories?

Yes.
Try setting "Store at least XX days of work" parameter as low as possible for BOINC Manager to start requesting for GPU work.
A detailed explanation can be found at Message #54300.


Allright! Now that is excellent advice, as I did indeed increase my work buffer at some point. And it did the trick allright, I got new WU's now.

Thanks, man!

Keith Myers
Send message
Joined: 13 Dec 17
Posts: 1297
Credit: 5,442,861,959
RAC: 9,970,037
Level
Tyr
Scientific publications
watwatwatwatwat
Message 61080 - Posted: 25 Jan 2024 | 16:56:28 UTC - in response to Message 61071.
Last modified: 25 Jan 2024 | 16:58:40 UTC

The last time I ran a 1080 Ti, the project only had the single acemd3 executable which only took a couple of hours max to compute.

The python tasks are a different animal. I don't know offhand how long a 1080 Ti takes to compute the ATMbeta tasks.

I would have to scour the hosts and see if I could find someone else running that card to compare times.

I still run a 2080 Ti on my two hosts along with some 3080's and never have seen any instance where I couldn't return tasks within 24 hours to get the credit bonus. I think my 2080 Ti's have taken as long as 14-18 hours sometimes.

The 1080 Ti is a generation older with slower clocks and memory. I could imagine it might take over a day to compute. But still within the five day expiration limit. So never should see a "won't finish" message.

Granted Linux is always faster than Windows.

Don't have any other theories for your experience now.


On average, my 1080Ti took 5hrs to complete an ATM WU, so I completed just shy of 5 WU's per day. This was back in October, when I still received them. So GPU age/performance is definitely not the issue. Unless of course they changed the WU size recently.

Yes, Steve said they did indeed increase the size of the tasks so they run longer now. Was only 50 atoms originally and is now doubled to 100.

[BAT] Svennemans
Send message
Joined: 27 May 21
Posts: 50
Credit: 332,759,517
RAC: 1,616,110
Level
Asp
Scientific publications
wat
Message 61091 - Posted: 26 Jan 2024 | 0:22:49 UTC - in response to Message 61080.


Yes, Steve said they did indeed increase the size of the tasks so they run longer now. Was only 50 atoms originally and is now doubled to 100.


I think that was even earlier, before October, because judging from the WU I coerced into running now (through some job.xml/run.bat shenanigans ;-) ) I will still finish in less than 5 hours.

1 hour runtime now and already at 22%

Post to thread

Message boards : Frequently Asked Questions (FAQ) : No tasks in quite a while.

//