Advanced search

Message boards : Graphics cards (GPUs) : Only getting 1 GPU work at a time even though set for 10 days of work.

Author Message
Sandman192
Send message
Joined: 26 Oct 09
Posts: 13
Credit: 32,914,359
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwatwatwatwatwatwat
Message 54285 - Posted: 10 Apr 2020 | 9:06:06 UTC

I only have GPU Grid to grab GPU work and set for 10 days of storing work and BOINC is only getting 1 until at a time until it's done.

I never had problems when I was using Prime GPU work.

Plus, I had no GPU work running so my GPU wasn't working on anything.
I get "Task won't finish it time" when it was trying to? I using a 1080Ti. I finally got work but it's just 1 WU. A Ti and set for 24H and it won't finish on time?

You have "New version of ACEMD" only running right now and I assume that's beta?

4/10/2020 3:05:25 AM | GPUGRID | Sending scheduler request: To fetch work.
4/10/2020 3:05:25 AM | GPUGRID | Requesting new tasks for NVIDIA GPU
4/10/2020 3:05:28 AM | GPUGRID | Scheduler request completed: got 0 new tasks
4/10/2020 3:05:28 AM | GPUGRID | No tasks sent
4/10/2020 3:05:28 AM | GPUGRID | Tasks won't finish in time: BOINC runs 96.5% of the time; computation is enabled 99.4% of that

Profile ServicEnginIC
Avatar
Send message
Joined: 24 Sep 10
Posts: 566
Credit: 5,940,852,024
RAC: 10,923,673
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 54286 - Posted: 10 Apr 2020 | 10:06:09 UTC - in response to Message 54285.

I only have GPU Grid to grab GPU work and set for 10 days of storing work and BOINC is only getting 1 until at a time until it's done.
.
I get "Task won't finish it time"...

As all GPUGrid's WUs have a deadline of 5 days, setting a higher buffer will cause BOINC Manager to "think" that any further tasks won't finish in time.
If you set buffer to 4,5 days or lower (for example), you'll probably receive an additional WU (two per GPU is the current maximum at GPUGrid)

Sandman192
Send message
Joined: 26 Oct 09
Posts: 13
Credit: 32,914,359
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwatwatwatwatwatwat
Message 54290 - Posted: 10 Apr 2020 | 22:56:05 UTC - in response to Message 54286.
Last modified: 10 Apr 2020 | 23:35:34 UTC

If you set buffer to 4,5 days or lower (for example), you'll probably receive an additional WU (two per GPU is the current maximum at GPUGrid)

That's not how that works when I have no GPU work to run. If I have it set for 10 days then it would only get at least 2 5 days of WUs to total 10 days. And not to have any work at all.

Plus, when I was using prime I got loads of work, even the ones that took over 5 days to complete along with still extra work for the GPU alone.

mikey
Send message
Joined: 2 Jan 09
Posts: 291
Credit: 2,040,791,115
RAC: 10,382,191
Level
Phe
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 54291 - Posted: 10 Apr 2020 | 23:03:30 UTC - in response to Message 54290.

If you set buffer to 4,5 days or lower (for example), you'll probably receive an additional WU (two per GPU is the current maximum at GPUGrid)

That's not how that works. If I have it set for 10 days then it would only get at least 2 5 days of WUs to total 10 days. And not to have any work at all.

Plus, when I was using prime I got loads of work, even the ones that took over 5 days to complete along with still extra work for the GPU alone.


The PROJECT has a 2 workunit per gpu maximum so they can maniplulate the next set of workunits quicker, your work affects the next set of workunits.

Each Boinc Project does things differently, it's not a clone thing.

mikey
Send message
Joined: 2 Jan 09
Posts: 291
Credit: 2,040,791,115
RAC: 10,382,191
Level
Phe
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 54292 - Posted: 10 Apr 2020 | 23:03:45 UTC - in response to Message 54290.

If you set buffer to 4,5 days or lower (for example), you'll probably receive an additional WU (two per GPU is the current maximum at GPUGrid)

That's not how that works. If I have it set for 10 days then it would only get at least 2 5 days of WUs to total 10 days. And not to have any work at all.

Plus, when I was using prime I got loads of work, even the ones that took over 5 days to complete along with still extra work for the GPU alone.


The PROJECT has a 2 workunit per gpu maximum so they can maniplulate the next set of workunits quicker, your work affects the next set of workunits.

Each Boinc Project does things differently, it's not a clone thing.

mikey
Send message
Joined: 2 Jan 09
Posts: 291
Credit: 2,040,791,115
RAC: 10,382,191
Level
Phe
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 54293 - Posted: 10 Apr 2020 | 23:08:27 UTC - in response to Message 54290.

If you set buffer to 4,5 days or lower (for example), you'll probably receive an additional WU (two per GPU is the current maximum at GPUGrid)

That's not how that works. If I have it set for 10 days then it would only get at least 2 5 days of WUs to total 10 days. And not to have any work at all.

Plus, when I was using prime I got loads of work, even the ones that took over 5 days to complete along with still extra work for the GPU alone.


The PROJECT has a 2 workunit per gpu maximum so they can maniplulate the next set of workunits quicker, your work affects the next set of workunits.

Each Boinc Project does things differently, it's not a clone thing.

mikey
Send message
Joined: 2 Jan 09
Posts: 291
Credit: 2,040,791,115
RAC: 10,382,191
Level
Phe
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 54294 - Posted: 10 Apr 2020 | 23:09:12 UTC - in response to Message 54290.

If you set buffer to 4,5 days or lower (for example), you'll probably receive an additional WU (two per GPU is the current maximum at GPUGrid)

That's not how that works. If I have it set for 10 days then it would only get at least 2 5 days of WUs to total 10 days. And not to have any work at all.

Plus, when I was using prime I got loads of work, even the ones that took over 5 days to complete along with still extra work for the GPU alone.


The PROJECT has a 2 workunit per gpu maximum so they can maniplulate the next set of workunits quicker, your work affects the next set of workunits.

Each Boinc Project does things differently, it's not a clone thing.

Sandman192
Send message
Joined: 26 Oct 09
Posts: 13
Credit: 32,914,359
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwatwatwatwatwatwat
Message 54295 - Posted: 11 Apr 2020 | 0:34:01 UTC - in response to Message 54293.

Mikey Umm... You posted 4 times.

The PROJECT has a 2 workunit per gpu maximum so they can maniplulate the next set of workunits quicker, your work affects the next set of workunits.

You still haven't explained why I have NO WUs as I stated before and GPUGRID says it can't get any more work because it can't finish on time.
I have NO WU's at all and GPUGRID can't get WUs due to deadline issues on a GTX 1080. What? And that 4 to 5 day set does not fix the problem.

Each Boinc Project does things differently, it's not a clone thing.

What??? Each project from any site still has the same effect. BOINC goes by how long it would take to finish and when's the deadline is. There are no differents on any WU from any worksite when it comes to time and deadline.

I think the deadline for some GPUGRID WUs is broken. I've seen it before in other projects other then GPUGRID where no one could get WUs because of a deadline error.

klepel
Send message
Joined: 23 Dec 09
Posts: 189
Credit: 4,196,461,293
RAC: 1,617,116
Level
Arg
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 54297 - Posted: 11 Apr 2020 | 2:15:58 UTC - in response to Message 54285.

I only have GPU Grid to grab GPU work and set for 10 days of storing work and BOINC is only getting 1 until at a time until it's done.

There is now reason to write in bold and big letters, we all try to help to solve problems!

Question: How is you priority set for GPUGRID? Might it be, that it is lower than priorities on your CPU projects? In my expirience BOINC gets confused sometimes when the priority for GPU work is lower than CPU work and does not download GPU WUs.

GPUGRID works best with a short workbuffer, I get the next WU about 30 minutes before the last ends.

And yes! GPUGRID needs shortest turn-around times possible, so you are asked not to bunker WUs, therefore the number of GPUGRID is limited to 2 per GPU at any given time.

Sandman192
Send message
Joined: 26 Oct 09
Posts: 13
Credit: 32,914,359
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwatwatwatwatwatwat
Message 54298 - Posted: 11 Apr 2020 | 4:52:58 UTC - in response to Message 54297.

There is now reason to write in bold and big letters, we all try to help to solve problems!

Apparently I do since no one reads or thinks before they respond.
I said I have NO GPU work running and I get this response. How is this relevant?

If you set buffer to 4,5 days or lower (for example), you'll probably receive an additional WU (two per GPU is the current maximum at GPUGrid)
No, that doesn't fix the problem. I get no work from time to time and my GPU sets there not working on anything. Changing settings does NOT fix a thing. Been there done that with other sites that had this same issue.

Question: How is you priority set for GPUGRID? Might it be, that it is lower than priorities on your CPU projects? In my experience BOINC gets confused sometimes when the priority for GPU work is lower than CPU work and does not download GPU WUs.

Since it's the only GPU work I'm requesting then it doesn't matter. If BOINC gets "confused" on running 1 GPU work site like GPUGRID then BOINC has a big bug in it. I can change it to 10,000 and that won't fix a thing. Again been there done that with other sites that had the same issue and they had to fix it on their side. This is nothing new on deadline issues.

GPUGRID works best with a short workbuffer, I get the next WU about 30 minutes before the last ends.

No, it doesn't. I've been doing this for years and this is a first.

And yes! GPUGRID needs shortest turn-around times possible, so you are asked not to bunker WUs, therefore the number of GPUGRID is limited to 2 per GPU at any given time.

What are you talking about? Are you saying my 1080 Ti can't do long turn arounds? Not to mention it's overclocked. That has nothing to do with the setting you give me to change.

Profile tito
Send message
Joined: 21 May 09
Posts: 16
Credit: 1,057,958,678
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwat
Message 54299 - Posted: 11 Apr 2020 | 7:35:46 UTC - in response to Message 54298.

Try other Boinc Manager. Maybe Your version has some problems.
Personally I use 7.12.1 without any issues.

Profile ServicEnginIC
Avatar
Send message
Joined: 24 Sep 10
Posts: 566
Credit: 5,940,852,024
RAC: 10,923,673
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 54300 - Posted: 11 Apr 2020 | 9:48:43 UTC - in response to Message 54295.

You still haven't explained why I have NO WUs as I stated before and GPUGRID says it can't get any more work because it can't finish on time.

Sometimes things are the way they are, not the way they should be...
I'll show you the way I like, with a practical example:

Usually my BOINC Manager's Computing preferences are set like this, for not to get too much tasks at queue.
In this situation, my list of running/Ready to start tasks is like this.
Then, if I change preferences to "Store at least 9.05 days of work", I get a message like the following:

Sat 11 Apr 2020 09:28:04 WEST | GPUGRID | Sending scheduler request: To fetch work.
Sat 11 Apr 2020 09:28:04 WEST | GPUGRID | Requesting new tasks for CPU and NVIDIA GPU
Sat 11 Apr 2020 09:28:06 WEST | GPUGRID | Scheduler request completed: got 0 new tasks
Sat 11 Apr 2020 09:28:06 WEST | GPUGRID | No tasks sent
Sat 11 Apr 2020 09:28:06 WEST | GPUGRID | Tasks won't finish in time: BOINC runs 100.0% of the time; computation is enabled 100.0% of that

The same you are complaining about.
And if I set preferences to "Store at least 4.05 days of work", the message changes to the following:

Sat 11 Apr 2020 09:29:55 WEST | GPUGRID | update requested by user
Sat 11 Apr 2020 09:30:00 WEST | GPUGRID | Sending scheduler request: Requested by user.
Sat 11 Apr 2020 09:30:00 WEST | GPUGRID | Requesting new tasks for CPU and NVIDIA GPU
Sat 11 Apr 2020 09:30:02 WEST | GPUGRID | Scheduler request completed: got 1 new tasks
Sat 11 Apr 2020 09:30:04 WEST | GPUGRID | Started download of 4zohA04_320_3-TONI_MDADpr4sz-4-conf_file_enc
Sat 11 Apr 2020 09:30:04 WEST | GPUGRID | Started download of 4zohA04_320_3-TONI_MDADpr4sz-4-4zohA04_320_3-TONI_MDADpr4sz-3-10-RND7698_1
.
.
.
Sat 11 Apr 2020 09:30:11 WEST | GPUGRID | Started download of 4zohA04_320_3-TONI_MDADpr4sz-4-prmtop_file
Sat 11 Apr 2020 09:30:12 WEST | GPUGRID | Finished download of 4zohA04_320_3-TONI_MDADpr4sz-4-prmtop_file
Sat 11 Apr 2020 09:30:34 WEST | GPUGRID | Sending scheduler request: To fetch work.
Sat 11 Apr 2020 09:30:34 WEST | GPUGRID | Requesting new tasks for CPU and NVIDIA GPU
Sat 11 Apr 2020 09:30:35 WEST | GPUGRID | Scheduler request completed: got 0 new tasks
Sat 11 Apr 2020 09:30:35 WEST | GPUGRID | No tasks sent
Sat 11 Apr 2020 09:30:35 WEST | GPUGRID | This computer has reached a limit on tasks in progress

After that, my list of running/Ready to start tasks is like this.
I've got an extra GPUGrid's WU, and then no more, due to current policy of maximum two tasks per GPU. This is so for me, and for every other GPUGrid's crunchers...

I hope this can help you to manage your preferences to your best criteria.
Kind regards,

rod4x4
Send message
Joined: 4 Aug 14
Posts: 266
Credit: 2,219,935,054
RAC: 0
Level
Phe
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 54301 - Posted: 11 Apr 2020 | 11:04:21 UTC - in response to Message 54300.
Last modified: 11 Apr 2020 | 11:04:58 UTC

Excellent explanation ServicEnginIC

You went above and beyond to explain the problems reported by OP.

Accurate and fully details why OP settings don't work with GPUGRID.

Hat tip to other responders who's posts were correct and supplied with desire to provide genuine assistance (that was rebuffed).

We are all here to assist this project and fellow volunteers. Hopefully the OP recognises this point.

Sandman192
Send message
Joined: 26 Oct 09
Posts: 13
Credit: 32,914,359
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwatwatwatwatwatwat
Message 54306 - Posted: 12 Apr 2020 | 0:39:31 UTC

Ok, so what you are all saying is that.

1. After having no work done that GPUGRID is going to say "No work for you anyway and sorry your GPU is not going to work on anything". I said that before in my first statement. Even as I'm posting this GPUGRID has 242,886 WUs to send. Ya, no bug there. Wink Wink.

2. This is new. I only get 2 WU that only takes 2 hours to work on at a time. I NEVER had that problem with other BOINC project sites that have large (2 to 6 days which I bearly get and still running out) or small WU (2 to 10 hours).

3. I only have GPUGRID for my GPU only. No other work for it.

Don't post here unless you work for GPUGRID. Sorry, I really don't want to be rude but... NONE of you know what you are talking about.
Expert for what is OP?

Ian&Steve C.
Avatar
Send message
Joined: 21 Feb 20
Posts: 1031
Credit: 35,522,207,483
RAC: 66,331,002
Level
Trp
Scientific publications
wat
Message 54307 - Posted: 12 Apr 2020 | 1:22:56 UTC - in response to Message 54306.
Last modified: 12 Apr 2020 | 1:33:21 UTC

you wont ever get more than 2 WUs at a time on your system. as has been said before, the project enforces a maximum of 2 WUs per GPU. since you only have 1 GPU, you only will ever have 2 max. that's it, you can't get more without adding more GPUs. I would recommend you set your cache level to something like 1 day. that should be small enough to not get the error about time limit, and more than large enough to keep you at 2 WUs.

also keep in mind that since SETI shut down, this project has seen a large influx of the number of hosts, which increases load on the project servers. causing timeouts and trouble uploading and downloading new work. sometimes in this situation BOINC will put your system into a backoff mode and not request new work if you have stuck transfers or if you have had several unsuccessful connections. this can lead to the your GPU sitting idle for a while until the backoff expires and it tries to establish a connection with the project again.

you can use the boinccmd program to periodically retry stuck transfers and re-initiate schedule requests to minimize extended backoffs. that's what I do.

also, historically GPUGRID has not had consistent work available, that has changed recently, and there has been steady work here, but it could go away again. if you are worried about your GPU sitting idle, perhaps you should setup a backup project to failover to when you dont get work from GPUGRID.
____________

Profile Retvari Zoltan
Avatar
Send message
Joined: 20 Jan 09
Posts: 2343
Credit: 16,201,255,749
RAC: 7,520
Level
Trp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 54309 - Posted: 12 Apr 2020 | 11:12:07 UTC - in response to Message 54306.
Last modified: 12 Apr 2020 | 11:23:39 UTC

Don't post here unless you work for GPUGRID.
There is no such authorization on a public forum which lets anyone select the people who are allowed to post on a public thread.
This rule you think you're authorized to set includes you, so it wouldn't allow even you to post in your own thread.
By the way all of the people who answered your question "work" for GPUGrid, as we are all the volunteers of this project, having better knowledge of how to crunch for it than the staff.
Don't expect that the staff will answer you, as there are ten thousands of hosts working fine, so the bug is inevitably on your side.
Sorry, I really don't want to be rude but... NONE of you know what you are talking about.
A man was driving down highway 407 when a radio announcer said: “be warned of a car driving the wrong way along the highway”
The man, peeking out the window, scoffs to himself as he thinks “just one? All these idiots are going the wrong way!”
Expert for what is OP?
You've been given expert advice before. Now all you have to do is follow them, and report back the result.
If they don't work, you can detach from the projects you don't crunch for, and reset the rest.

Profile Retvari Zoltan
Avatar
Send message
Joined: 20 Jan 09
Posts: 2343
Credit: 16,201,255,749
RAC: 7,520
Level
Trp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 54310 - Posted: 12 Apr 2020 | 11:13:44 UTC - in response to Message 54306.
Last modified: 12 Apr 2020 | 11:16:26 UTC

The forum is sluggish.

Aurum
Avatar
Send message
Joined: 12 Jul 17
Posts: 399
Credit: 13,024,100,382
RAC: 766,238
Level
Trp
Scientific publications
watwatwat
Message 54321 - Posted: 14 Apr 2020 | 16:11:54 UTC
Last modified: 14 Apr 2020 | 17:06:40 UTC

I didn't see where the sandman said what BOINC CPU project he's running. If the queue is set to 10/? days to get lots of CPU WUs then they start to run high priority when their deadline is 48 hours out. If BOINC needs to dedicate to Running High Priority CPU WUs it will not use the CPUs to support a GPU WU. I set my queue to 1.0/0.1 days and have a constant flow of work.

I don't know where BOINCmgr tells you it's Running High Priority. In BoincTasks it's prominently displayed in the Status column on the Tasks tab.

A quick fix is to Suspend all waiting CPU WUs.

Sandman192
Send message
Joined: 26 Oct 09
Posts: 13
Credit: 32,914,359
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwatwatwatwatwatwat
Message 54331 - Posted: 15 Apr 2020 | 3:57:13 UTC - in response to Message 54321.

I didn't see where the sandman said what BOINC CPU project he's running. If the queue is set to 10/? days to get lots of CPU WUs then they start to run high priority when their deadline is 48 hours out. If BOINC needs to dedicate to Running High Priority CPU WUs it will not use the CPUs to support a GPU WU. I set my queue to 1.0/0.1 days and have a constant flow of work.

I don't know where BOINCmgr tells you it's Running High Priority. In BoincTasks it's prominently displayed in the Status column on the Tasks tab.

A quick fix is to Suspend all waiting CPU WUs.


I don't have to show you on what CPU work I'm running. That's irrelevant.

I have no work from GPUGRID at all and I get this "Tasks won't finish in time:" on a GTX 1080 Ti.

There is still work to send with 242,857 waiting to send.

And changing the queue changes nothing. Also irrelevant.

4/14/2020 9:12:27 PM | GPUGRID | Project requested delay of 31 seconds
4/14/2020 9:13:01 PM | GPUGRID | No tasks sent
4/14/2020 9:13:01 PM | GPUGRID | Project requested delay of 31 seconds
4/14/2020 9:20:33 PM | GPUGRID | Project requested delay of 31 seconds
4/14/2020 9:21:06 PM | GPUGRID | No tasks sent
4/14/2020 9:21:06 PM | GPUGRID | Tasks won't finish in time: BOINC runs 97.7% of the time; computation is enabled 99.6% of that
4/14/2020 9:21:06 PM | GPUGRID | Project requested delay of 31 seconds
4/14/2020 9:49:26 PM | GPUGRID | No tasks sent
4/14/2020 9:49:26 PM | GPUGRID | Project requested delay of 31 seconds
4/14/2020 9:59:40 PM | GPUGRID | No tasks sent
4/14/2020 9:59:40 PM | GPUGRID | Project requested delay of 31 seconds
4/14/2020 10:17:47 PM | GPUGRID | No tasks sent
4/14/2020 10:17:47 PM | GPUGRID | Project requested delay of 31 seconds

Profile Retvari Zoltan
Avatar
Send message
Joined: 20 Jan 09
Posts: 2343
Credit: 16,201,255,749
RAC: 7,520
Level
Trp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 54344 - Posted: 16 Apr 2020 | 0:37:15 UTC - in response to Message 54331.
Last modified: 16 Apr 2020 | 1:02:37 UTC

I don't have to show you on what CPU work I'm running. That's irrelevant.
There's only one queue for workunits in BOINC.
There's no separate queue for CPU tasks and GPU tasks.
That's why it's relevant how much CPU work you have queued, how long their (estimated) aggregated processing time is.

I have no work from GPUGRID at all and I get this "Tasks won't finish in time:" on a GTX 1080 Ti.
The reason for this message is that your work queue is full.

And changing the queue changes nothing. Also irrelevant.
You are wrong.
Reducing the queue length won't have an immediate effect, as after reducing the queue length the workunits still will be there, filling your queue multiple times.
You should do the following:
1. reduce your cache settings to 1.0+0.1 days
2. Set "No new tasks" for other projects temporarily (to prevent them refilling your queue)
3. Abort as many CPU tasks needed to get well below 1 day (per CPU cores)

Note that you won't receive more than 2 tasks per GPU from GPUGrid. (Regardless of your queue length settings.)

Some projects tend to "overfill" the queue inhibiting other projects from sending work. The other factor is that initially the BOINC manager doesn't know exactly how long it takes to process a GPUGrid WU on your GPU. Probably it overestimates the time needed to process one GPUGrid workunit, so you may have to suspend your CPU projects (perhaps even abort all of their tasks) for a couple of days, until the BOINC client establishes the real processing rate of GPUGrid workunits on your GPU.

Post to thread

Message boards : Graphics cards (GPUs) : Only getting 1 GPU work at a time even though set for 10 days of work.

//