Guide to running multiple GPU work units

[Jstn7477]

Did I understand correctly that the new WU should use more of the card without the "tweaks"? Would be nice considering my 7950 would get at least 3x the points it is getting right now, hopefully. I want my slice of pie back already!

[KieX]

Originally Posted by hat

It behaved normally, like the app_info file wasn't there. Oh well, if there's going to be an update that does this without finicky tweaking then I'll just wait for that.

Hmm. Is the file name coming up in the log? If not, then either you need to remove and re-attach WCG from BOINC (finish all WU or lose them) or re-check the file is in the right folder/name/file type..

Originally Posted by Jstn7477

Did I understand correctly that the new WU should use more of the card without the "tweaks"? Would be nice considering my 7950 would get at least 3x the points it is getting right now, hopefully. I want my slice of pie back already!

The BETAs look to have 2 jobs to process but not simultaneously. Up to the 50% stage is one job, and then the second job runs up to the 100% completion. I would guess that app_info still offers more performance as a result.

But if it keeps sending the BETA WU to your BOINC with app_info, your HCC GPU may run dry

[[Ion]]

I've picked up a few of the BETAs in my queue--although, strangely enough, they aren't taking high priority over the standard ATI GPU WUs like betas usually do.

[F150_Raptor]

Well guys, my buffer went completely dry and won't download any wu's with the app_info file. I've been running 2 wu's per gpu for a while. I set it back to stock, only 1 wu per gpu. I don't know if it has some thing to do with the beta's. A little is better than none though.

[animal007uk]

Strange did i miss something? Been running none stop for the last few days and work units keep coming in, Not changed anything in the settings either apart from having network usage set to always avalible.

[mstenholm]

Originally Posted by F150_Raptor

Well guys, my buffer went completely dry and won't download any wu's with the app_info file.

I had the same problem and so did I guess everybody else who had run beta-testing ticked off and used the app_info file. I got one rig back to normal (3 WUs) but is still struggling with the other.

[KieX]

I found the app_info code for the BETA on the problem discussion thread. I've got an 8hr- long queue to complete before I can test this though. This would resolve the no new WU for those of us affected.

The concept is simple though. Reset project, before getting new tasks add the code below to your app_info, then restart BOINC:

Spoiler

<app>
<name>beta3</name>
<user_friendly_name>Help Conquer Cancer</user_friendly_name>
</app>
<file_info>
<name>wcg_beta3_img_7.05_windows_intelx86__ati_hcc 1</name>
<executable/>
</file_info>
<file_info>
<name>hcckernel.cl.7.05</name>
<executable/>
</file_info>
<app_version>
<app_name>beta3</app_name>
<version_num>705</version_num>
<platform>windows_intelx86</platform>
<plan_class>ati_hcc1</plan_class>
<avg_ncpus>1.0</avg_ncpus>
<max_ncpus>1.0</max_ncpus>
<coproc>
<type>ATI</type>
<count>1</count>
</coproc>
<file_ref>
<file_name>wcg_beta3_img_7.05_windows_intelx86__at i_hcc1</file_name>
<main_program/>
</file_ref>
<file_ref>
<file_name>hcckernel.cl.7.05</file_name>
<open_name>hcckernel.cl</open_name>
</file_ref>
</app_version>

If anyone wants to experiment, let us know the results. Else I'll try this out later on and post working app_info files to the OP

[t_ski]

both of mine are still churning away

[Norton]

The Beta was to be a very small batch from what I read on the WCG website.... so hopefully it doesn't cause anyone too much trouble. I only have 1 wu left on one of my rigs that hasn't been completed yet (2 pages of them processed last night)

Wouldn't unchecking Beta testing on the WCG site keep these from causing issues?

[F150_Raptor]

Originally Posted by KieX

I found the app_info code for the BETA on the problem discussion thread. I've got an 8hr- long queue to complete before I can test this though. This would resolve the no new WU for those of us affected.

The concept is simple though. Reset project, before getting new tasks add the code below to your app_info, then restart BOINC:



If anyone wants to experiment, let us know the results. Else I'll try this out later on and post working app_info files to the OP

I took me an hour the last time I added the app_info file into the rig to get it to work. I would get no work at all, only the gpu's and no cpu work, or a problem where only 1 of the 3 cards would work. It's just a pain in the donkey. I was tired last night so I just took the file out, instead of getting frustrated at it. I might try that app_info later though, Thanks.

Originally Posted by Norton

The Beta was to be a very small batch from what I read on the WCG website.... so hopefully it doesn't cause anyone too much trouble. I only have 1 wu left on one of my rigs that hasn't been completed yet (2 pages of them processed last night)

Wouldn't unchecking Beta testing on the WCG site keep these from causing issues?

Unchecking the beta would work, but for how long. If you have to keep changing the file every time some changes, it's not worth it. You lose what you have in your buffer every time, which is work that has to be redistributed through the system again. They need to create a new version of boinc that would just allow us to change how many work units we want to run on the gpu's on the fly. Then we'd have no lost work, and could run more(or other) projects easily. Hcc is winning but the other projects are losing bad from this.

[mstenholm]

Originally Posted by F150_Raptor

I took me an hour the last time I added the app_info file into the rig to get it to work. I would get no work at all, only the gpu's and no cpu work, or a problem where only 1 of the 3 cards would work. It's just a pain in the donkey. I was tired last night so I just took the file out, instead of getting frustrated at it. I might try that app_info later though, Thanks.



Unchecking the beta would work, but for how long. If you have to keep changing the file every time some changes, it's not worth it. You lose what you have in your buffer every time, which is work that has to be redistributed through the system again. They need to create a new version of boinc that would just allow us to change how many work units we want to run on the gpu's on the fly. Then we'd have no lost work, and could run more(or other) projects easily. Hcc is winning but the other projects are losing bad from this.

This morning (Europe time) I saw a patch on WGC forum that would let beta work run with app_info. I can't find it now and I don't know if it works but I'm sure next time a new batch is released the discussion will be on again.

BTW I got my other AMD card to play nice again but I'm sure that there is a lot of people that are running high-priority GFAM projects now

[KieX]

Originally Posted by F150_Raptor

It's just a pain in the donkey. I was tired last night so I just took the file out, instead of getting frustrated at it. I might try that app_info later though, Thanks.

Unchecking the beta would work, but for how long. If you have to keep changing the file every time some changes, it's not worth it. You lose what you have in your buffer every time, which is work that has to be redistributed through the system again. They need to create a new version of boinc that would just allow us to change how many work units we want to run on the gpu's on the fly. Then we'd have no lost work, and could run more(or other) projects easily. Hcc is winning but the other projects are losing bad from this.

Totally agree with you. I went back to default with this too. From what I read it's down to the way the WCG servers are setup. In most other projects the app_info sends information back to the server so the project computers know what to send you. WCG.. haven't implemented this which is why some of us luck out when we're spammed with WU we didn't request.

Hopefully they'll take users feedback on-board because as you said, this is unfortunately not in the best interest of all the other sciences. Will work on this app_info but if it carries on giving hassle will go back to the default.

Originally Posted by mstenholm

This morning (Europe time) I saw a patch on WGC forum that would let beta work run with app_info. I can't find it now and I don't know if it works but I'm sure next time a new batch is released the discussion will be on again.

BTW I got my other AMD card to play nice again but I'm sure that there is a lot of people that are running high-priority GFAM projects now

I think the patch you mention is the one I posted above. It's something you add to a current app_info file to add the project. Saw you were browsing the same thing this morning (not stalking )

[mstenholm]

Originally Posted by KieX

Totally agree with you. I went back to default with this too. From what I read it's down to


I think the patch you mention is the one I posted above. It's something you add to a current app_info file to add the project. Saw you were browsing the same thing this morning (not stalking )

Yes that was the one and I should have added it right away. They just sent out some betas again and I got them on my app_info PCs. Clear all and restart, 1 hour lost. Now, at least untill I get the beta add-on running I'm not in the beta program. The patch will only save you from this beta run (HCC 7.05) but not from whatever they might send you in the future.

[KieX]

Originally Posted by mstenholm

And I should have added it right away. They just sent out some betas again and I got them on my app_info PCs. Clear all and restart, 1 hour lost. Now, at least until I get the beta add-on, I'm not in the beta program.

I was being sent BETA even with it deselected Anyway, just 2hr left before my cache clears, then will try out creating a new app_info with those BETA.

[Norton]

Originally Posted by F150_Raptor

Unchecking the beta would work, but for how long. If you have to keep changing the file every time some changes, it's not worth it. You lose what you have in your buffer every time, which is work that has to be redistributed through the system again. They need to create a new version of boinc that would just allow us to change how many work units we want to run on the gpu's on the fly. Then we'd have no lost work, and could run more(or other) projects easily. Hcc is winning but the other projects are losing bad from this.

I was just suggesting unchecking the Beta work as a temporary measure until a good work-around could be figured out (didn't realize that you lose what's in your buffer when you do that).

I'm actually running all projects and only one HCC gpu job on my 7870 and 6870. I'm cool with the 25k or so of extra ppd that's giving me but like you said, didn't want the other work to miss out on not getting done

[manofthem]

Well, I just ran out of work, not sure why. I have everything set to how it's been set for days, and it literally just happened within the last 20 minutes. Weird

[the54thvoid]

I'm also getting no tasks now.

Message file reads:

[error] No application found for task: windows_intelx86 705 ati_hcc1; discarding

[KieX]

***Update on project:

Greetings,

We have updated the science application for Help Conquer Cancer. This application change will allow work units to process multiple images. The purpose of this change is to increase the average runtime of the work units and decrease the server scheduler requests.

This will be a two part change.

1. The application will be released and process single image work units.
2. In about 3 days we will start sending multiple image work units.

The application version being loaded is 7.05.

Thanks,
-Uplinger

In order for the current app_info files to work:

1. copy app info file outside the current location (e.g. to the Desktop)
2. change all 6.56 to 7.05 and 656 to 705 in it
3. detach all 6.56 WU's
4. detach and re-attach WCG
4.a. detach all new WU's
5. close BOINC
6. copy updated app info file to its folder
7. start BOINC

Don't have the time now, but I'll update all the OP files with that. For now.. please try the above to resolve the issue. It's just a matter of the version of WU specified in app_info being updated.

Testing patience to the limits
Source: http://www.worldcommunitygrid.org/fo...d_thread,34225

[manofthem]

Originally Posted by the54thvoid

I'm also getting no tasks now.

Message file reads:

[error] No application found for task: windows_intelx86 705 ati_hcc1; discarding

http://img.techpowerup.org/121109/Untitled.png

Same errors I started getting

Originally Posted by KieX

***Update on project:



In order for the current app_info files to work:



Don't have the time now, but I'll update all the OP files with that. For now.. please try the above to resolve the issue. It's just a matter of the version of WU specified in app_info being updated.

Testing patience to the limits
Source: http://www.worldcommunitygrid.org/fo...d_thread,34225

I'll give this a shot too, thanks!

[animal007uk]

Seems to work took a few trys.

[mstenholm]

I got the HCC GPU only to work on one rig. Tried several times to get a CPU/GPU to run but got errors and no downloads.

[manofthem]

So is everyone still running mutli WUs or not? Mine stopped, as the54thvoid showed his errors. So I just did the steps that KieX described above, editing the app_info file with the 7.05 info. Now it's acting like it does automatically, no multi work units, like there's no app_info file. Am I doing something wrong or missing something?

Edit: So I'm not sure what i did, maybe close it and recopy file, but it seems to be working. Who's to say how long it will work though
Let's hope that it keep sending me the work! After it all levels out and it goes for awhile, I'll have a better idea. Thanks KieX for postnig the fix!

[F150_Raptor]

Originally Posted by Norton

I was just suggesting unchecking the Beta work as a temporary measure until a good work-around could be figured out (didn't realize that you lose what's in your buffer when you do that).

I was referring to changing the app_info file, not unchecking beta's.

[ThE_MaD_ShOt]

Mine is down too. Trying to work on it now.

Screw it I really don't have the patience to try and figure it out right now. Maybe next year or something. And since my numbers are going to fall anyway I am going to take some rigs down next week.

[t_ski]

Mine was down when I got home, and got the same errors. Trying your fix now.

EDIT: Just relaunched the client and I get two notices:

11/9/2012 9:09:52 PM | World Community Grid | File referenced in app_info.xml does not exist: wcg_hcc1_img_7.05_windows_intelx86__ati_hcc1
11/9/2012 9:09:52 PM | World Community Grid | File referenced in app_info.xml does not exist: hcckernel.cl.7.05