2 workers on 1 GPU

Message boards : Number crunching : 2 workers on 1 GPU
Message board moderation

To post messages, you must log in.

1 · 2 · Next

AuthorMessage
Dimitar Stoynev
Avatar

Send message
Joined: 7 Jan 09
Posts: 19
Credit: 336,531
RAC: 0
Bulgaria
Message 1899397 - Posted: 5 Nov 2017, 20:52:37 UTC
Last modified: 5 Nov 2017, 20:53:02 UTC

Hi everyone,

I'm trying to run 2 workers on 1 AMD GPU. For some reason I always get this error:

11/5/2017 10:46:20 PM | SETI@home | Entry in app_config.xml for app 'setiathome_v8', plan class '' doesn't match any app versions

Here is the app_config.xml:

<app_config>
    <app_version>
        <app_name>setiathome_v8</app_name>
        <avg_ncpus>0.5</avg_ncpus>
        <ngpus>0.5</ngpus>
    </app_version>
</app_config>

Many guys on the webz claim this config is working for them, and looking through BOINC's wiki it looks like a valid config file to me.

Any ideas what's wrong?
ID: 1899397 · Report as offensive
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5517
Credit: 528,817,460
RAC: 242
United States
Message 1899399 - Posted: 5 Nov 2017, 20:59:20 UTC - in response to Message 1899397.  
Last modified: 5 Nov 2017, 20:59:37 UTC


Any ideas what's wrong?


Might want to unhide your computers so we can get a better look at your work results. Without knowing that, what kind of work units are you running? Cuda, SoG? Mix?
ID: 1899399 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13720
Credit: 208,696,464
RAC: 304
Australia
Message 1899408 - Posted: 5 Nov 2017, 21:56:45 UTC

I thought AMD hardware always had issues running more than 1 Seti WU at a time on a video card?
Grant
Darwin NT
ID: 1899408 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1899409 - Posted: 5 Nov 2017, 21:59:31 UTC - in response to Message 1899397.  

11/5/2017 10:46:20 PM | SETI@home | Entry in app_config.xml for app 'setiathome_v8', plan class '' doesn't match any app versions

Here is the app_config.xml:

<app_config>
    <app_version>
        <app_name>setiathome_v8</app_name>
        <avg_ncpus>0.5</avg_ncpus>
        <ngpus>0.5</ngpus>
    </app_version>
</app_config>
Have another look at the manual. Although it says the <plan_class> is optional, I'd suggest it's required if not null - and all GPU apps have a plan_class. It's the only thing which can identify which <app_version> you're working on.

Please report back, and I'll update the wiki if that's the ambiguity.
ID: 1899409 · Report as offensive
Profile Brent Norman Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester

Send message
Joined: 1 Dec 99
Posts: 2786
Credit: 685,657,289
RAC: 835
Canada
Message 1899416 - Posted: 5 Nov 2017, 22:27:27 UTC

Try this, take out astropulse if you don't crunch it, but it need to be listed if you do.
<app_config>

	<app>
	<name>setiathome_v8</name>
	<gpu_versions>
		<gpu_usage>0.5</gpu_usage>
		<cpu_usage>0.2</cpu_usage>
	</gpu_versions>
	</app>

	<app>
	<name>astropulse_v7</name>
	<gpu_versions>
		<gpu_usage>1</gpu_usage>
		<cpu_usage>1</cpu_usage>
	</gpu_versions>
	</app>

</app_config>
ID: 1899416 · Report as offensive
woohoo
Volunteer tester

Send message
Joined: 30 Oct 13
Posts: 972
Credit: 165,671,404
RAC: 5
United States
Message 1899427 - Posted: 5 Nov 2017, 23:36:50 UTC

<app_config>
<app_version>
<app_name>setiathome_v8</app_name>
<plan_class>opencl_ati5_SoG_nocal</plan_class>
<avg_ncpus>0.5</avg_ncpus>
<ngpus>0.5</ngpus>
</app_version>
</app_config>

of course there are eight other plan classes than the one that I just mentioned

and depending on what card and driver that you're using, you might generate errors when running multiple wu per gpu
ID: 1899427 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6534
Credit: 196,805,888
RAC: 57
United States
Message 1899471 - Posted: 6 Nov 2017, 5:47:47 UTC - in response to Message 1899397.  

Hi everyone,

I'm trying to run 2 workers on 1 AMD GPU. For some reason I always get this error:

11/5/2017 10:46:20 PM | SETI@home | Entry in app_config.xml for app 'setiathome_v8', plan class '' doesn't match any app versions

Here is the app_config.xml:

<app_config>
    <app_version>
        <app_name>setiathome_v8</app_name>
        <avg_ncpus>0.5</avg_ncpus>
        <ngpus>0.5</ngpus>
    </app_version>
</app_config>

Many guys on the webz claim this config is working for them, and looking through BOINC's wiki it looks like a valid config file to me.

Any ideas what's wrong?

That specific app_config doesn't generate an error on my system.
However without specifying a plan class these settings also get applied to the CPU apps. So there will be 2 CPU apps running on each core/thread of the CPU.
Using <app></app> settings instead of <app_version></app_version> normally make a lot more sense. Unless you have specific plan_class settings you want to adjust.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1899471 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6534
Credit: 196,805,888
RAC: 57
United States
Message 1899473 - Posted: 6 Nov 2017, 5:51:23 UTC - in response to Message 1899408.  

I thought AMD hardware always had issues running more than 1 Seti WU at a time on a video card?

it depends on the hardware and driver used. Pretty much all drivers since the release of the Rx 400 series are no good for running more than 1 task per GPU.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1899473 · Report as offensive
Dimitar Stoynev
Avatar

Send message
Joined: 7 Jan 09
Posts: 19
Credit: 336,531
RAC: 0
Bulgaria
Message 1899491 - Posted: 6 Nov 2017, 9:15:58 UTC
Last modified: 6 Nov 2017, 9:21:26 UTC

Thanks everyone!

Since this forum's multiquote capabilities are zilch, I'll try and address everyone here.

@Zalster I was running pretty much what the server threw at me, mostly SoG. I had no idea that hiding my machines also hides my work done.

@Brent Norman Thanks, BOINC still throws
Entry in app_config.xml for app 'setiathome_v8', plan class '' doesn't match any app versions
when I use your code. I deleted the Astropulse part, though. Even though there is no plan class here, it still insists on one in the error output.

@woohoo Same thing with your code... :(

@Richard Haselgrove I'm not sure if it's the wiki, but every code suggested here comes up with the same error.

It looks like something's really broken with my install/setup. Even if I somehow make it work, HAL and Grant have some disturbing news for AMD owners (sigh).

Anyway, the fight goes on! I'll make it work if only to see it crashing later!

EDIT: I'm running a 8GB RX 580 (hence my rant here https://setiathome.berkeley.edu/forum_thread.php?id=81962) @stock (17.7.2) on a trusty Ryzen 1700X (OC).
ID: 1899491 · Report as offensive
Profile Brent Norman Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester

Send message
Joined: 1 Dec 99
Posts: 2786
Credit: 685,657,289
RAC: 835
Canada
Message 1899497 - Posted: 6 Nov 2017, 10:35:23 UTC - in response to Message 1899491.  

Do you have a "hand made" app_info file that doesn't have 'setiathome_v8' in it? (I can't even see that working)
It's all I can think of since that is the same app_config that I use on my ATI R7-240 and my Nvidia cards without any issues.
ID: 1899497 · Report as offensive
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5517
Credit: 528,817,460
RAC: 242
United States
Message 1899505 - Posted: 6 Nov 2017, 11:38:28 UTC - in response to Message 1899491.  
Last modified: 6 Nov 2017, 11:38:43 UTC

I kind of figured you were running stock but wasn't sure. That is why I didn't go into the plan class as I wasn't sure what you were running.

Now that we know that, I'd suggest you use the installer from Mike's World to install the SoG app.

http://mikesworld.eu/download.html

As you go thru it make sure you check and click off the correct applications for your hardware, I know in default it will sometimes pick the cuda application.

Once you do that, you can try the following in your app_config.xml

<app_config>
<app_version>
<app_name>setiathome_v8</app_name>
<plan_class>opencl_nvidia_SoG</plan_class>
<avg_ncpus>1</avg_ncpus>
<ngpus>0.5</ngpus>
<cmdline></cmdline>
</app_version>
<app_version>
<app_name>astropulse_v7</app_name>
<plan_class>opencl_nvidia_100</plan_class>
<avg_ncpus>1</avg_ncpus>
<ngpus>0.5</ngpus>
<cmdline></cmdline>
</app_version>
</app_config>

I've removed the commandlines as you will need to get those from Mike as to what to use for the RX 580

Z
ID: 1899505 · Report as offensive
Dimitar Stoynev
Avatar

Send message
Joined: 7 Jan 09
Posts: 19
Credit: 336,531
RAC: 0
Bulgaria
Message 1899558 - Posted: 6 Nov 2017, 17:29:37 UTC - in response to Message 1899505.  

I kind of figured you were running stock but wasn't sure. That is why I didn't go into the plan class as I wasn't sure what you were running.

Now that we know that, I'd suggest you use the installer from Mike's World to install the SoG app.

http://mikesworld.eu/download.html

Z


Thanks! I've installed Lunatics' app with BOINC stopped in advance. Mike's stuff was installed correctly, at least according to logs.

Now, except for the HUGELY INCREASED crunch times I got nothing else. lolwut? All my GPU & CPU tasks got way longer to finish, compared to native things, and that's with the same GFLOPS/task as before. What perversion is this? A CPU task (Local:setiathome_v8 8.00) now takes 11 hours to complete, when it took 2h before? And that is with app_config left untouched. Mike's app_config.xml is enormous! I literally have no idea where to put the stuff now.

Even SMP is not detected/kicking in. What did I miss?
ID: 1899558 · Report as offensive
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5517
Credit: 528,817,460
RAC: 242
United States
Message 1899560 - Posted: 6 Nov 2017, 17:47:35 UTC - in response to Message 1899558.  

The remaining time to complete will be grossly exaggerated until you complete at least 11 work units.

Then the times will come down into a more reasonable time frame.

Have any of the work units completed yet with the new set up?

Z
ID: 1899560 · Report as offensive
Dimitar Stoynev
Avatar

Send message
Joined: 7 Jan 09
Posts: 19
Credit: 336,531
RAC: 0
Bulgaria
Message 1899569 - Posted: 6 Nov 2017, 18:31:14 UTC - in response to Message 1899560.  

The remaining time to complete will be grossly exaggerated until you complete at least 11 work units.

Z


Thanks for staying up with me, Z! Funny thing how much my nickname resembles yours, although I don't use it here. :-) It's just a couple of letters different. First time seeing your nick I thought I was going crazy. I was like "LOL, when did I make an account here with that nick?"

On the topic - 1.) I get different tasks than before for both CPU & GPU, although in the same GFLOPS ballpark. For GPU I'm only getting "opencl_ati5_sah", while I was getting "_SoG_nocal" and "_nocal" before. New "_ sah" takes considerably more time to complete for the same GFLOPS.
2.) CPU tasks are now just "Localhost: setiathome_v8 8.00" and they take to complete roughly as before.
3.) I haven't payed attention before, but now the GPU utilization is very shoddy. I get 20%, 70%, 91% (and again), according to HWiNFO64. I've set 85% for max CPU usage in BOINC, so the GPU tasks are not starved from CPU time.

I'll drop a line to Mike. Maybe he can shed some light here as well. Also, can I "lock" 1 CPU core for the mixed CPU+GPU tasks? I don't think my current settings or scheduler are doing a good job at it.

And the new "app_config.xml" is HUGE. I've no friggin idea where to put the GPU tags.
ID: 1899569 · Report as offensive
Profile Kissagogo27 Special Project $75 donor
Avatar

Send message
Joined: 6 Nov 99
Posts: 715
Credit: 8,032,827
RAC: 62
France
Message 1899591 - Posted: 6 Nov 2017, 20:11:32 UTC

last time i've to run 2 task on my HD7750 , it takes twice the time to crunch the 2 WU than once at a time .

and with longest results with mixed Wu Arecibo /Blc

example of app_info for 2 wu on one gpu i'd use with the <coproc> tag

<app_version>
<app_name>setiathome_v8</app_name>
<version_num>800</version_num>
<api_version>7.5.0</api_version>
<avg_ncpus>0.04</avg_ncpus>
<max_ncpus>0.2</max_ncpus>
<plan_class>opencl_ati5_nocal</plan_class>
<coproc>
<type>ATI</type>
<count>0.5</count>
</coproc>
<file_ref>
<file_name>MB8_win_x86_SSE2_OpenCL_ATi_r3330.exe</file_name>
<main_program/>
</file_ref>
<file_ref>
<file_name>libfftw3f-3-3-4_x86.dll</file_name>
</file_ref>
<file_ref>
<file_name>MultiBeam_Kernels_r3330.cl</file_name>
</file_ref>
<file_ref>
<file_name>mb_cmdline_win_x86_SSE2_OpenCL_ATi.txt</file_name>
<open_name>mb_cmdline.txt</open_name>
</file_ref>
</app_version>


with <count>1</count> only 1 be crunched .
ID: 1899591 · Report as offensive
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5517
Credit: 528,817,460
RAC: 242
United States
Message 1899592 - Posted: 6 Nov 2017, 20:13:12 UTC - in response to Message 1899569.  

Let's talk about the last one first. You don't need to add any GPU tags to the app_info.xml They should already be in there.

Using the app_config.xml allows you to override certain parts of the app_info.xml without having to go thru it(app_info) and change things there.

You can lock a cpu core for the GPU to use -cpu_lock which gets place in the commandline section but I haven't used it in a long time. Mike maybe able to help you with that part.

GPU utilization will vary by how many tasks are running. I don't know with ATI how it affect them directly since I'm a nvidia person.

CPU tasks shouldn't be any different other than name, so no worries there.

_sah was the old app before SoG came out. I'm thinking you might have selected the wrong app in the ATI section for the installer. I would rerun the installer and double check what options there are in the ATI section and see if you find a SoG application and select it. Don't worry about the work units, once the installer is finished, the work units should run on the new application even thou the name on it say _sah.

As for the name, Yes there are many Zalster out in the world, run into a few over the years but I claim it as my own, lol.......
ID: 1899592 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13720
Credit: 208,696,464
RAC: 304
Australia
Message 1899594 - Posted: 6 Nov 2017, 20:17:26 UTC - in response to Message 1899592.  

GPU utilization will vary by how many tasks are running.

Or what command line values you use.
The defaults are set to give good performance without impacting on system responsiveness. If it's a dedicated cruncher, output can be increased significantly, at the expense of system responsiveness.
Grant
Darwin NT
ID: 1899594 · Report as offensive
Dimitar Stoynev
Avatar

Send message
Joined: 7 Jan 09
Posts: 19
Credit: 336,531
RAC: 0
Bulgaria
Message 1899613 - Posted: 6 Nov 2017, 21:53:09 UTC - in response to Message 1899594.  
Last modified: 6 Nov 2017, 21:55:29 UTC

Thanks, kissagogo! Now that I have app_config.xml up and running with Mike's app, I don't really know how to configure it properly. This file has TONS of stuff in it. Thanks, Mike! :-) For some reason(s) my stock app didn't want to work with a custom app_config.

If it's a dedicated cruncher, output can be increased significantly, at the expense of system responsiveness.

I'd really like to see how, even if I have to sacrifice responsiveness. Most of the time my rig isn't doing anything anyway. My goal is to maximize the crunching output. If things start crashing I'll lift the throttle a bit, until I reach an equilibrium.

I also hope Mike get back in touch with me. I dropped him an email and hopefully he'll shed some more light on the topic.
ID: 1899613 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13720
Credit: 208,696,464
RAC: 304
Australia
Message 1899618 - Posted: 6 Nov 2017, 22:23:56 UTC - in response to Message 1899613.  

If it's a dedicated cruncher, output can be increased significantly, at the expense of system responsiveness.

I'd really like to see how, even if I have to sacrifice responsiveness. Most of the time my rig isn't doing anything anyway. My goal is to maximize the crunching output. If things start crashing I'll lift the throttle a bit, until I reach an equilibrium.

In the Project directory will be a Docs folder.
There will probably be a ReadMe_Multibeam_AMD_SoG or similar file with some suggested values to try.
Grant
Darwin NT
ID: 1899618 · Report as offensive
Dimitar Stoynev
Avatar

Send message
Joined: 7 Jan 09
Posts: 19
Credit: 336,531
RAC: 0
Bulgaria
Message 1899620 - Posted: 6 Nov 2017, 22:33:37 UTC - in response to Message 1899618.  

In the Project directory will be a Docs folder.
There will probably be a ReadMe_Multibeam_AMD_SoG or similar file with some suggested values to try.


Either I haz the dumb today and I can't brain, or the aforementioned Docs are a bit vague in terms of maximizing performance.

Thanks everyone for the insights and the help. At this point I think only Mike could help me out with this ~70KB .xml of his. Or if only AMD were to get their hands on the matter.
ID: 1899620 · Report as offensive
1 · 2 · Next

Message boards : Number crunching : 2 workers on 1 GPU


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.