Collatz does not use all available GPUs
log in

Advanced search

Message boards : Unix/Linux : Collatz does not use all available GPUs

Previous · 1 · 2 · 3 · Next
Author Message
Zdenek Hrbek
Send message
Joined: 28 Mar 10
Posts: 4
Credit: 1,012,504,481
RAC: 846,808
Message 22489 - Posted: 30 May 2016, 12:20:06 UTC - in response to Message 22480.

Dear Veit

Thanks for Your time. Tried to modify this right now, but nothing changes.
Project Einstein uses both GPUs, Moo Wrapper also both GPUs, but when
I change to collatz it runs only on one GPU. And really, one of two GPUs is unused and cold.
____________

vdvogt
Send message
Joined: 10 Jan 16
Posts: 38
Credit: 1,090,698,551
RAC: 0
Message 22497 - Posted: 31 May 2016, 17:49:04 UTC

Hi Zdenek Hrbek,
I agree with you.
When I let run GPUGRID it uses both GPUs.
But Collatz ONE GPU ONLY.

I do not know how to change that and it seams that no one of the developers do take care of this problem ;-(((((

regards
Veit

Helix Von Smelix
Send message
Joined: 2 Aug 10
Posts: 43
Credit: 10,188,485,336
RAC: 1,477,398
Message 22592 - Posted: 19 Jun 2016, 18:07:34 UTC

any solution to this? I suddenly have this issue.

vdvogt
Send message
Joined: 10 Jan 16
Posts: 38
Credit: 1,090,698,551
RAC: 0
Message 22595 - Posted: 20 Jun 2016, 14:02:13 UTC

Hi,
in the last time I had made some tests:

1. shutdown the computer, let them cool down (especially the grafics card) and reactivate it.
Than Collatz first uses both GPUs, but only a few task were calculated by GPU1 and afterwards GPU0 was active only.

2. pause GPU calculation for some minutes.
Sometimes Collatz started with both GPUs, sometimes only one.

I got these message from Boinc multiple times:

Mo 20 Jun 2016 13:52:46 UTC | Collatz Conjecture | [coproc] Assigning NVIDIA instance 1 to collatz_sieve_2726103852208506273792_6597069766656_0

But this task is never going to be calculatet by GPU1

I think this means, that Collatz tries to calculate with GPU1, but something prevents it.

What could that be?????

Any suggestions?

regards
Veit

Zdenek Hrbek
Send message
Joined: 28 Mar 10
Posts: 4
Credit: 1,012,504,481
RAC: 846,808
Message 22609 - Posted: 23 Jun 2016, 7:21:16 UTC - in response to Message 22595.

Last week I checked situation, and suppose the problém was in the application (tasks), because no chages from my side were made, and now both GPUs are running together. So it seems problém is solved.
Best regards
____________

vdvogt
Send message
Joined: 10 Jan 16
Posts: 38
Credit: 1,090,698,551
RAC: 0
Message 22610 - Posted: 23 Jun 2016, 9:19:05 UTC - in response to Message 22609.

Hi Zdenek Hrbek,
NO, the problem is NOT solved.

In the meantime I heard that other people with DUAL-GPU grafig card have the same problem.

One wrote me to write

sleep=2

to the file

collatz_sieve_1.21_x86_64-pc-linux-gnu__opencl_nvidia_gpu.config

in

/var/lib/boinc-client/projects/boinc.thesonntags.com_collatz

I tried that, but had no success!

Conclusion: The problem is still there!

regards
Veit

mmonnin
Send message
Joined: 29 May 16
Posts: 8
Credit: 166,876,282
RAC: 891
Message 22621 - Posted: 24 Jun 2016, 12:13:50 UTC
Last modified: 24 Jun 2016, 12:14:23 UTC

Have you tried using this in app_config.xml ?

<gpu_usage>.5</gpu_usage>



gpu_usage
The number of GPU instances (possibly fractional) used by GPU versions of this app. For example, .5 means that two jobs of this application can run at once on a single GPU.

https://boinc.berkeley.edu/wiki/Client_configuration

vdvogt
Send message
Joined: 10 Jan 16
Posts: 38
Credit: 1,090,698,551
RAC: 0
Message 22625 - Posted: 25 Jun 2016, 9:30:47 UTC - in response to Message 22621.

Hi mmonnin,
where is app_config.xml?
I can not find it.

regards
Veit

vdvogt
Send message
Joined: 10 Jan 16
Posts: 38
Credit: 1,090,698,551
RAC: 0
Message 22626 - Posted: 25 Jun 2016, 9:54:43 UTC - in response to Message 22625.

Hi,
while searching for app_config.xml I found in
/var/lib/boinc-client/coproc_info.xml
this message at the and:

<warning>NVIDIA library reports 2 GPUs</warning>

Why is there a warning?

Could this be one of the reasons that Collatz does not use both GPUs?

regards
Veit

mmonnin
Send message
Joined: 29 May 16
Posts: 8
Credit: 166,876,282
RAC: 891
Message 22634 - Posted: 27 Jun 2016, 13:16:35 UTC

It is in that projects directory. It may be located in different locations based on how BOINC was installed.

Is it here?
/var/lib/boinc-client/projects/boinc.thesonntags.com_collatz/

You have 2 GPUs on one card. BOINC is running 1 task per card. Make it use 2 tasks on a card.

One of the 1st links when googling came to my own teams optimization thread. See this post which says to do the same thing I suggested.
http://www.overclock.net/t/1414835/collatz-conjecture-project-support-thread/30_30#post_21783131

vdvogt
Send message
Joined: 10 Jan 16
Posts: 38
Credit: 1,090,698,551
RAC: 0
Message 22638 - Posted: 28 Jun 2016, 8:35:40 UTC - in response to Message 22634.

Hi mmonnin,
I had written that app_config.xml to the directory
/var/lib/boinc-client/projects/boinc.thesonntags.com_collatz/
It was not there before.

But there must be more than this single line

<gpu_usage>.5</gpu_usage>

I think ther should be beginning and ending lines.

Please send me an example.

regards
Veit

vdvogt
Send message
Joined: 10 Jan 16
Posts: 38
Credit: 1,090,698,551
RAC: 0
Message 22639 - Posted: 28 Jun 2016, 9:22:02 UTC - in response to Message 22638.

Hi mmonnin,
I just added these lines to app_config.xml so that it is now:

<app_config>
<gpu_usage>.5</gpu_usage>
</app_config>

But I think there must be more because I got this error message:


Di 28 Jun 2016 09:18:56 UTC | Collatz Conjecture | Unknown tag in app_config.xml: gpu_usage


regards
Veit

vdvogt
Send message
Joined: 10 Jan 16
Posts: 38
Credit: 1,090,698,551
RAC: 0
Message 22640 - Posted: 28 Jun 2016, 9:38:41 UTC - in response to Message 22639.

Hi mmonnin,
its me once again;-)

I just found that:

<app_config>
<app>
<name>solo_collatz</name>
<max_concurrent>1</max_concurrent>
<gpu_versions>
<gpu_usage>1.0</gpu_usage>
<cpu_usage>0.2</cpu_usage>
</gpu_versions>
</app>
</app_config>

But I do not understand the following line:

<name>solo_collatz</name> ?????

What is the meaning of that?

And:
Why is there the cpu line in the <gpu_versions>

<gpu_versions>
<gpu_usage>1.0</gpu_usage>
<cpu_usage>0.2</cpu_usage> ?????
</gpu_versions>

What are the meaning of the two values?

<gpu_usage>1.0</gpu_usage>
does it mean: run on one GPU only?
Than I should set it to 2.0?

<cpu_usage>0.2</cpu_usage>
Does it mean: use 20% of the CPU?
I would like to run 100% of the CPUs. Or is it counterproductive because every GPU needs an amount of a CPU as assistance.


regards
Veit

Profile mikey
Avatar
Send message
Joined: 11 Aug 09
Posts: 3242
Credit: 1,686,599,333
RAC: 6,251,296
Message 22644 - Posted: 28 Jun 2016, 10:51:28 UTC - in response to Message 22640.

Hi mmonnin,
its me once again;-)

But I do not understand the following line:

<name>solo_collatz</name> ?????

What is the meaning of that?


That's the name of one of the kinds of files Collatz sends out.

And:
Why is there the cpu line in the <gpu_versions>

<gpu_versions>
<gpu_usage>1.0</gpu_usage>
<cpu_usage>0.2</cpu_usage> ?????
</gpu_versions>

What are the meaning of the two values?

<gpu_usage>1.0</gpu_usage>
does it mean: run on one GPU only?


No it means to run one workunit on each gpu, putting a 0.5 for example would mean to run 2 workunits per gpu, a 0.33 would mean to run 3 units per gpu and a 0.25 would mean to run 4 workunits at the same time on each gpu.


Than I should set it to 2.0?


NO, that would mean running 1/2 of a workunit on each gpu which is physically impossible.

<cpu_usage>0.2</cpu_usage>
Does it mean: use 20% of the CPU?
I would like to run 100% of the CPUs. Or is it counterproductive because every GPU needs an amount of a CPU as assistance.


Yes it means to only use upto 20% of the cpu to keep the gpu fed, BUT even if you did leave this set of lines out the cpu will NEVER reach 100% because the gpu is doing the crunching, the cpu is just sending and receiving the data to and from the gpu.

regards
Veit


NONE of this has ANYTHING to do with why only only gpu is being used in your system, it ALL has to do with tweaking the gpu's that are being used. I have one gpu crunching here, and AMD 6870, and a single workunit is using @96% of the total gpu resources. So there is no point in trying to run multiple workunits at once on it. As for tweaking the cpu usage I always leave a cpu core free for each gpu to use as it sees fit in my machines, so no need for that kind of tweaking either.

vdvogt
Send message
Joined: 10 Jan 16
Posts: 38
Credit: 1,090,698,551
RAC: 0
Message 22645 - Posted: 28 Jun 2016, 10:57:19 UTC - in response to Message 22640.

Hi mmonnin,
hi all,
I had read the complete postings on
http://www.overclock.net/t/1414835/collatz-conjecture-project-support-thread/30_30#post_21783131

and i'm confused completely now. ;-((

I do not understand these lines in the file
/var/lib/boinc-client/projects/boinc.thesonntags.com_collatz/collatz_sieve_1.21_x86_64-pc-linux-gnu__opencl_nvidia_gpu.config


verbose=1 that's clear!
kernels_per_reduction=128 what is a reduction? what is reduced?
threads=8 threads per?
lut_size=17 what is the meaning of lut?
sieve_size=30 what is the effect of varying sieve_size?
sleep=1 sleep after what? what is the dimension of 1: seconds?
reduce_cpu=0 why should I reduce CPU?

How does this all effect calculating on both GPUs on my Titan Z???

I need a basic instruction of these values.

regards
Veit

mmonnin
Send message
Joined: 29 May 16
Posts: 8
Credit: 166,876,282
RAC: 891
Message 22664 - Posted: 1 Jul 2016, 23:55:22 UTC

To me it seems like collatz is seeing your card as a 1 GPU but you have 2 actual chips on the card so you have to tell the app to run 2 tasks on your card.

Use this in app_config.xml

​<app_config>
<app>
<name>solo_collatz</name>
<max_concurrent>2</max_concurrent>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
</gpu_versions>
</app>
</app_config>


The means are explained in the Boinc Wiki
https://boinc.berkeley.edu/wiki/Client_configuration

The rest of the OCN thread was optimizing Collatz per task. Post 58 contains the above app_config.xml code to get 2 tasks to run per GPU.

A full explanation of the other options like items_per_kernal and lut

vdvogt
Send message
Joined: 10 Jan 16
Posts: 38
Credit: 1,090,698,551
RAC: 0
Message 22665 - Posted: 2 Jul 2016, 10:00:06 UTC - in response to Message 22664.

Hi mmonnin,
I had changed the app_config.xml as you recommended.
But I got this error message:

Sa 02 Jul 2016 09:52:09 UTC | Collatz Conjecture | Your app_config.xml file refers to an unknown application 'solo_collatz'. Known applications: 'collatz_sieve'

Solo_collatz is wrong.

When I enter the command

systemctl status boinc-client

I get this output:

● boinc-client.service - Berkeley Open Infrastructure Network Computing Client
Loaded: loaded (/lib/systemd/system/boinc-client.service; enabled; vendor preset: enabled)
Active: active (running) since Sa 2016-07-02 09:52:08 UTC; 1min 58s ago
Process: 15600 ExecStopPost=/bin/rm -f /var/lib/boinc-client/lockfile (code=exited, status=0/SUCCESS)
Process: 15633 ExecStartPre=/bin/chown boinc:boinc /var/log/boinc.log /var/log/boincerr.log (code=exited, status=0/SUCCESS)
Process: 15631 ExecStartPre=/usr/bin/touch /var/log/boinc.log /var/log/boincerr.log (code=exited, status=0/SUCCESS)
Main PID: 15637 (sh)
CGroup: /system.slice/boinc-client.service
├─15637 /bin/sh -c /usr/bin/boinc --dir /var/lib/boinc-client >/var/log/boinc.log 2>/var/log/boincerr.log
├─15638 /usr/bin/boinc --dir /var/lib/boinc-client
├─16044 ../../projects/boinc.thesonntags.com_collatz/collatz_sieve_1.21_x86_64-pc-linux-gnu
├─16045 ../../projects/boinc.thesonntags.com_collatz/collatz_sieve_1.21_x86_64-pc-linux-gnu
├─16046 ../../projects/boinc.thesonntags.com_collatz/collatz_sieve_1.21_x86_64-pc-linux-gnu
├─16048 ../../projects/boinc.thesonntags.com_collatz/collatz_sieve_1.21_x86_64-pc-linux-gnu
├─16050 ../../projects/boinc.thesonntags.com_collatz/collatz_sieve_1.21_x86_64-pc-linux-gnu
├─16052 ../../projects/boinc.thesonntags.com_collatz/collatz_sieve_1.21_x86_64-pc-linux-gnu
├─16055 ../../projects/boinc.thesonntags.com_collatz/collatz_sieve_1.21_x86_64-pc-linux-gnu
├─16057 ../../projects/boinc.thesonntags.com_collatz/collatz_sieve_1.21_x86_64-pc-linux-gnu
└─16059 ../../projects/boinc.thesonntags.com_collatz/collatz_sieve_1.21_x86_64-pc-linux-gnu__opencl_nvidia_gpu

Jul 02 09:52:08 Ci7-293 systemd[1]: Starting Berkeley Open Infrastructure Network Computing Client...
Jul 02 09:52:08 Ci7-293 systemd[1]: Started Berkeley Open Infrastructure Network Computing Client.

Probably I have to enter collatz_sieve instead?

I will test that.

regards
Veit

vdvogt
Send message
Joined: 10 Jan 16
Posts: 38
Credit: 1,090,698,551
RAC: 0
Message 22666 - Posted: 2 Jul 2016, 10:07:29 UTC - in response to Message 22665.

Hi mmonnin,
collatz sieve is right.
But with the actual settings in app_config.xml

TWO tasks are running on GPU0, NONE on the CPUs and GPU1 is still IDLE.

I'll restore the old version of app_config.xml to run at least 8 CPU tasks and ONE GPU task until I got the right settings.

regards
Veit

vdvogt
Send message
Joined: 10 Jan 16
Posts: 38
Credit: 1,090,698,551
RAC: 0
Message 22668 - Posted: 2 Jul 2016, 11:41:37 UTC - in response to Message 22666.

HELP! HELP! HELP!

Hi mmonnin,
since I resetted app_config.xml (I renamed it so that boinc can not see it any more after a restart.)

Only the CPUs are calculating and the two GPU tasks of GPU0 are suspended.

Could it be that boinc keeps the app_config.xml anywhere else?

regards
Veit

mmonnin
Send message
Joined: 29 May 16
Posts: 8
Credit: 166,876,282
RAC: 891
Message 22678 - Posted: 4 Jul 2016, 19:01:55 UTC

What happens if you put this into app_config.xml above or below the GPU usage line?

<cpu_usage>1</cpu_usage> ?????


I use my CPUs for other CPU only projects.

I don't know of another place it's stored. Has Boinc manager been restarted or the project updated?

Previous · 1 · 2 · 3 · Next
Post to thread

Message boards : Unix/Linux : Collatz does not use all available GPUs


Main page · Your account · Message boards


Copyright © 2018 Jon Sonntag; All rights reserved.