Only 1 GPU is used for OpenCL though cc_config.xml is set properly. ver. 7.11, 7.9, even 7.8 on Fedora 28!

Message boards : BOINC client : Only 1 GPU is used for OpenCL though cc_config.xml is set properly. ver. 7.11, 7.9, even 7.8 on Fedora 28!
Message board moderation

To post messages, you must log in.

AuthorMessage
Tetsuji Maverick Rai

Send message
Joined: 29 Aug 05
Posts: 14
Japan
Message 86197 - Posted: 13 May 2018, 5:11:17 UTC
Last modified: 13 May 2018, 6:06:58 UTC

Hi all,

I am encountering a strange phenomenon. I installed two NVIDIA cards in my rig running Fedora 28 and run boinc with cc_config.xml
<cc_config>
    <options>
        <use_all_gpus>1</use_all_gpus>
    </options>
</cc_config>

note use_all_gpus is set to 1.

Anyway "coproc debug info"-enabled log also says
Sun 13 May 2018 01:54:21 PM JST | collatz | [coproc] NVIDIA instance 0; 1.000000 pending for collatz_sieve_ab75f0e3-7368-4642-a127-1626d5a1bf3e_0
Sun 13 May 2018 01:54:21 PM JST | collatz | [coproc] NVIDIA instance 0: confirming 1.000000 instance for collatz_sieve_ab75f0e3-7368-4642-a127-1626d5a1bf3e_0
Sun 13 May 2018 01:54:21 PM JST | collatz | [coproc] Assigning NVIDIA instance 1 to collatz_sieve_272c3169-bcc5-4fbc-a45d-65bee7a17426_0


But as you see, NVIDIA instance 1 is not confirmed and actually it does not appear in the task list. The task is left in "Ready to Start" status, waiting to be crunched.

In the begging of the log, boinc client detects all gpus.
Sun 13 May 2018 02:06:49 PM JST |  | CUDA: NVIDIA GPU 0: GeForce GTX 1070 (driver version 396.24, CUDA version 9.2, compute capability 6.1, 4096MB, 3986MB available, 6463 GFLOPS peak)
Sun 13 May 2018 02:06:49 PM JST |  | CUDA: NVIDIA GPU 1: GeForce GTX 750 Ti (driver version 396.24, CUDA version 9.2, compute capability 5.0, 2002MB, 1964MB available, 1421 GFLOPS peak)
Sun 13 May 2018 02:06:49 PM JST |  | OpenCL: NVIDIA GPU 0: GeForce GTX 1070 (driver version 396.24, device version OpenCL 1.2 CUDA, 8120MB, 3986MB available, 6463 GFLOPS peak)
Sun 13 May 2018 02:06:49 PM JST |  | OpenCL: NVIDIA GPU 1: GeForce GTX 750 Ti (driver version 396.24, device version OpenCL 1.2 CUDA, 2002MB, 1964MB available, 1421 GFLOPS peak)
Sun 13 May 2018 02:06:49 PM JST |  | OpenCL: Intel GPU 0: Intel(R) HD Graphics Haswell GT2 Desktop (driver version 1.3, device version OpenCL 1.2 beignet 1.3, 2048MB, 2048MB available, 160 GFLOPS peak)
Sun 13 May 2018 02:06:49 PM JST |  | OpenCL CPU: pthread-Intel(R) Core(TM) i7-4790K CPU @ 4.00GHz (OpenCL driver vendor: The pocl project, driver version 1.1, device version OpenCL 1.2 pocl HSTR: pthread-x86_64-unknown-linux-gnu-haswell)



Several versions with other projects (SETI@home, Amicable Numbers also) have this problem, so my setting must be wrong.

What is wrong?

Thanks in advance!

EDIT: For testing I added SETI@home/beta for Intel gpu and the task is "Waiting to run". instead of "Ready to run" or "Running." Why?

BTW I am running X Windows on Intel GPU.

EDIT2: Another problem which existed before this is that boinc client shuts down when transitioning to graphic mode or returning back to text mode. So I need to restart boinc service manually.
-Tetsuji Maverick Rai
Salt of the earth. Luckiest in the world.
ID: 86197 · Report as offensive
Tetsuji Maverick Rai

Send message
Joined: 29 Aug 05
Posts: 14
Japan
Message 86199 - Posted: 13 May 2018, 9:42:00 UTC - in response to Message 86197.  

Self reply.

I install Fedora 27 and all the problems are gone! Fedora 28 seems to have a problem on this.
-Tetsuji Maverick Rai
Salt of the earth. Luckiest in the world.
ID: 86199 · Report as offensive
Tetsuji Maverick Rai

Send message
Joined: 29 Aug 05
Posts: 14
Japan
Message 86205 - Posted: 14 May 2018, 0:58:55 UTC
Last modified: 14 May 2018, 1:35:34 UTC

Again, it happened with Fedora 27!! After running 9 or 10 hours, it doesn't recognize more than 1 GPU. Just one hour ago, it recognized 2 GPUs, but now it doesn't. Reinstalling new version (7.11.0) or old version (7.9.3) doesn't help. I guess this isn't a Fedora's bug. It's due to boinc client.

How can I fix it?

EDIT: I cleared all files in /var/lib/boinc, and setup again, and now boinc-client began to use 2 GPUs. But I botched cached workunits.
-Tetsuji Maverick Rai
Salt of the earth. Luckiest in the world.
ID: 86205 · Report as offensive
Tetsuji Maverick Rai

Send message
Joined: 29 Aug 05
Posts: 14
Japan
Message 86206 - Posted: 14 May 2018, 2:55:23 UTC - in response to Message 86205.  

This happens with Collaz Conjecture project. IIRC, it didn't happen with Amicable Numbers for months. So Collaz might be wrong.
-Tetsuji Maverick Rai
Salt of the earth. Luckiest in the world.
ID: 86206 · Report as offensive

Message boards : BOINC client : Only 1 GPU is used for OpenCL though cc_config.xml is set properly. ver. 7.11, 7.9, even 7.8 on Fedora 28!

Copyright © 2019 University of California. Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.