2 GPU 1 doing work

Message boards : GPUs : 2 GPU 1 doing work
Message board moderation

To post messages, you must log in.

AuthorMessage
Apteryx

Send message
Joined: 15 Mar 12
Posts: 2
United States
Message 43033 - Posted: 15 Mar 2012, 1:59:03 UTC

/14/2012 20:03:04 | | OS: Microsoft Windows 7: Home Premium x64 Edition, Service Pack 1, (06.01.7601.00)
3/14/2012 20:03:04 | | Memory: 8.00 GB physical, 16.00 GB virtual
3/14/2012 20:03:04 | | Disk: 298.08 GB total, 217.92 GB free
3/14/2012 20:03:04 | | Local time is UTC -5 hours
3/14/2012 20:03:04 | | NVIDIA GPU 0: GeForce GTX 260 (driver version 29610, CUDA version 4020, compute capability 1.3, 896MB, 583 GFLOPS peak)
3/14/2012 20:03:04 | | NVIDIA GPU 1 (not used): GeForce 9800 GTX+ (driver version 29610, CUDA version 4020, compute capability 1.1, 512MB, 470 GFLOPS peak)

just installed 9800 GTX+ in this system how do I use this to run another Einstein work unit.

I have rebooted a couple times stop/closed/restarted BOINC a few times

Thanks
Appy
ID: 43033 · Report as offensive
Profile Jord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15478
Netherlands
Message 43036 - Posted: 15 Mar 2012, 9:07:54 UTC - in response to Message 43033.  

By default BOINC only uses the best GPU. To use other GPUs, you need to resort to a client configuration file and the <use_all_gpus> switch. See this FAQ for the how-to.
ID: 43036 · Report as offensive
Apteryx

Send message
Joined: 15 Mar 12
Posts: 2
United States
Message 43038 - Posted: 15 Mar 2012, 10:02:18 UTC - in response to Message 43036.  

thanks

appy
ID: 43038 · Report as offensive
Profile mitrichr
Avatar

Send message
Joined: 21 May 07
Posts: 349
United States
Message 44719 - Posted: 4 Jul 2012, 13:21:54 UTC

Jord-

Trying to isolate a problem, I tried to find the proper line for the cc_config file.

I am running with two GTX580's and the cc_config command <use_all_gpus>.

But, I want to try running only one card, so I need to know the proper command for the cc_config file.

Thanks

>>RSM
http://sciencesprings.wordpress.com
http://facebook.com/sciencesprings

ID: 44719 · Report as offensive
Profile Jord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15478
Netherlands
Message 44720 - Posted: 4 Jul 2012, 13:37:47 UTC - in response to Message 44719.  
Last modified: 4 Jul 2012, 13:38:16 UTC

If you don't care which of the two to use, remove the <use_all_gpus/> line from cc_config.xml, save changes and restart BOINC.

If you know which of the two you want to use, use the <ignore_nvidia_dev/> option. GPU count is from zero onwards. So first GPU in the system is GPU 0, second is GPU 1 and so forth.

The below will ignore (not use) Nvidia GPU#1 (second in the case).
<cc_config>
<log_flags>
</log_flags>
<options>
<ignore_nvidia_dev>1</ignore_nvidia_dev>
</options>
</cc_config>


Save changes to cc_config.xml and exit & restart BOINC.
ID: 44720 · Report as offensive
Profile mitrichr
Avatar

Send message
Joined: 21 May 07
Posts: 349
United States
Message 44723 - Posted: 4 Jul 2012, 14:00:16 UTC

Thanks-

So replace <use_all_gpus>1</use_all_gpus) with

<ignore_nvidia_dev>1</ignore_nvidia_dev> ?

I do not have <log_flags> and </log_flags> in my file, just <options>, the gpu line and </options>.

Do I need the <log_flags> lines? If so, do they both go above <options> and as two lines as you have shown them?

Thanks
http://sciencesprings.wordpress.com
http://facebook.com/sciencesprings

ID: 44723 · Report as offensive
Profile Jord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15478
Netherlands
Message 44727 - Posted: 4 Jul 2012, 15:00:53 UTC - in response to Message 44723.  

I add the <log_flags/> lines for completion of the cc_config.xml file. If you use them, they go above the <options/>section.

They aren't required for use.
ID: 44727 · Report as offensive
Profile mitrichr
Avatar

Send message
Joined: 21 May 07
Posts: 349
United States
Message 44734 - Posted: 4 Jul 2012, 16:41:50 UTC

I made the change in cc_config.xml to
<ignore_nvidia_dev>1</ignore_nvidia_dev>

BOINC Messages says on start up

"Unrecognized tag in cc_config.xml: <ignore_nvidia-dev>"

What now?

Thanks.
http://sciencesprings.wordpress.com
http://facebook.com/sciencesprings

ID: 44734 · Report as offensive
Profile mitrichr
Avatar

Send message
Joined: 21 May 07
Posts: 349
United States
Message 44735 - Posted: 4 Jul 2012, 17:04:34 UTC

I did a search and found that "nvidia" starts in 6.13.X, I am in 6.12.34, so I changed "nvidia" to "cuda", as the wiki item said, and now the line in Messages reads

"Config: ignoring GPU 1"

Now let us see what happens.

I am doing this because I started to get video kernel errors/crashes, which would last a few seconds and then recover. So, I am hoping that leaving the second card free to the computer will solve that.

I should have told you that I am using 6.12.34.

I am having just tons of problems and am trying to work through them. this is all on my fantastic Maingear Shift Super Stock machine. Those guys at Maingear, a really great bunch of very helpful people think that this machine can do anything. but, they have no knowledge of BOINC. Just as we discovered that we needed to modify CPU usage for heat as the newer processors began to run hotter (my oldest processor is now an i7-920), so too we need to understand that if we use all gpu capability for BOINC swe are leaving nothing for the rest of the computer's use.

As you can see, I have done just tons of credits since March 2012, when i had about 7,000,000, for work dating from 2007. This is just really four months to the day tomorrow since I started using this machine. My average day's credits have been now 135,000, and that includes down time. I am at 99.8 percentile in all BOINC history. I can afford to slow it down to using one GPU.
http://sciencesprings.wordpress.com
http://facebook.com/sciencesprings

ID: 44735 · Report as offensive
Profile mitrichr
Avatar

Send message
Joined: 21 May 07
Posts: 349
United States
Message 44743 - Posted: 5 Jul 2012, 0:05:10 UTC

Things are much quieter in the digiteria since I changed down to using one GPU.

The biggest problem that I see is that there is little configuration of GPU's possible in BOINC, which is far different from what is possible with CPU usage, where one can control the number of threads and the percentage of usage. I had thought that there was some configuration possible at the projects in Computing Preferences; but I checked all of my projects running on GPU and found nothing of the sort.

The only setting I can find anywhere for GPU is in Tools\Computing Preferences where there is the line "Only after computer has been idle for X minutes". I did recently change this from 3 to 1, and that might have been an error on my part. If so, you might advise me.

The thing that I do not understand if this change to one GPU is a good idea, how is it that for most of the time I have had this machine, I got away with using both GPU's with no difficulty? Did I in fact inflict my own pain by changing down from a 3 minute delay to a one minute delay on GPU usage?
http://sciencesprings.wordpress.com
http://facebook.com/sciencesprings

ID: 44743 · Report as offensive
Profile Jord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15478
Netherlands
Message 44746 - Posted: 5 Jul 2012, 1:10:11 UTC - in response to Message 44743.  

You never said what the problem was that you were debugging, and I didn't ask. If you don't want to tell in a post, why should I pry? ;-)

That there's little configuration options for the GPU isn't something to yell to the developers about. There's just no config options in any of the manufacturer's APIs. Although some projects use hacks, such as Einstein, where you can state how many tasks you want to run on any given GPU.

While you can configure the CPU to use 3 cores, you can't do so on the GPU. It will use all cores, or none. So even if you run two tasks at the same time on that GPU, it'll switch between the two tasks to run their data. It won't split half of the GPU cores to work on one task, and the other half on the other task. It'll just be all cores running part of task 1, then all cores running part of task 2, etc.
ID: 44746 · Report as offensive
Profile mitrichr
Avatar

Send message
Joined: 21 May 07
Posts: 349
United States
Message 44747 - Posted: 5 Jul 2012, 1:52:03 UTC - in response to Message 44746.  

You never said what the problem was that you were debugging, and I didn't ask. If you don't want to tell in a post, why should I pry? ;-)


I was just trying to solve this thing where the monitor would go dark for about ten seconds and then recover and Windows would tell me that there had been a problem in the display module kernel with nvidia. That has disappeared for the rest of the day since I changed to BOINC just using the one GPU.

I was wondering if that change to a 1 minute delay from a 3 minute delay might have caused that problem, because I had never seen it before while running BOINC with 2 GPU's.

Anyway, I have read through just tons of your replies to people, you are a genius and I value your guidance and friendship.

http://sciencesprings.wordpress.com
http://facebook.com/sciencesprings

ID: 44747 · Report as offensive
Profile Jord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15478
Netherlands
Message 44753 - Posted: 5 Jul 2012, 10:30:32 UTC - in response to Message 44747.  

I was just trying to solve this thing where the monitor would go dark for about ten seconds and then recover and Windows would tell me that there had been a problem in the display module kernel with nvidia.

That sounds like a problem with your drivers. Was it on the videocard that the monitor is attached to, or don't you know? Or are your cards connected in SLI mode? Which driver version?
ID: 44753 · Report as offensive
Profile mitrichr
Avatar

Send message
Joined: 21 May 07
Posts: 349
United States
Message 44754 - Posted: 5 Jul 2012, 10:52:18 UTC - in response to Message 44753.  

I am at work right now, so not at the computer with the problem. I do not know whether it is the card with the monitor. The driver is 301.42 which I believe I first saw recommended at GPUgrid.

What is SLI mode?

Anyway, last night passed without incident.


http://sciencesprings.wordpress.com
http://facebook.com/sciencesprings

ID: 44754 · Report as offensive
Profile mitrichr
Avatar

Send message
Joined: 21 May 07
Posts: 349
United States
Message 44762 - Posted: 5 Jul 2012, 21:34:54 UTC

Just got home, everything is running smoothly. Right now BOINC is using GPU 0 and ignoring GPU 1. Maingear wants the monitor to be definitely attached to GPU 0. I want to move BOINC to GPU 1.

Can I do that by changing the 1 in <ignore_cuda_dev>1</ignore_cuda_dev> to read <ignore_cuda_dev>0</ignore_cuda_dev>?

Or is there something else I need to do? I know that sometimes these numbers have specific meanings in these commands, and I do not want to assume anything.
http://sciencesprings.wordpress.com
http://facebook.com/sciencesprings

ID: 44762 · Report as offensive
Profile mitrichr
Avatar

Send message
Joined: 21 May 07
Posts: 349
United States
Message 44763 - Posted: 5 Jul 2012, 23:03:59 UTC

O.K., news from Maingear. Apparently, both cards are in fact in use because they are running with (in?, under?) SLI, Scalable Link Interface.

So, maybe the question of GPU 0 or GPU 1 is moot?

http://en.wikipedia.org/wiki/Scalable_Link_Interface
http://sciencesprings.wordpress.com
http://facebook.com/sciencesprings

ID: 44763 · Report as offensive
Profile Jord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15478
Netherlands
Message 44764 - Posted: 6 Jul 2012, 13:49:40 UTC - in response to Message 44763.  

So, maybe the question of GPU 0 or GPU 1 is moot?

No, as BOINC will detect all GPUs independently. So you can define to use them independently.

So in your case:
Actual 1st GPU in your system == GPU device 0.
Actual 2nd GPU in your system == GPU device 1.

If instead of GPU device 0 you want to use GPU device 1, then you change the line in cc_config.xml from <ignore_cuda_dev>1</ignore_cuda_dev> to <ignore_cuda_dev>0</ignore_cuda_dev> and restart BOINC. That will ignore GPU device 0 and allow you to use only GPU device 1.
ID: 44764 · Report as offensive
Profile mitrichr
Avatar

Send message
Joined: 21 May 07
Posts: 349
United States
Message 44765 - Posted: 6 Jul 2012, 15:34:56 UTC - in response to Message 44764.  

Thank you ever so much for all of your help.
http://sciencesprings.wordpress.com
http://facebook.com/sciencesprings

ID: 44765 · Report as offensive

Message boards : GPUs : 2 GPU 1 doing work

Copyright © 2024 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.