Multiple WU's on one card only

Message boards : Number crunching : Multiple WU's on one card only
Message board moderation

To post messages, you must log in.

AuthorMessage
Ryan Munro

Send message
Joined: 5 Feb 06
Posts: 63
Credit: 18,519,866
RAC: 10
United Kingdom
Message 2011705 - Posted: 12 Sep 2019, 9:14:06 UTC

If I have two cards in one machine is this possible?
I have a Titan XP and a GT1030 in my machine, one unit almost overloads the 1030, the memory load is regularly at 100% but the Titan isn't being anywhere near maxed, 80% GPU load and minimal memory load.

Can I run 2 maybe 3 units at a time whilst only running the one on the 1030 or is it a global thing?
ID: 2011705 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13736
Credit: 208,696,464
RAC: 304
Australia
Message 2011709 - Posted: 12 Sep 2019, 9:40:35 UTC - in response to Message 2011705.  

Can I run 2 maybe 3 units at a time whilst only running the one on the 1030 or is it a global thing?
It's a global thing.
You'd be better off not processing any work on the GT 1030- just leave it for driving the display, and making use of some command line values to improve the output of your Titan XP.

In the project directory (if you installed it on the C: drive).
C:\ProgramData\BOINC\projects\setiathome.berkeley.edu

will be a file called
mb_cmdline-8.22_windows_intel__opencl_nvidia_SoG.txt

Open it with Notepad (not Word or Word Pad) put the following values in it, and save it, then the next WU that starts will use those values, and should be processed considerably quicker.
-tt 1500 -hp -period_iterations_num 1 -high_perf -high_prec_timer -cpu_lock -sbs 2048 -spike_fft_thresh 4096 -tune 1 64 1 4 -oclfft_tune_gr 256 -oclfft_tune_lr 16 -oclfft_tune_wg 256 -oclfft_tune_ls 512 -oclfft_tune_bn 64 -oclfft_tune_cw 64



If the system becomes sluggish, missing key strokes or slow display refreshes, change the -period_iterations_num 1 to something like 3 or 5. The next WU to start processing will then use the new values. If it's OK, try a lower value till it affects the system, then bump it up 1 or 2 & then stick with that value. If you use the GT 1030 to drive the display, then you won't have any system responsiveness issue no mater how aggressive the crunching settings are for the Titan.


I don't bother with limiting my CPU usage using the percentage options the others use, I just reserve 1 CPU core for each GPU WU & let the system take care of itself.

I use a file named
app_config.xml
Easiest just to open a new file using Notepad, copy the settings below, and then Save as app_config.xml, Using the BOINC Manager, just select Options, Read config files & it will use those settings.

<app_config>
 <app>
  <name>setiathome_v8</name>
  <gpu_versions>
  <gpu_usage>1.0</gpu_usage>
  <cpu_usage>1.0</cpu_usage>
  </gpu_versions>
 </app>
 <app>
  <name>astropulse_v7</name>
  <gpu_versions>
  <gpu_usage>1.0</gpu_usage>
  <cpu_usage>1.0</cpu_usage>
  </gpu_versions>
 </app>
</app_config>



Then see how the system runs for a while, keeping an eye on the type of WU being processed. Once you've got a baseline, try running 2 WUs at a time and see if the output improves or falls. I've found no benefit with my cards, but others with similar hardware have. All you can do is try it for yourself and see.
Grant
Darwin NT
ID: 2011709 · Report as offensive
Ryan Munro

Send message
Joined: 5 Feb 06
Posts: 63
Credit: 18,519,866
RAC: 10
United Kingdom
Message 2011713 - Posted: 12 Sep 2019, 10:10:17 UTC - in response to Message 2011709.  

Cheers this is my gaming rig so I cant attach all my screens to the 1030, I do use it to power my second and third screen though, is it possible to tell Boinc / Seti to totally ignore that card at all?
Failing that I guess ill remove it and grab some adapters to hook the other screens up to the Titan.
ID: 2011713 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13736
Credit: 208,696,464
RAC: 304
Australia
Message 2011714 - Posted: 12 Sep 2019, 10:23:13 UTC - in response to Message 2011713.  
Last modified: 12 Sep 2019, 10:23:55 UTC

is it possible to tell Boinc / Seti to totally ignore that card at all?
With cards of such different capabilities, you usually have to use a configuration option in order to make use of the lower powered card- the default is to use only the most capable. It is possible to exclude particular GPUs, but someone more knowledgeable will have to help you with that option.
Grant
Darwin NT
ID: 2011714 · 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 2011746 - Posted: 12 Sep 2019, 16:31:02 UTC - in response to Message 2011714.  
Last modified: 12 Sep 2019, 16:31:31 UTC

The first question is, are you using an cc_config on that machine? That is usually the only way to get boinc to use both cards. If you are, then remove that so that only the Titan is being used. That should prevent BOINC from using the lower end card for work.

If that fails then the next step would be to modify the cc_config with the ignore option. But for that you would need to figure out which slot the low end card is assigned to. I believe Keith just recently found a way to modify the way BOINC list the cards so that it matches how the computer list the cards. Why is that important? It makes it easier for you to know which card is in which slot to place that value in the cc_config for boinc to ignore. Sometimes card 0 isn't listed in slot 0 so it can be confusing when sometimes.

Try removing the cc_config first, reboot and see if that works.
ID: 2011746 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14650
Credit: 200,643,578
RAC: 874
United Kingdom
Message 2011749 - Posted: 12 Sep 2019, 18:01:40 UTC - in response to Message 2011746.  

Before you do that, just read the startup messages in the BOINC Event Log. It will tell you which what devices have been detected, what device number they have been assigned (you will need this, not the physical slot number), and whether or not BOINC will use them in the current configuration. Start from there.
ID: 2011749 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 2011750 - Posted: 12 Sep 2019, 18:03:38 UTC - in response to Message 2011746.  

You don't have to care about how the cards are numbered by the host system. All you have to care about is how BOINC identifies the cards. The Event Log startup lists the cards detected by the client starting with GPU0 and then GPU1. Since the Titan XP is the most powerful card, it will be assigned GPU0 identifier and the GT1030 will be assigned the GPU1 identifier. That is all you have to know. Put in an exclude statement in cc_config.xml in the Options section for GPU1 and you should be good to go.

  <exclude_gpu>
        <url>http://setiathome.berkeley.edu/</url>
        <device_num>1</device_num>
        <type>NVIDIA</type>
    </exclude_gpu

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2011750 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13736
Credit: 208,696,464
RAC: 304
Australia
Message 2011818 - Posted: 13 Sep 2019, 7:05:59 UTC
Last modified: 13 Sep 2019, 7:06:40 UTC

I just had a quick look through your task list, and I can't see any sign of the GT 1030 processing any WUs at all, so no need to worry about use_all_gpus or exclude_gpu options.
Just add the command line vales to the appropriate file & let your Titan really get stuck in to some work.
Grant
Darwin NT
ID: 2011818 · Report as offensive
Ryan Munro

Send message
Joined: 5 Feb 06
Posts: 63
Credit: 18,519,866
RAC: 10
United Kingdom
Message 2011831 - Posted: 13 Sep 2019, 13:07:02 UTC - in response to Message 2011709.  



Open it with Notepad (not Word or Word Pad) put the following values in it, and save it, then the next WU that starts will use those values, and should be processed considerably quicker.
-tt 1500 -hp -period_iterations_num 1 -high_perf -high_prec_timer -cpu_lock -sbs 2048 -spike_fft_thresh 4096 -tune 1 64 1 4 -oclfft_tune_gr 256 -oclfft_tune_lr 16 -oclfft_tune_wg 256 -oclfft_tune_ls 512 -oclfft_tune_bn 64 -oclfft_tune_cw 64




Wow thats almost halved my times without any rise in fan speeds / temps, I was getting 5 min times, now its 2:40 odd, is there a post anywhere explaining what all the commands actually do?

Its even shaved 10 mins of the GT1030 times
ID: 2011831 · Report as offensive
Ryan Munro

Send message
Joined: 5 Feb 06
Posts: 63
Credit: 18,519,866
RAC: 10
United Kingdom
Message 2011832 - Posted: 13 Sep 2019, 13:12:21 UTC - in response to Message 2011831.  

Also, are these settings Nvidia specific or would they benefit at all on AMD cards?
ID: 2011832 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 2011847 - Posted: 13 Sep 2019, 15:23:01 UTC - in response to Message 2011832.  

Also, are these settings Nvidia specific or would they benefit at all on AMD cards?

If you look in the project /Docs directory, there is a readme explaining the parameters. And yes, the same parameters are used on AMD cards. The same application is used for all gpus. Just named differently for the different brands. There is a dedicated thread for AMD tuning parameters.
https://setiathome.berkeley.edu/forum_thread.php?id=84209
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2011847 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13736
Credit: 208,696,464
RAC: 304
Australia
Message 2011873 - Posted: 13 Sep 2019, 21:57:58 UTC - in response to Message 2011831.  

Its even shaved 10 mins of the GT1030 times
I went through several pages of results before, and couldn't see any for the GT 1030, and this time, there they are!
Grant
Darwin NT
ID: 2011873 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13736
Credit: 208,696,464
RAC: 304
Australia
Message 2011915 - Posted: 14 Sep 2019, 4:36:06 UTC

I noticed 2 WUs that have errored out, which is due to the video driver you are using.
Anything later than v431.60 results in some types of WUs exceeding their estimated processing time.
Grant
Darwin NT
ID: 2011915 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 2011921 - Posted: 14 Sep 2019, 5:37:11 UTC - in response to Message 2011915.  

Interesting. So another species of Arecibo VHAR tasks don't run with the 436 drivers. See that these are AR=4.6. Wonder where the cutoff is. I wonder what it is the lowest AR that runs correctly on the 436 drivers. Anybody see any VHAR tasks lower than the 2.7 AR that we first noticed failing. Any VHAR's in the 1.x range that fail? Any VHAR's in the 1.x range that crunch correctly?
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2011921 · Report as offensive

Message boards : Number crunching : Multiple WU's on one card only


 
©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.