Two nvidia cards and only one is doing math?!

[BulldogPO]

I have edited cc_config.xml in C:/programfiles/boinc/ folder to:

<cc_config>
<options>
<use_all_gpus>1</use_all_gpus>
</options>
</cc_config>

And I?m using Boinc 7.0.42 with app_config.xml in C:\ProgramData\BOINC\projects\www.worldcommunitygrid.org folder just like this:

<app_config>
   <app>
      <name>hcc1</name>
      <max_concurrent>4</max_concurrent>
      <gpu_versions>
          <gpu_usage>.250</gpu_usage>
          <cpu_usage>1.000</cpu_usage>
      </gpu_versions>
   </app>
</app_config>

And still going like this:


So where is the problem?

[nanoprobe]

I know there are some projects that have a problem with running 2 Nvidia cards at the same time but I'm not sure if WCG is one of them. Try adding the line of code below to your cc_config.xml file which is located in the BOINC folder not the project folder.

<use_all_gpus>1</use_all_gpus>

If you don't have a cc_config file in the BOINC folder copy and paste the code below into notepad, rename it cc_config.xml and place it into the BOINC folder. You'll have to shut down and restart boinc for the changes to take effect.

<cc_config>
<options>
<use_all_gpus>1</use_all_gpus>
</options>
</cc_config>

[BulldogPO]

I had that cc_config in boinc folder already, like I wrote in first post.

[bluestang]

Have you tried max_concurrent set to 8 ?

[nanoprobe]

I had that cc_config in boinc folder already, like I wrote in first post.

My bad. Was in a hurry and missed that. If that doesn't get both cards going I have no other suggestions.

[BulldogPO]

Changed Max Concurrent to 8 and one packet went to second card, but only one.

[bluestang]

You have a 3570k so cpu_usage should be .50 to run 4 per card.

You can probably put concurrent back to 4 then.

[BulldogPO]

Damn, is it this CPU, i5-3570K..
Now it does three packets per GPU as I changed app_config to following:

<app_config>
   <app>
      <name>hcc1</name>
      <max_concurrent>6</max_concurrent>
      <gpu_versions>
          <gpu_usage>.333</gpu_usage>
          <cpu_usage>.667</cpu_usage>
      </gpu_versions>
   </app>
</app_config>

[BulldogPO]

Now looks good, and does four threads per GPU (four on GTX 660 Ti and four on GTX 670)



With this:

<app_config>
   <app>
      <name>hcc1</name>
      <max_concurrent>8</max_concurrent>
      <gpu_versions>
          <gpu_usage>.250</gpu_usage>
          <cpu_usage>0.500</cpu_usage>
      </gpu_versions>
   </app>
</app_config>

CPU might be bottlenecking if I add one card more?

[Bun-Bun]

Which card? My GTX670 didn't benefit from any more than 2 threads. I suspect you would be better off running 2 threads per card.

[nanoprobe]

Which card? My GTX670 didn't benefit from any more than 2 threads. I suspect you would be better off running 2 threads per card.

Agreed. Nvidia has horrible OpenCl driver support. If you do some testing I think you'll find that 2 is the sweet spot for them.

[STEvil]

I'm surprised you can stand running it on your nVidia cards at all. Everything turns into a lag fest on the desktop when I use my GTX670's..

[Snow Crash]

In my testing I found the best results for NVidia is to only run 1 WU per CPU thread as their implementation of OpenCL is really a pig.

[nanoprobe]

I'm surprised you can stand running it on your nVidia cards at all. Everything turns into a lag fest on the desktop when I use my GTX670's..

You can get around that if you use 1 to crunch and the other to run the just the monitor. Not the best scenario but better than the lag fest. If you have room for a third card that is older and can't run the tasks maybe you could use that for the monitor while both the 670s crunch.

[BulldogPO]

nanoprobe, how do you exclude one specific card from WCG?

[nanoprobe]

nanoprobe, how do you exclude one specific card from WCG?

Add an exclude GPU tag to you cc_config.xml file in the ProgramData BOINC folder.

<exclude_gpu>
<url>project_www.worldcommunitygrid.org</url>
<device_num>N</device_num> N= device number either 0 or 1
<type>nvidia|ati</type> One or the other, not both
<app>appname</app> GPU app name
</exclude_gpu>