Looking through the wiki here, I do not see anything like <device_number> or <device_nums> for app_config.xml

I have two similar nvidia boards but one had 1/2 the memory the other one has. I can run 4 concurrent tasks from a certain project on the larger board but only 2 concurrent on the smaller board from that same project (they would all error out if 4 tried to run).

I do not see how to do that. The best I can think of is to exclude the project from the smaller board and use another project instead

for example: I want 4 Einstein on the 4gb board but only 2 on the 2gb board.

what I managed to do was 4 Einstein on the larger and 2 Milkyway on the smaller. I had to exclude each project from the corresponding "other" board.

The best I can think of is to exclude the project from the smaller board and use another project instead

I believe that is the best that can be done with the available options.

I want 4 Einstein on the 4gb board but only 2 on the 2gb board.

Make sure the 4GB board is #0, then use app_config to set a maximum of 6 concurrent tasks. If you're lucky #0 is filled first, resulting in a 4+2 distribution.

I want 4 Einstein on the 4gb board but only 2 on the 2gb board.

Make sure the 4GB board is #0, then use app_config to set a maximum of 6 concurrent tasks. If you're lucky #0 is filled first, resulting in a 4+2 distribution.

This worked but I could not use it as I only had 4 cores and 6 einstein was too many for this system.

So I set it to max of 4 concurrent tasks and ran into a scheduler problem with milkyway
When the last milkyway job completes it reports nvidia job que is full (or some such wording) and no more milkyway jobs are downloaded. If I temporarily suspend einstein then milkyway downloads a boatload of tasks but eventually they are all processed and the second video board is not being used.

I think this is a bug in the scheduler as the other project is not aware the first one is limited to only 4 tasks and thinks the gpu queue is full

I ended up excluding einstein from the smaller board. The 2 additional milkyway tasks consumes very little cpu so this system works ok with 4+2

I ended up excluding einstein from the smaller board.

You could also try to exclude the board for a specific application.

<exclude_gpu>
   <url>project_URL</url>
   [<device_num>N</device_num>]
   [<type>NVIDIA|ATI|intel_gpu</type>]
   [<app>appname</app>]
</exclude_gpu>

<code>app</code> specifies the short name of an application (i.e. the <code>name</code> element within the <code>app</code> element in client_state.xml). If specified, only tasks for that app are excluded.

---

You could also try using multiple clients. However, someone else would have to chime in on the best way to configure the GPU's for each client as I've not done that one yet.

---