Guide to running multiple GPU work units

[t_ski]

Like mstenholm, using m&m's utility got me up and running. Currently running 8 WU on my 7970 and 8 WU on my 2600k.

http://img.techpowerup.org/130103/8wu.jpg

That looks like it's only running 8 GPU WU's. I'm assuming it's set up to switch over to CPU WU's if the GPU WU's run out.

 

[KieX]

Hello everybody, I did a utility to set all the settings required faster.
The utility works with Vista/7/8 and with the new app_config.xml (BOINC version 7.0.40 or above).

Basically, it creates the app_config.xml with all your settings.

It should (I hope) save you time.

Happy new year!

Awesome utility, good to see it already resolved a few problems. Updated the OP to include your post and util as the preferred method

Like mstenholm, using m&m's utility got me up and running. Currently running 8 WU on my 7970 and 8 WU on my 2600k.

http://img.techpowerup.org/130103/8wu.jpg

As t_ski said, you're setup to run the GPU and CPU together, so it will only do CPU work when GPU WU run out. You might want to lower the number of GPU WU until you've completed the CPU tasks otherwise they will not see any CPU time until they become high priority tasks.

Best results will come from GPU only (and setting the WCG profile to match). If you want to use CPU and GPU you'll need to reduce the WU for GPU to avoid "returned late" results from CPU.

 

[Norton]

As t_ski said, you're setup to run the GPU and CPU together, so it will only do CPU work when GPU WU run out. You might want to lower the number of GPU WU until you've completed the CPU tasks otherwise they will not see any CPU time until they become high priority tasks.

Best results will come from GPU only (and setting the WCG profile to match). If you want to use CPU and GPU you'll need to reduce the WU for GPU to avoid "returned late" results from CPU.

Quick note- the app_config setup is much better at reducing the # of "late" results on the CPU.
Example- I've been running 8 GPU and 2 CPU (HCC only atm) on my FX-8350/7870 rig for a while now and the batch of CPU HCC work units ready to start now stay at a reasonable level.... much better the older app_info setup

*I don't think I've had a late result on the rig since I switched over.

 

[brandonwh64]

When using two nvidia cards, does it matter if they are in SLI or not?

 

[mstenholm]

You might need to hit Update since a BOINC sever was down for some hours and your buffer could have run dry. I did run dry on one rig and if I hadn't asked for new jobs I would have lost yet another two hours on top of the 1½ hours I lost already. I had +1300 wus waiting to be sent from tree rigs.

 

[Norton]

You might need to hit Update since a BOINC sever was down for some hours and your buffer could have run dry. I did run dry on one rig and if I hadn't asked for new jobs I would have lost yet another two hours on top of the 1½ hours I lost already. I had +1300 wus waiting to be sent from tree rigs.

I had 350 wu's unreported on one rig... meaning it had 50 or so left to work on. The other two rigs were reporting properly.

Thanks for the heads up!

 

[[Ion]]

I lost about four and a half hours on one of my rigs--not sure about the others yet

@Brandon: No, at least it didn't with my GTX460s.

 

[t_ski]

My main rig ended up switching over to CCPU WU's at some point and ran that way for over an hour, then switched back to GPU WU's. At least with the new config file it will do that and the whole day isn't a loss.

 

[HammerON]

I didn't know that. I like that feature

 

[Norton]

FYI- BOINC did a software update yesterday (1/9). The 7.0.42 version BOINC Manager has been replaced with version 7.0.44.

The former version no longer exists- I'm still looking for a list of changes between the versions but haven't found anything yet. :shadedshu

 

[agent00skid]

This thread is good for that.

http://boinc.berkeley.edu/dev/forum_thread.php?id=6698&sort_style=&start=60

 

[BarbaricSoul]

My main rig ended up switching over to CCPU WU's at some point and ran that way for over an hour, then switched back to GPU WU's. At least with the new config file it will do that and the whole day isn't a loss.

My 2600k system is running 4 CCPU WUs and 8 GPU WUs at the sametime. Has been ever since I got multiple WUs going on my 7970.

 

[brandonwh64]

Trying to get laughingmans WCG to work

He has a FX 8 core
two 7770's

And he wants to run 4 WU's per card. what would be the proper xml file for him.

WE are pulling our hair out trying to get it to work.

 

[Norton]

Trying to get laughingmans WCG to work

He has a FX 8 core
two 7770's

And he wants to run 4 WU's per card. what would be the proper xml file for him.

WE are pulling our hair out trying to get it to work.

MJost of this info is on Post #1 of this thread:

Download and install BOINC 7.0.44
-Attach to WCG account- do the initial downloads, click no new tasks, and shut down BOINC
- Make a cc_config.xml file (put in Program Data> BOINC folder)

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

Use this app_config.xml (1 cpu per 0.25 gpu- total of 8 gpu wu's)

<app_config>
   <app>
      <name>hcc1</name>
      <max_concurrent>8</max_concurrent>
      <gpu_versions>
          <gpu_usage>.25</gpu_usage>
          <cpu_usage>1.0</cpu_usage>
      </gpu_versions>
   </app>
</app_config>

Restart BOINC
*Note- Check the project profile on WCG to make sure HCC project and GPU work is selected

 

[brandonwh64]

His shows its running 1 ATI and the rest off the cpu LOL

 

[Norton]

His shows its running 1 ATI and the rest off the cpu LOL

Shutdown BOINC and restart it (BOINC- not the rig)- close it down from the system tray too and make sure stop running tasks on shutdown check box is checked.

It may have to be restarted once or twice to start picking them up

 

[TheLaughingMan]

Ok, not it is not working with the GPUs at all. I give up.

 

[Norton]

Ok, not it is not working with the GPUs at all. I give up.

There are a few spots that need to be checked to get it going- it's pretty basic but if you miss a step you're stuck. Drop me a PM if you want some help walking through it....

 

[m&m's]

Trying to get laughingmans WCG to work

He has a FX 8 core
two 7770's

And he wants to run 4 WU's per card. what would be the proper xml file for him.

WE are pulling our hair out trying to get it to work.

Using the utility.

How many work units do you want to run at a time? 8
How many work units do you want per card? 4
How many threads does your computer have? 8

It should work since as for now I've not received a single complaint.
If it doesn't work, it will at least mean that the problem is not the app_config.xml.

 

[BarbaricSoul]

Using the utility.

How many work units do you want to run at a time? 8
How many work units do you want per card? 4
How many threads does your computer have? 8

It should work since as for now I've not received a single complaint.
If it doesn't work, it will at least mean that the problem is not the app_config.xml.

I wasn't able to get multiple WU going without using M&Ms utility

 

[[Ion]]

I didn't use the utility--I just manually created the config file on my i7-860 Thursday. I do believe I had some issues at first, but resetting the project fixed it.

Doing 8 WUs on the HD7930 w/ a slight OC (1.04GHz) and 3 other WUs on the i7 and I'm getting WUs done in ~9:00-10:00 most of the time
It looks nearly as good as the HD7950--but I'll need to do some more tests to know for sure.

 

[stinger608]

Also it should be noted that when starting M&M's utility to right click on your mouse and "run as administrator" to insure the utility opens the command prompt as the administrator.

 

[ThE_MaD_ShOt]

Hell I am still "old school" I am still on the app_info file. Hell it's still working so I won't mess with it until it breaks.

 

[Norton]

Here's a late X-Mas gift from WCG for all of you guys running low/running out of wu's... They upgraded the Quota to 1200 wu's cached per GPU

See below:

We are caught up now and work is flowing freely.

In order to help volunteers keep their machines contributing during these outages, we have expanded some setting that control how much can be cached. We are now using the following settings"

<daily_result_quota>300</daily_result_quota>
<gpu_multiplier>15</gpu_multiplier> <initial_daily_result_quota>5</initial_daily_result_quota>
<max_wus_to_send>30</max_wus_to_send> <max_wus_in_progress>90</max_wus_in_progress>
<max_wus_in_progress_gpu>1200</max_wus_in_progress_gpu>

*Note- the code listed is on their end- we don't have to change anything on our end

Link:
http://www.worldcommunitygrid.org/forums/wcg/viewpostinthread?post=408493

 

[johnspack]

Need help.. installed newest version of boinc, used the utility for multiple gpu units, checked the xml file and it looks the same as op, but I can't get any hcc wus at all. Tried resetting multiple times, no good. This is on a fresh install of windows with a fresh install of 7.0.44. The manager also reports an error of: app hcc1 not found in app_config.xml. I'm confused?