App_Config

[0ne.shot]

Because App_info is dated and won't be used much longer and the more efficient way to run multiple WUs on a GPU is the App_config way, what is the code to be placed into a .txt file to run multiple GPUs for HCC for AMD and Nvidia cards? And what directory should this file be placed in?

[Snow Crash]

http://www.xtremesystems.org/forums/...=1#post5174409

[OldChap]

Remove the app_info and put app_config in the same place.

Take a look at this:

http://www.xtremesystems.org/forums/...=1#post5174409

AMD and nVidia are not treated differently

[0ne.shot]

I got it to work. This app_config.xml guide should be stickied. I thought it was .txt and not .xml so it wasn't working. After I changed the filetype all was well. Thanks, guys.

[Mumak]

Is there a way how to setup different CPU/GPU allocation in a multi-GPU system? Let's say a system has HD7950 + HD7770 and I want to allocate 8 CPU threads to HD7950 and 4 threads to HD7770. The app_config doesn't seem to support GPU index # per set.

[OldChap]

Yes there is if there are just 2 dissimilar cards.

set up for the bigger card and run that to confirm that your 7950 is seen as the primary card then, leaving the gpu usage number alone alter the concurrent and cpu usage to suit what you are trying to acheive.

Examples:

<app_config>
<app>
<name>hcc1</name>
<max_concurrent>16</max_concurrent>
<gpu_versions>
<gpu_usage>0.0625</gpu_usage>
<cpu_usage>0.5</cpu_usage>
</gpu_versions>
</app>
</app_config>

The above should be running the 7950 only with 16 wu's (controlled by max concurrent) each using 0.5 cpu threads to total 8 cpu threads. If you now increase max concurrent to 20 the 7950 should continue to run 16 wu's and the 7770 should start to run another 4 wu's

<app_config>
<app>
<name>hcc1</name>
<max_concurrent>20</max_concurrent>
<gpu_versions>
<gpu_usage>0.0625</gpu_usage>
<cpu_usage>0.5</cpu_usage>
</gpu_versions>
</app>
</app_config>

I believe that the cpu usage number is the minimum allocated amount of cpu power each wu gets to work with but that if there is more available (due to offsetting the wu's in time) then they will use it when they need it. This being so I do not think you need to attempt to get the cpu allocation just right in this instance. However feel free to prove me wrong

summary:

gpu usage to get the right number of wu's on the primary card

cpu usage to get enough cpu for the total number of wu's running on the rig

Max concurrent to control the total number of wu's running so this effectively is controlling how many wu's run on the second card.

[Mumak]

Thanks, OC. I thought the app_config is per GPU, so expects symmetric configuration. Will try that soon...

[yojimbo197]

Does the same app_config.xml code work for Linux?

If it does, where do you drop the app_config.xml?

[OldChap]

Yep, and same place...... in the projects folder where I seem to remember everything looks the same as in windows

[StitchExperimen]

How do I code an app_config file for GPUGRID (or is it just "GPUGRID").
I'm running one 660 Ti Nvidia card

and then how do I write after the GPU stuff to do 13 threads of WCG, Human Proteome Folding - Phase 2.
I have a 8/16 core intel processor

<app_config>
<app>
<name>???????????????????????????????????????????? ????????????????</name>
<max_concurrent>1</max_concurrent>
<gpu_versions>
<gpu_usage>1</gpu_usage>
<cpu_usage>1</cpu_usage>
</gpu_versions>
</app>
</app_config>


Thanks for the help!

[OldChap]

app_config is WCG only I believe.



Beyond that I have no clue if the two can be mixed.

The app_config would probably be:

<app_config>
<app>
<name>hpf2</name>
<max_concurrent>13</max_concurrent>
</app>
</app_config>

I suspect that if you want to run GPU Grid and just one instance then just run the gpu grid app too. there should be no need to do anything other than tell WCG not to use GPU

[Snow Crash]

I think it is acemd but check your stout file for the real name

[Mumak]

For GPUGrid, there are the following app names:
acemdlong - Long run tasks
acemdshort - Short run tasks
So you can do something like this:

<app_config>

<app>
<name>acemdlong</name>
<max_concurrent>1</max_concurrent>
<gpu_versions>
<gpu_usage>1</gpu_usage>
<cpu_usage>1</cpu_usage>
</gpu_versions>
</app>

<app>
<name>acemdshort</name>
<max_concurrent>1</max_concurrent>
<gpu_versions>
<gpu_usage>1</gpu_usage>
<cpu_usage>1</cpu_usage>
</gpu_versions>
</app>

</app_config>

[Movieman]

Hypothetically of course but say you had TWO 7970 cards on a very fast 3960X what would you use for a app.config file?
Just hypothetically of course..

[OldChap]

Based on the supposition that anyone trying this might want to start with a moderate number and then progress upwards looking for the most efficient use of the hardware, I would recommend the following settings as a starting point:

<app_config>
<app>
<name>hcc1</name>
<max_concurrent>24</max_concurrent>
<gpu_versions>
<gpu_usage>.083333</gpu_usage>
<cpu_usage>0.5</cpu_usage>
</gpu_versions>
</app>
</app_config>

If on the other hand the potential user wanted to post a rather impressive screenshot but was less concerned about maximising their daily output and was happy to let the separation of work units fall where it may (which in my experience may indeed be rather good), I would probably suggest something more akin to this:

<app_config>
<app>
<name>hcc1</name>
<max_concurrent>96</max_concurrent>
<gpu_versions>
<gpu_usage>.01</gpu_usage>
<cpu_usage>0.12</cpu_usage>
</gpu_versions>
</app>
</app_config>

If you should ever hear of someone wanting to do such a thing, you really should try to get them to post a screenie or two

[jeanguy2]

Hypothetically of course but say you had TWO 7970 cards on a very fast 3960X what would you use for a app.config file?
Just hypothetically of course..

MM, Theoretically I suggest: (Try this for 3 days)

<app_config>
<app>
<name>hcc1</name>
<max_concurrent>24</max_concurrent>
<gpu_versions>
<gpu_usage>0.0833</gpu_usage>
<cpu_usage>0.5</cpu_usage>
</gpu_versions>
</app>
</app_config>

and then: for other 3 days:

<app_config>
<app>
<name>hcc1</name>
<max_concurrent>20</max_concurrent>
<gpu_versions>
<gpu_usage>0.1</gpu_usage>
<cpu_usage>0.6</cpu_usage>
</gpu_versions>
</app>
</app_config>

Compare PPD after the two trial periods and fine tune up or down. Make sure that you space the WUs daily so that they do not bunch up. I now use 20 WUs concurrent with ny two 7970 in Xfire.

Have fun MM...

PS: MM, adding a second card will raise the temp of the top card by 10 to 12 deg C. Monitor invalids closely after installing second 7970 and be ready to up fan speed and or voltage.

[Movieman]

Thanks guys.......hypothetically of course!

[bluestang]

Wouldn't you be better off putting them in separate hypothetical rigs?

[Movieman]

Wouldn't you be better off putting them in separate hypothetical rigs?

If I had that hypothetical option, yes..

[[XC] serlv]

Just started crunching, yesterday, with a new card, on a new install of Win 7.
Seems to be going smoothly; downloading, crunching and uploading.
However, this appears, in red, in my event log:

app beta not found in app_config.xml

Is that anything to worry about?



Also, will this app_ config file work best for me? Or should I make some adjustments?
I'm running a dual Harpertown rig with a Sapphire 7850 OC Edition.

<app_config>
<app>
<name>hcc1</name>
<max_concurrent>16</max_concurrent>
<gpu_versions>
<gpu_usage>0.0625</gpu_usage>
<cpu_usage>0.25</cpu_usage>
</gpu_versions>
</app>

<app>
<name>beta</name>
<max_concurrent>16</max_concurrent>
<gpu_versions>
<gpu_usage>0.0625</gpu_usage>
<cpu_usage>0.25</cpu_usage>
</gpu_versions>
</app>

</app_config>

[Movieman]

Scott:
" app beta not found in app_config.xml"
Non issue pal..
The rest I'm not up on enough to comment on.

[[XC] serlv]

Ok, then. I'll be looking to see if this helps at all. I see you'll be passing my ass, like today. That is not good.
Whood a thought, eh?

[Movieman]

Ok, then. I'll be looking to see if this helps at all. I see you'll be passing my ass, like today. That is not good.
Whood a thought, eh?

I passed you earlier today and honestly, not something I ever expected or wanted to do..
emailed you earlier on this..

To the rest of you guys especially the newer guys, Scott is one of the people who founded the DC teams here on XS.
A very special guy and someone I'm proud to call a friend and no, thats not becaus he has put me up on my three trips to CES..

[OldChap]

Just to clarify a little Scott, Most folks who have an interest in getting badges are short of runtime on Beta's. The app_config I posted was to help these guys in the event that any Beta testing is done. The downside is that for folks that run exclusively GPU tasks this will only help if any GPU Beta's are run but for those folks running some cpu too this should help them snag a few normal cpu Beta's.

The reason it shows up in the log is that the rig has not done any yet. Once it has done some (and downloaded the associated files) the log entry goes away.

[StitchExperimen]

got it working