WCG with GPU: a guide

grumpydaddy

Folding Team Leader
Jan 20, 2008
3,166
0
You can just run a single instance on your GPU but for amd cards certainly, running multiple instances is more efficient.

First thing to do is stop getting new work (Projects>No new tasks) then complete the work you have waiting to run on your rig. then.....

What you need is an app_config file. FOR APP_CONFIG YOU NEED TO RUN BOINC 7.0.42 OR LATER !! Select the one for your os on this page: http://boinc.berkeley.edu/download_all.php There are advantages to using 7.0.54 or later in that you can make changes on the fly (boincmanager>advanced>read config file)

Users should be aware that due to the lack of driver support, namely open cl support, AMD 7xxx cards will not crunch using Windows XP.



Here is one that I have used on 3770K/7950

<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_config>

copy this into notepad then save it in the following folder: program data> Boinc> projects> www.worldcommunitygrid.org ensuring you select the "all files" option in the "save as type" box. Name the file app_config.xml

Unless you changed it already files and folders like this are hidden. Unhide them to get this done: (in search box type folder options and select that from results> you will see this



As in the highlighted area select show hidden files and folders. It also helps to uncheck the Hide Extensions of Known File Types box...see post 3 below.

When you stop and re-start boinc this should then start working on 16 wu's running concurrently on your graphics card but your rig may have a number of other projects waiting to run. Go to your my grid page and change your profile to stop cpu work then abort any cpu tasks you may have in cache.

To get the best out of your rig it is necessary to spend some time separating in time the concurrent wu's so that they are all at a different stage of completion at any one time. It needs to look a little like this:



May I suggest a figure of around 40 seconds between wu's to start with until you get a feel for how your rig is performing. It is the figure I use on an overclocked 3770K or 2600K. Unfortunately this is not an exact science, as you can maybe see from the pic, so you may find that, over time, the wu's tend to group together. If you are one that wants absolute best performance it is necessary to re-allign regularly (daily) or after you use graphics for anything else like watching vids

You may want to experiment with how many wu's to run concurrently so here is a little insight as to what each figure in the config file is doing. You can cut/paste this version with its descriptions as your app_config file if you wish. This one runs 8 concurrent GPU wu's and no cpu wu's on a 4 core no h/t cpu.

<app_config>

<app>

<!-- the name tag is the name of the application you want to control with this app_config file-->
<name>hcc1</name>

<!-- set max_concurrent to the total number of both CPU and GPU WUs to run at a time on your RIG-->
<!-- this line is unnecessary but adds control especially for those with two different cards in a rig-->

<max_concurrent>8</max_concurrent>

<!-- gpu_version tag is where you take control of how BOINC allocates your resources for this GPU app-->
<gpu_versions>

<!-- gpu_usage calucation is 1 divided by the number of WUs you wish to run on any single card-->
<!-- Don't change this when you add a card then each card will run this number. (if you have enough CPU)-->

<gpu_usage>0.125</gpu_usage>


<!-- cpu_usage is how much cpu resources BOINC will reserve to run each gpu WU-->
<!-- calculation is the number of CPU cores/threads you allow for all GPU work divided by the number of GPU WUs-->

<cpu_usage>0.5</cpu_usage>

</gpu_versions>

</app>

</app_config>


Feel free to ask any question if I have been unclear on something
EDIT at the time of writing there are probably around 50-60 days of work left for gpu at which time things may revert to cpu only at least for a while.
 
Last edited:

rickss69

New member
Feb 7, 2010
518
0
Grumpy - I have been using this auto-config utility... techPowerUp! Forums - View Single Post - Guide to running multiple GPU work units

Can you confirm it is doing it's job?

Edit: Put the 7970 in the 980X gamer - Running 8 WU's on the card it's taking about 9 minutes to complete...sound about right?

Revised - It's completing around 7 minutes now...probably need to add more WU's to the card.

Edit: Here it is with 12 WU's running to about 9 minutes completion...

7970-12WUs_zps82e2162f.jpg

 
Last edited:

grumpydaddy

Folding Team Leader
Jan 20, 2008
3,166
0
Show us a screenshot of boincmanager active tasks. It should be a variation of this:




See how the progress of each wu is offset in time. the above is not perfect as it showing what happened after 24 hours running. There is a tendency for wu's to clump together

I don't doubt that TPU idea works but once you are up and running it is now easy to adjust the numbers in notepad and save then from the advanced tab> read the config file.

If you installed in the standard place then go to program data> Boinc> projects> World Community Grid.org >app_config.xml and open with notepad make your changes and save (remembering the all files requirement) after this you will be able to open notepad, select open and the all files option and it will be there waiting for you to open and edit
 
Last edited:

rickss69

New member
Feb 7, 2010
518
0
I'll make a lower resolution screenshot later. I'm assuming I need to stop each WU for a brief period in order to separate them for completion time? What is the effect if you don't do this?
 

grumpydaddy

Folding Team Leader
Jan 20, 2008
3,166
0
The next step

To get the best results you might consider overclocking your GPU. There is no need to do anything with memory speed it seems to make no difference.

Core clock on the other hand is a different matter.

Folks have their favourite methods of overclocking but for WCG there is no need to use any testing like uniengine etc.

Cards have been shown to run WCG faster than they will run for games.

There are often two problems encountered, one is a voltage wall and the other is a clock wall imposed by the bios in conjunction with the overclocking software.

In my experience the most effective tool to use for overclocking GPU for use with WCG is Trixx https://www.sapphireselectclub.com/ssc/TriXX/ It may seem like a clunky interface but it gets the job done (be aware that to the right of the digital readout is a slider giving access to voltage control.)

You will need HWInfo or Gpu-z to monitor voltages/temps of all parts though.

My view of overclocking is to choose the voltage that you are prepared to run at then increase the clock 25 at a time let it run a couple of hours each time, check your results have no invalids, increase, repeat until you see invalids and then drop back a step. Fine tune if you want to.

Keep an eye on not only your gpu temps but your vrm temps too.

Have fun folks....
 
Last edited:

grumpydaddy

Folding Team Leader
Jan 20, 2008
3,166
0
I'll make a lower resolution screenshot later. I'm assuming I need to stop each WU for a brief period in order to separate them for completion time? What is the effect if you don't do this?


Wu's tend to bunch together unless like me you investigate the optimum number to run to combat this.

It may be necessary to re-align on a daily basis unless you run a large amount concurrently.

When running 16 or more 2, 3 or maybe 4 together is not the end of the world but more will reduce you output.

I believe that running 16, 24, 36, 48 resolves the problem for the most part or at least reduces the frequency of re-alignment

It may take a second or two longer on average to complete though

EDIT yes start each wu about 30 seconds apart to give you a feel for a better frequency.
 

grumpydaddy

Folding Team Leader
Jan 20, 2008
3,166
0
There are a couple of ways to approach this:

run the total number you want then do exactly that ... Highlight the wu in question and suspend it .... another will start.... watch the runtime of that one and highlight another when the time has elapsed to space it behind the first one you did suspend the second one. there is now a time gap between the two. continue to do this until all are spaced apart.

Alternately if you are starting from a clean page and activity suspend; within the tasks window select the first wu then shift+down (slider) to the last wu and using the button on the left suspend (all wu's) Now in activity run always.

You can now select each wu in turn and resume one wait 5 seconds and resume another wait 6 seconds and another and another adding a second each time. this will give you a spread that you can use the first method on to better space the wu's.
 
Top