coolstream, It's real easy. Do acquaint yourself also with the cc_config.xml available settings to control [if having multiple in system] on which card a project is allowed to run. The basic <ooptions> section code set is:

<exclude_gpu>
<url>project_URL</url>
[<device_num>N</device_num>]
[<type>nvidia|ati|intel_gpu</type>]
[<app>appname</app>]
</exclude_gpu>

Square brackets [ and ] indicate optional lines!

For all, *Strongly* recommended, if you're allowed to install BOINC with administrator privilege, do so. Right click in Windows explorer on file and it usually shows a "Run as administrator".

The latest installer 7.0.52 now more clearly warns that installing as "Service" [the label used in Task Manager], no longer with term "Protected application execution", will usually not permit GPU computing when installed as such, on Windows! In that case untick the service box [there occasionally is a registry key stuck to it being ticked]. See item 4 of just updated screenshot guided installer FAQ [Do read the introductory text!]: http://www.worldcommunitygrid.org/forums/wcg/viewthread_thread,28971

underwater, changing, or inserting app_info.xml would often lead to erasing work that was already on board. That's why there are these elaborate starts and stop procedures, most user-unfriendly [nanoprobe is well experienced with the proper how-to... see his posts]. For app_config, a future client will have the simple function to "read in" the config. The premiss of that is though that the client accepts the changes that way. Often, as with changes to the cc_config.xml <options> a client restart is required. The app_config will probably have additional features such as a line to control the number of threads used by a Multithreaded science... I started the debate at developers and it continues, but has to come since app_info is being banished where it could be set [for the few projects that are MT]. Getting way off topic, so leave this bit at that.

starting with a brand new BOINC installation:
Phase 1
use no app_info just start BoincManager and in settings, TICK the use GPU box
wait until at least one GPU task has downloaded then wait for it to start
once GPU unit has started, File | Exit BOINC

Phase 2
now put your app_info.xml into the the BOINC projects folder (e.g. C:\\Programdata\BOINC\projects)
restart BM and you should see more GPU units starting to be processed
you can check in Events Log window to see what BOINC is doing if you wish.

It's actually not that complicated as long as you have correctly written your xml file, but as Sekerob says, it's probably better for anyone starting this to now dive straight into the approved newer method.

Often it's hidden for the users, but the actual path is printed in the startup message log of BOINC. Type the location in the explorer addressbar most of the times gives access.

Ty again,
I'm in with the file crunching 3 gpu wu.
they took 11 min for 3 so 3.66 min apiece.
Faster than the 5 or 6 min for 1.
I'm going to try 4 and see what happens.
This machine is a I7-2500K no overclock.
but the cpus should handle it.

Edit: I'm getting the notice 'This project does not support computers of type anonymous' message
I'll read up I've seen that somewhere

This seems to be working well.
With cpu threads left for one more 7770 (coming soon)