Hello Nevada Bob,
Go to Advanced View - Projects and make sure that you are attached to WCG. If not, then go to Tools and attach to World Community Grid Make sure that your copy of BOINC is running using your webside profile. If something is not working the way you want it to, post the startup messages from BOINC Manager here so we can see what your setup is and tell us the problem.

You can not use your GPU on WCG yet because we do not have a project with GU code yet. We hope to run a Beta test with some GPU code in the near future.

Lawrence

Nevada Bob...the difference in BOINC versions is that the 6.12.34 is what Berkeley recommends (the creator of the software) and 6.10.58 is the "latest" (more like the last) that WCG officially recommends due to their testing standards. Both are usable for GPU if work is available. However, WCG does not yet have GPU capable work but we are told it is coming "soon". For now, you should take this opportunity to help out other projects with that GPU and get a little experience with it until WCG is ready.
Here is an outdated list: http://boinc.berkeley.edu/wiki/GPU_computing
Not on the list but still supports ATI/AMD cards is Albert@Home, Donate@Home, and POEM@Home

Edit: I forgot to mention, some of the projects require the newer test versions of BOINC which are the version 7.x.x series. These are not guaranteed to be stable, but some of the OPENCL based projects require them. POEM and ALBERT both do.

PrimeGrid is very bad choice for ATI cards, also like SETI. The performance is better than CPUs but very bad in compare with nVidia cards. For ATI cards are the best MilkyWay, Moo, Collatz. I mean that is wasting of ATI GPU power for "nVidia" projects.

Cheers

Some GPU projects for ATI are using one thread of CPU. I don`t have ATI GPU now, so I don`t know which application from which project is using one thread of CPU for GPU work. You can try another ATI GPU projects and you will see which GPU app is computing without CPU.

Edit: you can see here which app is using also CPU. Only Moo and on the hi-performance models HD69xx is MilkyWay is not using CPU for computing.

There is a differentiation of [normal] intense and non-intense CPU utilization of GPU tasks. What I understand is that the non-intense will allow [normal] CPU only based jobs to occupy the same core-thread as the non-intense GPU **, whereas the [normal] intense won't, i.e. there reserve the core. As it is, GPU's require the CPU to control the GPU tasks. The earlier 'petnek' reply points to these 'non-intense' CPU loading GPU projects.

[Sidebar, future talk], the techs of WCG want a [high] intense term added to the way the client manages jobs. This then to control how many of the heavies are allowed to run concurrent. For instance you can have then as many CEP2 jobs in the queue [currently limited through preferences in the website device profile], but then you can configure how many of these are allowed to run concurrent. E.g. you can configure to have a max of 6 in the queue, and 4 to run concurrent. If in FIFO processing the other CEP2 should have their turn, they will be held back until one other of the already running 4 CEP2 jobs finishes first. Suppose then the client scheduler would then advance the queued CEP2 job then first, if it were it's turn in [delayed] FIFO. This would greatly improve the share that CEP2 type science apps are allowed on a client.

At any rate, did not understand how you could have gotten 2 *core* clients (CC) to run concurrently... boinc.exe only ever by itself will allow to start 1 CC [madmen and special setups can configure 2 to run concurrent, but of course, they each get only part of the cycles and increase competition, increase overhead, double memory use, net reducing production compared to 1 client using it all]. Maybe you were talking about two boinc managers? That was possible, and maybe because of the naming convention would allow for them to have 2 started, but by default, from discussion, 2 boincmgr.exe should no longer be possible, unless each is connected to a different core client on launch... one local core client, other remote core client. This is why I've long disabled boincmgr.exe to start and have exclusively gone to the multi-client manager BOINCTasks, which can run on Windows, Linux+Mac [in e.g. wine session], can show all 'connected' clients simultaneous, so we have oddly, mostly at night, slight vibrations [it's the sun/moon I tell you, the moon/sun it is and the gravitational pull variance they exerts]

--//--

Edit: ** There also possibility to sub-divvy the power GPU's to run multiple GPU jobs on 1 card. Not in the know if these then are also classed into intense/non-intense. Get ready to study for a GPU crunching degree [which WCG is trying hard to avoid to be needed to participate here when it comes... we'll see if succeeding ;O].

Edit2: Think you could have ended up with 2 installs because the 64 bit client installs to C:\Program Files\BOINC, where the 32 bit WCG version goes to C:\Progarm Files(x86)\BOINC, but still only 1 boinc.exe would launch.

edit3: Non-intense, non-GPU using projects that can co-occupy a CPU core-thread are WUPROP [pure result statistics collection] and Quake Catcher-Network. Latter I run on a laptop with accelerometer [to protect the HD]. They use from observation 0.05% to 0.1% of 1 CPU thread and latter reports collected information in "trickle"... living here right on the border of a complex earth-crust fracture system. :O|