I think the problem most people are having is trying to add a CPU project with an app_info file already in the project folder. That will not work. You need to have some of each project you want to run in cache before adding the app_info so it can "see" the info it needs to work properly. You also need to make sure your app_info file has all the correct syntax to run GPU and CPU tasks at the same time.

Thanks Coleslaw and nanoprobe :-)

Very interesting. I was wondering if it could be made to converge on some stable state, but just the CPU limitation by itself doesn't do it. Unfortunately though, you need to be in a situation where you can saturate the GPU, which is often not the case. If this project were to continue long enough, it would be worthwhile for WCG to build the self-regulation into the tasks themselves, though that might take changes to BOINC too.

Thanks for all replies!
Well, I am pretty sure I won't make it to sapphire on HFCC if I leave that task up to the quad core with no dedicated GPU. I need like slightly less than 200 cpu days and there are not many days left as I understand...

However, at the moment I have now tried to run 12 parallel GPU WUs since my CPU can run 12 threads. Will see what kind of output I get, but it should logically produce more since both CPU and GPU utilization is more likely held at a higher degree of usage.

Well, if ever there will be an app_info file for HFCC I will perhaps have to consider running a few such CPU tasks together with some GPU tasks...depening on how my quad will manage in the mean time.

Yes, that I have done before also...and yes it works. I would however like to run, say, 6 GPU WUs at the same time together with, say, 6 HFCC CPU WUs at the same time. That way I'd have a decent throughput on the GPU side and also I would reach sapphire on HFCC...