dollar for dollar, I've understood that gpu tasks are "cheaper" than cpu tasks....eg. If I've got one dollar to spend getting "work" done then using a gpu is more efficient than a cpu.

How does this apply or not apply to gpugrid cpu-only tasks?

Difficult to compare considering we run totally different work on CPUs than GPUs. On CPUs we run docking algorithms while on GPUs we run full atom molecular dynamics simulations.

ya, I know there are a lot of variables.

Trying to decide if I should fund a new gpu for a box here at home or fund a cpu box in the cloud.

Actually, I can refine the question..

If I could spend $1,000 on running gpugrid tasks would you prefer I run GPU or CPU tasks?

If I could spend $1,000 on running gpugrid tasks would you prefer I run GPU or CPU tasks?

This is your choice, but I would buy a GPU. There are more CPUs available for BOINC projects, than GPUs. GPUs are far more power efficient than CPUs (i.e. you can do more science on GPUs by spending the same amount of money). The only factor that could alter this choice if there's no GPU application for the desired goal. So if you want to do protein folding, you have to spend on CPU - if you want to do the "usual" Molecular Dynamics you have to spend on GPU.

Actually, I can refine the question..

If I could spend $1,000 on running gpugrid tasks would you prefer I run GPU or CPU tasks?

Another thought is if you fund a cloud machine in the end your 1000 bucks is gone and you have nothing to show for it except a few stats. If you buy a new gpu you do have something physical to hang on to and use for anything else later on, even just playing games. Theoretically of course YOU could be THE ONE to find the key to everything with your cloud funded machine, but the chances are slim and none, and I hear slim left town a while ago.