can more than one work group execute on a compute unit concurrently?


In some forums i have read that, each compute unit can execute 8 workgroups concurrently, i thought we can run only one work group on a compute unit at a time, the reason for my understanding is , in the description of clGetDeviceInfo in opencl-1.2 document its mentioned as "A work-group executes on a single compute unit ". can any one clarrify my doubt.


Any given work-group will execute on a single compute unit. However, there is no reason why multiple work-groups cannot concurrently execute on the same compute unit, and indeed many devices will do this in order to improve device utilisation.