concept maximum work - group size in category opencl

This is an excerpt from Manning's book OpenCL in Action: How to Accelerate Graphics and Computation.
Another rule of thumb to follow when generating work-items is that the global size should be a multiple of the maximum work-group size. Chapters 3 and 4 discussed work-groups in detail, but for this chapter, there are five main points to keep in mind:
No matter how many work-items you generate for a kernel, the maximum number of work-items in a work-group remains the same. If your global size exceeds the maximum work-group size, OpenCL will create additional work-groups. An application may generate a nearly unlimited number of work-groups, but the number of work-groups that can execute in parallel is determined by the number of compute units on the device.
The maximum work-group size depends on two things: the resources provided by the device and the resources required by the kernel. In general, the resources that constrain work-item availability are local memory and private memory. The more memory a kernel requires, the fewer work-items will be available to execute it. Your work-items could access global/constant memory instead of local/private memory, but then memory bandwidth drops. This is an important trade-off.