Slowdown when increasing the number of compute units

We can modify the IP file about the dispatch module to achieve relatively fair schedule.

--- Quote Start ---

The scheduler dispatches on workgroup units and it's based on compute unit availablity. The ordering of the scheduling to each compute unit is not easily predicable for users so I wouldn't count on any ordering. Expect the scheduler to do a good job keeping the compute units fed with work.

--- Quote End ---