The answer most likely is: it depends.
Presently (AFAIK) only NV cards support running of different kernels concurrently, such as display/compute. But even with this support, if the program is written in an aggressive manner, it will flood the GPU with compute kernels, which have higher priority to the thread scheduler, and as a result they can block display kernels to reach the ALU.
Out-of-order exec will not change this issue, however new architecture will be able to do both of them at the same time, however that will require OpenCL apps to be written in a manner that they intentionally underutilize the GPU, otherwise compute kernels will still prevent display refresh.
It was asked that OpenCL should be able to set priority to kernels in order to let apps run "in the background" of display without tricky coding.
