Before today, AMD says a high-priority task on a Radeon graphics card would have to request GPU resources using preemption. In that case, the GPU would temporarily suspend its other work, process the interruption, and return to its regular workload. AMD says preemption is a sub-optimal approach for time-sensitive tasks, since it doesn't provide any guarantee that a task will start and end within a given time frame. The company says that task-switching overhead and other delays associated with this method could also manifest as stuttering or lag in an application.
The Quick Response Queue, on the other hand, gives developers a special asynchronous queue where tasks can get preferential treatment from the GPU while the chip continues to perform other work. AMD says that since GCN asynchronous compute engines are both programmable and manage scheduling in hardware, it can enable the Quick Response Queue with nothing more than a software update on second-generation GCN GPUs and newer.
As an example of what this queue can do, AMD notes that Oculus implemented the asynchronous timewarp (or ATW) feature in version 1.3 of its SDK using the Quick Response Queue for AMD hardware. The company claims that using this feature makes it more likely that the ATW shader will be able to complete before the next vsync interval, even if it's submitted late in the rendering process. That's important since ATW is meant to reduce immersion-breaking judder in one's VR experience, and a warped frame is better than a dropped one. The company also touts the fact that running the ATW shader asynchronously means that the graphics card can continue to perform other tasks at the same time, like starting work on a new frame.
)
