cuda - Is it possible to handle GPU priority between processes -

I know from CUDA 5.5 that it is possible to have a higher priority kernel, but I think it's only for calls GPU But by the same reference, i.e. this other process does not affect the priority for kernel launch, as long as they have enough CPU time released.

Is it possible to have GPUs on high priority applications, so how can you set the OS to give priority to specific threads?

Heinrich Andresen

operation kernel in a higher priority stream to determine the lower priority currents in the competition. The kernel can be launched in it, the device can contain copy operations and events. I do not believe that, however, after expanding any differences in performance priority, the kernel has been launched on the device, it is completely a driver side-stream scheduling estimator control. This is present in concurrent kernel execution when increasing the range of kernels running on low priority streams on the device.

To do the best of my knowledge, there is currently no single way to influence a single reference in the stream, and there is no way to influence the processes that compete for access to the GPU, at the driver level Priority will be given. There is only one warning that I add it that CUDA 6 probably has changed it, and I still do not have the opportunity to see new things in that release.