Trying to get an educated guess how long it should take to process a 4K (3840x2160) RGB image using a cuda kernel that uses a 3D texture that represents a 3D lookup table. Assume one thread per pixel for processing and the maximum allowed threads running in parallel. Reminder this is for the Jetson Nano.
We don’t have existing data of this usecase to share now. There are samples in
It might be possible to apply patch of running 3D texture to the samples and do profiling. Please share what 3D texture APIs you use in your usecase.
Well my implementation is more than fast enough for 4K so guess I don’t need a number to compare against anymore.