Hello. I am a developer in Korea.
I would like to ask you a question about GPU performance.
Please check the following picture.
The picture is the easiest picture I have of understanding the operation of the hardware encoding of the CUDA API (if you misunderstand, I would appreciate it).
Input frame data into the buffer.
When the CPU memory calls the CUDA encoding, the buffer is transferred to the GPU memory.
GPU memory returns the completed data to CPU memory.
The right frame data means input into the buffer in 30 GOP units of work.
The decoded frame data will continue to be pushed into the encoder in units of 30 GOP.
Can I overflow GPU memory at this time?
The size of one frame itself is 4k, but it will cut out some areas from 30 frames.
Sorry for the inconvenience.