I have two memory questions about hardware encode:
1. How many memory bandwidth will be cost when encode 1080p30fps avc?
2. Intel CPU is Unified memory architecture, it means GPU and CPU use the same memory, Why the performance are different when using system memory or video memory?
I don't think there is an answer to your question #1 since this related to the configuration of the pipeline, how many buffer is allocated, if the system memory only or system/GPU mixed memory. But I think you might do some test or calculation.
To the question#2, the reason is, this is because the processing(encode/decode/vpp) happens in the GPU, the frame buffer is required in the GPU; if the input/output buffer is in the system memory, the extra copying must be executed before or after the processing, this costs the time and memory space.
Does this answer your question?