I’m dealing with image processing with c ++ / cuda. For about a year, I have been trying to keep the stream coming from the gstreamer directly in the gpu memory without putting it in the cpu memory. And I couldn’t handle it. I am using Ubuntu OS and I have a 2080ti graphics card. I am using an nvh265 decoder. I see you can do that on cards like Jetson. How can I do this on the computer? Right now I am getting the data from gstreamer with appsink with opencv videocapture. It outputs Mat. I upload the Mat to the GpuMat. This is without the need for upload. Decode the incoming data in Gpu. I want to keep it in gpu memory. There are 68 videos from different sources. The performance drops a lot when I upload instantly for all of them. Apart from these, is there any other method that I can capture data from FPGA and keep it directly in gpu? Thank you very much in advance.