NVIDIA Optical Flow SDK - How to get effective performance?

Hello, I’m doing theoretical research based on A100 optical flow algorithm,and I am using A100 to test the optical flow performance, the performance is only about 30fps, and NVIDIA official website said 150fps@4k (or GTC PPT: 300fps@4k )The difference is too far. I used AppOFCuda sample under nvofbasic samples to test. The configuration is as follows:
#a.fast mode;
#b.don’t save visual image;
#c.don’t save output;
I would like to ask if you can provide effective use of sample, or give some effective guidance, thank you very much;