How to increase GPU utilization for reference application

For the reference application, how can I increase the gpu utilization?, I see the performance reported by Nvidia on T4 is around 35 FPS and ~35% of gpu utilization, but what is the bottleneck preventing to have a maximum gpu utilization and therefore improving throughput performance? ID0E0BB0HA](NVIDIA DeepStream SDK Developer Guide — DeepStream 6.1.1 Release documentation)

the bottleneck is on decoder capacity, you can run deepstream-app as the doc config set, and monitor decode engine utility, nvidia-smi dmon -i “GPU id”
below is run with source number 35 and built-in stream 1080p file on T4, config same as doc config set but without nvinfer and tracker.

gpu pwr gtemp mtemp sm mem enc dec mclk pclk

Idx W C C % % % % MHz MHz

0 40 33 - 5 11 0 64 5000 1590
0 40 33 - 5 12 0 64 5000 1590
0 38 33 - 4 10 0 54 5000 1590
0 39 33 - 5 11 0 66 5000 1590
0 41 34 - 4 10 0 61 5000 1590
0 38 34 - 5 11 0 60 5000 1590
0 40 34 - 5 12 0 66 5000 1590
0 39 34 - 5 11 0 64 5000 1590
0 39 34 - 5 12 0 65 5000 1590