The performance data such as GPU utilization will change with different pipelines, models and configurations. The performance data in https://docs.nvidia.com/metropolis/deepstream/dev-guide/index.html#page/DeepStream_Development_Guide/deepstream_performance.html# is to show deepstream capabilities on different Nvidia platforms. It is not to show the limitations of the platform. For example, on T4, even with the same configuration and model in the test case, more than 35 streams can be supported.
The performance data means nothing if no precondition is identified.
Deepstream will use various accelerators such as VIC, GPU, DLA, NVDEC and NVENC in the Nvidia platform, any of the resources and CPU can be bottleneck for different application, different models and different configurations.