Looking for details on Tegrastats GPU Utilization


I have been measuring load, temp etc. from the Jetson Nano GPU/CPU during inference to get some insight on how the system is performing. There however is not any information on how the number read by tegrastats from e.g. /sys/devices/platform/host1x/gpu.0/load is calculated. I’m aware that it is a percentage telling the utilization rate, but from what time period and how is it calculated? NVIDIA-SMI/NVML apparently calculated the same statistic from sample period of 1/6 to 1 second, but is it the same here? These are the same things I’m looking for, but Tegrastats instead of SMI.

here’s developer guide for your reference, tegrastats Utility.

Hi, thanks for the reply!

Yes, I’m aware of that documentation. The tegrastats manual however does not tell us how the value in GPU sysnode is calculated (which tegrastas polls). For the CPU as far as i know the load is calculated by comparing active and idle processes between two reads from /proc/stat. What about the GPU, is it active and idle processes too and what interval?

hello joni.suominen,

you’ll see the actual GPU frequency via sysnode, i.e. $ cat /sys/kernel/debug/clk/gpcclk/clk_rate
that’s utility to gather and report the average usage within the period,
for example, you may running with the interval options to collect more generic results. $ tegrastats --interval <int>

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.