Question about tegrastats information CPU clock frequency

Robotics & Edge Computing Jetson & Embedded Systems Jetson TX2
1

Hi,

I am using Jetson TX2 in the production environment and have set the nvpmoddel to Max-N. I have also started jetson_clocks.sh for max performance.

I observe the tegrastats and see the the CPU frequency is not constant (as compared to TX1, which is sometimes used for minor testing)

RAM 2064/7851MB (lfb 922x4MB) cpu [8%@2031,0%@2035,0%@2035,5%@2036,9%@2035,3%@2035] EMC 50%@1866 APE 150 NVDEC 268 GR3D 99%@1300
RAM 2064/7851MB (lfb 922x4MB) cpu [6%@2035,0%@2034,0%@2037,7%@2035,4%@2036,6%@2034] EMC 49%@1866 APE 150 NVDEC 268 GR3D 31%@1300
RAM 2064/7851MB (lfb 922x4MB) cpu [6%@1974,0%@2035,0%@2035,3%@1993,4%@1986,8%@1992] EMC 50%@1866 APE 150 NVDEC 268 GR3D 99%@1300
RAM 2064/7851MB (lfb 922x4MB) cpu [7%@1997,0%@2034,0%@2036,5%@2006,8%@2001,5%@2003] EMC 50%@1866 APE 150 NVDEC 268 GR3D 99%@1300
RAM 2064/7851MB (lfb 922x4MB) cpu [6%@2036,0%@2034,0%@2034,6%@2035,6%@2035,6%@2035] EMC 50%@1866 APE 150 NVDEC 268 GR3D 99%@1300
RAM 2064/7851MB (lfb 922x4MB) cpu [5%@1997,0%@2035,0%@2034,5%@1997,5%@1997,6%@1998] EMC 50%@1866 APE 150 NVDEC 268 GR3D 99%@1300
RAM 2064/7851MB (lfb 922x4MB) cpu [6%@2031,0%@2034,0%@2035,6%@2033,6%@2037,7%@2040] EMC 50%@1866 APE 150 NVDEC 268 GR3D 99%@1300
RAM 2063/7851MB (lfb 922x4MB) cpu [6%@1997,0%@2035,0%@2035,5%@2000,6%@1998,7%@1998] EMC 49%@1866 APE 150 NVDEC 268 GR3D 68%@1300
RAM 2063/7851MB (lfb 922x4MB) cpu [7%@2023,0%@2035,0%@2035,3%@2038,5%@2029,5%@2036] EMC 50%@1866 APE 150 NVDEC 268 GR3D 99%@1300
RAM 2063/7851MB (lfb 922x4MB) cpu [4%@2024,0%@2034,0%@2035,8%@2036,6%@2031,6%@2037] EMC 50%@1866 APE 150 NVDEC 268 GR3D 99%@1300
RAM 2064/7851MB (lfb 922x4MB) cpu [10%@1997,0%@2034,0%@2034,5%@1998,7%@1999,4%@1998] EMC 50%@1866 APE 150 NVDEC 268 GR3D 99%@1300
RAM 2064/7851MB (lfb 922x4MB) cpu [6%@2031,0%@2036,0%@2036,6%@2037,6%@2037,6%@2041] EMC 50%@1866 APE 150 NVDEC 268 GR3D 99%@1300
RAM 2063/7851MB (lfb 922x4MB) cpu [8%@2031,0%@2035,0%@2034,4%@2039,4%@2034,8%@2034] EMC 50%@1866 APE 150 NVDEC 268 GR3D 99%@1300
RAM 2064/7851MB (lfb 922x4MB) cpu [8%@2002,0%@2034,0%@2035,5%@2003,8%@1999,7%@1998] EMC 50%@1866 APE 150 NVDEC 268 GR3D 99%@1300
RAM 2063/7851MB (lfb 922x4MB) cpu [6%@1996,0%@2035,0%@2035,6%@2004,7%@2000,3%@2000] EMC 50%@1866 APE 150 NVDEC 268 GR3D 99%@1300
RAM 2064/7851MB (lfb 922x4MB) cpu [5%@2030,0%@2035,0%@2034,6%@2036,5%@2039,8%@2034] EMC 50%@1866 APE 150 NVDEC 268 GR3D 99%@1300
RAM 2064/7851MB (lfb 922x4MB) cpu [5%@2030,0%@2036,0%@2035,6%@2037,5%@2034,6%@2036] EMC 50%@1866 APE 150 NVDEC 268 GR3D 99%@1300
RAM 2064/7851MB (lfb 922x4MB) cpu [4%@2031,0%@2035,0%@2034,6%@2040,8%@2037,8%@2039] EMC 50%@1866 APE 150 NVDEC 268 GR3D 99%@1300
RAM 2064/7851MB (lfb 922x4MB) cpu [6%@2030,0%@2035,0%@2034,4%@2035,9%@2036,7%@2033] EMC 50%@1866 APE 150 NVDEC 268 GR3D 99%@1300
RAM 2064/7851MB (lfb 922x4MB) cpu [6%@2031,0%@2035,0%@2035,4%@2038,4%@2037,8%@2035] EMC 50%@1866 APE 150 NVDEC 268 GR3D 99%@1300
RAM 2064/7851MB (lfb 922x4MB) cpu [7%@1995,0%@2034,0%@2034,5%@1997,5%@1998,2%@1997] EMC 49%@1866 APE 150 NVDEC 268 GR3D 99%@1300
RAM 2063/7851MB (lfb 922x4MB) cpu [9%@2035,0%@2036,0%@2035,5%@2035,7%@2034,7%@2035] EMC 50%@1866 APE 150 NVDEC 268 GR3D 99%@1300
RAM 2064/7851MB (lfb 922x4MB) cpu [8%@1997,0%@2035,0%@2035,5%@1999,7%@1998,5%@1999] EMC 50%@1866 APE 150 NVDEC 268 GR3D 97%@1300

Although it is still ~2.0GHz, I don’t see such fluctuations on TX1 where it is fixed to 1734MHz.

RAM 1150/3983MB (lfb 507x4MB) cpu [13%,2%,0%,1%]@1734 EMC 2%@1600 APE 25 GR3D 0%@998
RAM 1150/3983MB (lfb 507x4MB) cpu [4%,3%,0%,0%]@1734 EMC 2%@1600 APE 25 GR3D 0%@998
RAM 1150/3983MB (lfb 507x4MB) cpu [2%,3%,0%,0%]@1734 EMC 2%@1600 APE 25 GR3D 0%@998
RAM 1150/3983MB (lfb 507x4MB) cpu [1%,1%,0%,0%]@1734 EMC 2%@1600 APE 25 GR3D 0%@998
RAM 1150/3983MB (lfb 507x4MB) cpu [2%,1%,0%,0%]@1734 EMC 2%@1600 APE 25 GR3D 0%@998
RAM 1150/3983MB (lfb 507x4MB) cpu [2%,1%,0%,1%]@1734 EMC 2%@1600 APE 25 GR3D 0%@998
RAM 1150/3983MB (lfb 507x4MB) cpu [0%,2%,0%,0%]@1734 EMC 2%@1600 APE 25 GR3D 0%@998
RAM 1150/3983MB (lfb 507x4MB) cpu [15%,9%,11%,20%]@1734 EMC 3%@1600 APE 25 GR3D 7%@998
RAM 1172/3983MB (lfb 502x4MB) cpu [20%,10%,22%,12%]@1734 EMC 5%@1600 APE 25 GR3D 0%@998
RAM 1172/3983MB (lfb 502x4MB) cpu [1%,2%,0%,0%]@1734 EMC 4%@1600 APE 25 GR3D 4%@998
RAM 1172/3983MB (lfb 502x4MB) cpu [2%,2%,1%,2%]@1734 EMC 3%@1600 APE 25 GR3D 0%@998
RAM 1172/3983MB (lfb 502x4MB) cpu [0%,1%,0%,1%]@1734 EMC 3%@1600 APE 25 GR3D 0%@998
RAM 1172/3983MB (lfb 502x4MB) cpu [2%,1%,0%,2%]@1734 EMC 3%@1600 APE 25 GR3D 3%@998
RAM 1172/3983MB (lfb 502x4MB) cpu [0%,2%,0%,1%]@1734 EMC 2%@1600 APE 25 GR3D 0%@998

Is this behavior normal or I am missing something?

Thanks.

2

Hi bhargavK, are you seeing the fluctuation with only the Max-N nvpmodel, or other models as well? What about when running ~/jetson_clocks.sh or when there is load on the CPU? In theory what it is changing are the dynamic frequency governor limits, not setting the exact clocks themselves. You can also look around in the sysfs nodes (you can find them referenced inside jetson_clocks.sh) to play around with these clock governor settings.

3

Good point dusty_nv, I will check it.

Currently, I run ~/jetson_clocks.sh and the stats above are reported after running the script. The production environment is running a program like detectnet-video but I imagine it is not CPU intensive.

However, though, I haven’t checked the stats with other nvpmodel.