Throttling on custom carrier board

Hi

We have developed a custom carrier board for the TX2 module. The compute performance of the TX2 module is however very poor in combination with this board. It appears as the module is performing some kind of CPU frequency throttling even though none of the build-in temperature sensors report more than 60 degrees during operation. By first executing '$nvpmodel -m 0 && ./jetson_clocks.sh’and testing with ‘stress -c 6' there is a major difference between the dev. board and our custom board regarding the CPU clock freq when using './tegrastats’:

Dev. board: RAM 908/7851MB (lfb 1493x4MB) cpu [100%@2030,100%@2035,100%@2035,100%@2030,100%@2031,100%@2033] EMC 0%@1600 APE 150 GR3D 0%@114
Cust. board: RAM 800/7851MB (lfb 1600x4MB) cpu [100%@1021,100%@1115,100%@1356,100%@1337,100%@858,100%@705] EMC 0%@1866 APE 150 GR3D 0%@114

The cpuinfo_max_freq are set to 2035200 in both cases, and the crit_current_limit_* are set to 8190. A full readout from the power management chip during idle can be found below.

Ina3221 @ 0x40

crit_current_limit_0 16380
crit_current_limit_1 16380
crit_current_limit_2 16380
crit_power_limit_0 114397
crit_power_limit_1 114397
crit_power_limit_2 78624
in_current0_input 8
in_current1_input 40
in_current2_input 8
in_power0_input 55
in_power1_input 279
in_power2_input 38
in_voltage0_input 6984
in_voltage1_input 6984
in_voltage2_input 4800

Ina3221 @ 0x41
crit_current_limit_0 8190
crit_current_limit_1 16380
crit_current_limit_2 16380
crit_power_limit_0 57198
crit_power_limit_1 114397
crit_power_limit_2 78624
in_current0_input 208
in_current1_input 8
in_current2_input 40
in_power0_input 1452
in_power1_input 55
in_power2_input 192
in_voltage0_input 6984
in_voltage1_input 6984
in_voltage2_input 4800

I have observed that the in_voltage0_input, in_voltage1_input and in_voltage2_input drops to 5744, 5752 and 4808 during load on both addr 0x40 and 0x41.

Even if I try to fully load both the GPU and CPU simultaneously I am not able to get the TX2 to draw more than 8 watts. I do not see any dmesg errors related to this issue during boot or during full load.

Any suggestions that can help identify the root cause are welcomed, thanks.

Hi, did you check the difference between your board and dev kit? It’s not easy to tell what might cause this…

We are checking the power supply components at the moment.