Additional note: it is not obvious how nvpmodel and jetson_clocks interact.
jeston_clocks.sh boosts the clocks for you current config, that depends on last nvpmodel selected.
nvpmodel tries to use a preset of online cores / clocks for optimizing performance/power on different profiles (you may check the details in /etc/nvpmodel.conf).
The bad thing (at least on R28.2-DP) is that if you are running high speed clocks and then select a low power mode with nvpmodel, it can fail. For example, if I select MAXN mode and boost clocks (max performance), then it fails to properly switch to MAXP_CORE_DENVER mode:
sudo nvpmodel -m0
sudo nvpmodel -m4
NVPM WARN: write PARAM GPU: ARG MAX_FREQ: PATH: /sys/devices/17000000.gp10b/devfreq/17000000.gp10b/max_freq VAL: 1134750000
- probably failed due to it is less than current min value 1300500000
NVPM ERROR: Error writing to /sys/devices/17000000.gp10b/devfreq/17000000.gp10b/max_freq: 22
NVPM ERROR: failed to write PARAM GPU: ARG MAX_FREQ: PATH: /sys/devices/17000000.gp10b/devfreq/17000000.gp10b/max_freq VAL: 1134750000
NVPM ERROR: failed to set power mode!
NVPM ERROR: Failed to exec option requests!
So it is better to store the default (slow) clocks after boot with:
sudo ./jetson_clocks.sh --store
that will save the current config into ~/l4t_dfs.conf and you can restore later with:
sudo ./jetson_clocks.sh --restore
In a similar way, if you change the nvpmodel for a performance mode from low power mode, you may have to relaunch jetson_clocks to get max boost.