I suggest referring to system-monitoring-center for how to obtain these statistics for more devices in a semi-uniform way

https://github.com/hakandundar34coding/system-monitoring-center/blob/e0c6246d38af487a378adf486eed61485eea7566/src/Gpu.py#L432-L711

Doesn't look like system-monitoring-center uses that path either, but just wraps nvidia-smi instead.
If nvidia-smi works then mangohud should also work as they both use the nvml library.
Do you have nvml enabled in your mangohud build?

Nvml and Nvidia smi don't exist on the Jetsons.
I helped write system monitoring center so can guarantee it does use the /load path for this system. Read my linked code again and scroll to the bottom because you missed it. It is the last case

@theofficialgman hi german, salvador here. you had it easier then. on my sbc, load also shows both freq and load. I used this config to show it.

### Display custom text
custom_text=Mali-G610
### Display output of Bash command in next column
exec=awk '{split($1,a,"@"); printf "GPU usage: %d%%, frequency: %.2f MHz\n", a[1], a[2]/1000000}' /sys/class/devfreq/fb000000.gpu/load

it's a decent workaround

@neofeo I personally just forked mangohud a long time ago since making this issue and implemented it and lots of other tegra specific clocks/loading myself. Ram bandwidth, clocks, gpu load, gpu frequency, VIC/NVDEC/NVENC frequencies, cpu temp
master...theofficialgman:MangoHud:master-l4t