jetson nano

CPU and GPU ram is shared
Tensorflow tends to dislike this, sees cache memory as used.
sudo bash -c "sync; echo 1 > /proc/sys/vm/drop_caches" helps.
Unrelated, but dhclient -r -v eth0 && rm /var/lib/dhcp/dhclient.* ; dhclient -v eth0 to renew dhcp lease
Getting “Cannot allocate memory in static TLS block”? export LD_PRELOAD=/usr/lib/aarch64-linux-gnu/libgomp.so.1
Temp area details can be found on https://docs.nvidia.com/jetson/l4t/index.html#page/Tegra%20Linux%20Driver%20Package%20Development%20Guide/power_management_nano.html (search “BSP-Specific Thermal Zones”). Ignore zone 4 - PMIC-die as it is locked to 100C.
Bunch of stuff that tegrastats accesses can be found here: https://forums.developer.nvidia.com/t/source-for-tegrastats-and-or-info-about-querying-overall-gpu-utilization/43000/4

fans

We recommend 15mm long screws for installing the NF-A4x10 and 25mm long screws for installing the NF-A4x20.

M2.5 “YHB” seems to be the best option.

/etc/automagic-fan/config.json

{
"FAN_OFF_TEMP":35,
"FAN_MAX_TEMP":50,
"UPDATE_INTERVAL":2,
"MAX_PERF":0
}

sudo systemctl status automagic-fan etc

/sys/devices/pwm-fan/target_pwm

0-255