I have a stock image of the orin Nano/NX and two devkit : 1x Orin Nano 8GB, 1x Orin NX 8GB.
I am running benchmarks on both and I get better CPU results on 7z benchmark with the NX, as expected.
But while trying glmark2-es2 benchmark, I’m getting significantly worst results on the NX.
I get 2575 with the Nano, and 2000 with the NX.
I also get the Overcurrent warning on 2 of the tests in the benchmark. I tried using the power configuration tool with MAXN option, but I get the same results.
I also tried to create my own image using the SDK with the config jetson-orin-nano-devkit, and still same results.
I may be doing something wrong. Can someone enlighten me on what could happen?
Further than setting MAXN power mode, did you later maxed the clocks with:
sudo jetson_clocks
Sorry I only have an AGX Orin running a standard L4T R35.3.1 from NVME for trying.
I built and installed with:
cd ~/Desktop
git clone --recursive https://github.com/glmark2/glmark2.git
cd glmark2
# Using meson 1.2.1
meson setup build -Dflavors=x11-gl,x11-glesv2 --prefix=/home/nvidia/Desktop/glmark2-install
ninja -C build
ninja -C build install
cd ../glmark2-install/bin
./glmark2
./glmark2-es2
For reference, glmark2 gave score 3644, all tests passed, while glmark2-es2 returned score 3329 but with tests conditionals, function and loop failing to run (no idea what is the root cause).
Not sure, but I’d advise to have similar disks for rootfs (performance may vary a lot even for reading if using different SD cards) and use same display monitor for benchmarking.
Someone more skilled may better help.
Please set to this mode and give it a try. In MAXN mode, all engines can run to maximum clock and if either engine drains significant current, it may trigger over current in high system loading condition.
Hi, doing jetson_clocks yielded better results than the nano. I also had to fiddle with the power settings like DaneLLL suggested. The 20w profile had better results than before, but less than “unlocking” everything.