Occupancy Grid on dGPU


is it possible to execute the Occupancy Grid algorithm on the dGPU? I’m wondering because there is a lot of CPU RAM allocated no matter what the value of CUDA_VISIBLE_DEVICES is. Furthermore, tegrastats says that the dGPU is not being used (it’s frequency is always at 0%).

Thank you.

Hi m.muehlbauer,
The CUDA_VISIBLE_DEVICES can make CUDA kernels to launch on selected device. The GL calls (display calls) runs on iGPU only. You can also notice in the source code, for vibrante it uses GL-> NVmedia image streamer where as for other cases, it uses GL -> CUDA imagestreamer. As there is no CUDA calls invloved on vibrante, you don’t notice any change if you change CUDA_VISIBLE_DEVICES.