Week 15 Overview

During this fifteenth week, I started working on other assigned tasks within the RADI engine of the Unibotics platform.

First, I was tasked with conducting CPU and GPU consumption tests within RADI to observe if performance and consumption were optimal. I performed these tests on my Windows computer (with a 14-core CPU) and my Linux computer (with a 4-core CPU), monitoring the CPU usage percentage with the ‘htop’ tool. The results showed very high consumption, especially with more modest CPUs, which were pushed almost to their limits.

Additionally, I conducted tests focusing on GPU usage percentage using the ‘nvtop’ tool. The results indicated very low GPU usage overall, suggesting that there is room for improvement in GPU utilization to optimize CPU consumption.

Lastly, upon reviewing the configuration files for RADI launches with Docker-Compose, I noticed they were only set up for launches using the CPU or integrated GPU (Intel). I developed a new configuration to enable launches with Nvidia GPU, adding the necessary code to expose the GPU to the Docker container, and sent it to the development team.

deploy:
  resources:
    reservations:
      devices:
        - driver: nvidia
          count: all
          capabilities: [gpu]