Currently we restrict ourselves to GPU nodes due to the performance issues of software openGL in CPU only nodes.
There are far more CPU nodes, and switching to the CPU based VMs will possibly increase the robustness of the infrastructure (would avoid issues relating to nvidia driver updates, or broken virtualGL).
Currently we restrict ourselves to GPU nodes due to the performance issues of software openGL in CPU only nodes.
There are far more CPU nodes, and switching to the CPU based VMs will possibly increase the robustness of the infrastructure (would avoid issues relating to nvidia driver updates, or broken virtualGL).
We might start with building and testing this library: https://gallium.readthedocs.io/en/latest/gallium/drivers/openswr/faq.html