Can the same process be used to build a galaxy installation on a VM with graphics accelerator hardware as that for non-graphics accelerator hardware? In other words, if I build a test installation on non-graphics enabled hardware, do I have to do anything additional / different when building an installation on graphics-enabled hardware to get the tools to use the graphics hardware? Will the tools automatically detect the graphics hardware and use it?
This would be on google compute engine VMs running ubuntu 24.04.
The simple answer is there should not really be a difference, but the more complex answer is that it can depend how the tool you’re using accesses the GPU and whether you’re using any kind of containerization with Galaxy. Is the GPU-enabled tool you’re planning to use with Galaxy already available as a Galaxy tool? Are you running the entire Galaxy instance and the tools on a single instance or clustering? Are you running jobs in Docker or Singularity, or using Conda for dependencies?
In this case the gpu-enabled tool(s) are already available as Galaxy tools. At least the ones I know about. The entire instance and tools would be running on a single GCE ubuntu VM. Not planning on using Docker, Singularity or Conda, at least at the moment. Seems pointless if there is only going to be one instance for a number of years. The VM would start as one of Google’s accelerator-enabled VMs.