I want people to look at the HPC cluster as a tool, like a shovel or a drill. The thing is, it’s a tool that’s not really easy to use, so I see my role here as enabling people to use this tool, helping them get more out of it. I’m proud that I don’t just make software tools available, I also take care that they run in the best possible way. I take care of all the optimisations, tuning for the particular CPU instruction set, all the technical details. From my point of view, scientists should not even be bothered about these details. They should just be able to log in, run their software, and have it do its job in a reasonable time frame, in a repeatable manner.
Shortly after I started at EMBL, I approached the scientists and asked them what their needs were. From their responses, it was clear that people doing image processing were asking for GPUs. So based on that feedback we bought a GPU machine. It took a few months to set up the whole chain of software and libraries so that we could use it. But thanks to that investment, image analysis that used to take about 30 days to run on 250 CPUs now takes around 30 hours on the GPU. We’ve since acquired another two GPU machines, as demand has risen.