Felix: At a conference this week. As David suggested found TF which is processed on CPU with peak RSS ~15 GB but not on GPU, identified where it goes OOM and saw also other opportunities to spare memory (at the cost to clear mm more often), will fix soon ~next week, then ask for another round of tests.

Gabriele: Results skewed by bug in the ITS trackleting without primary vertices. Will fix with Felix next week. In the meantime, will check quality of the vertex function with the tracklets computed by the old vertexer