Roundtable
Alex
University-stuff
Stephen
Managed HepMC-support for covfi. Now a lot of people got interested in it and asked about support.
Started with testbed documentation.
Anton
Setting up test-playground for gpu-stuff.
Wiped some space to be able to continue with work.
Benjamin
Midcycle report went fine.
(Meetings in general more about hardware choice not ML ws handwritten code. ML might favour GPUs more than FPGAs)
Seems so far to work for the legacy chain. Resolution looks fine, pulls sometimes off.
Attila: How do you get space-points from ACTS? ->
ACTS does not generate spacepoints from strips, ATHENA makes the strip-spacepoints. Dump ntuples and read xyz-coordinates into examples.
Alexis added many overlapping rules. Might be complicated to translate to traccc.
Difficulty are e.g. overlaps between different eta and phi wheels. This is problematic for performance. GNN is already facing performance issues in the strips.
Raises the question, since we are making a hardware-choice, if we should give different weight to space-point formation and fitting. Current spacepoint-formation is quite detector specific, maybe we can generalise it?
Next focus:
- Check status of traccc-fitter for a new student.
- sofie-meeting
Attila
Preparing for GPU-tutorial, still some things that do not work yet. Using promata supercomputer. aroung 1/4 of H100/person.
Paul
Gen3: finished on generic detector. Navigation on generic is also working and didn't require a lot of additions to then Gen1 navigation.
Performance measurements: additional intersections slow down navigation. Optimising data structures because push-backs can be bad for compiler. Now intersection performance is good enough to be similar to old navigation.
Focus on getting material for workshop ready.
HepMC misses still a few things for data-generation.
Luthien
University classes.
Looking now into tracking efficency this week. Started to run IDPVM.
Andreas Stefl
ACTS-ATHENA-performance:
sympy brought us into 5% range but then we got worse to 18% slower. The steps don't correlate to special ACTS-releases, so maybe ATHENA changes?
Run with 100 events, so multi-threading causes fluctuation.
There should be no major development on ATHENA anymore.
Machines change every year and we have exclusive access.
Alex: Maybe calculate correlation to get that out? -> maybe going into too much detail.
