Alice Weekly Meeting: Software for Hardware Accelerators

Wednesday 6 Aug 2025, 10:00 → 11:00 Europe/Zurich

10:00 → 10:20 Discussion

Speaker: David Rohr (CERN)

10:20 → 10:25 TPC ML Clustering

Speaker: Christian Sonnabend (CERN, Heidelberg University (DE))

Busy writing thesis (especially MC chapter to give it to you (David) and afterwards Silvia).

 

Investigation on cluster flags:

• First ideas about training on flags directly failed because for most clusters the flags are 0 -> Network just learns to produce 0 for everything, so this needs downsampling
• Second idea learn the cluster overlap. This works to some extend

• Two problems need mitigation:
  • Also many clusters with overlap value of exactly 0 -> Removed from training data set now to avoid overtraining. 
  • Network output not bound to be between 0 and 1 without non-linear activation in the last layer: Outputs < 0 are mapped to 0 (sometimes incorrectly)
• Ultimate problem: Cluster overlap does not nicely correlate with cluster flags:

10:25 → 10:30 GPU Parameter Optimizations

Speaker: Gabriele Cimador (CERN, Università and INFN Torino)

10:30 → 10:35 Efficient Data Structures

Speaker: Dr Oliver Gregor Rietmann (CERN)

Summer Student Milla Bramsted

• She is working on benchmarking SoA code on GPUs
• We will track her project in this google doc
• She is adding CUDA kernels to this repo
• 5 CUDA kernels are now running in our benchmark framework
• They are running in AoS and SoA data layout
• Next steps:
  • Plot the results
  • Add manual AoS and SoA as baseline

ALICE O2 CI-Pipelines on NGT Cluster

• A fork of the AliceO2 repo is not in the  NextGenTrigggers (NGT) GitHub organization
• It has a GitHub action running the standalone benchmark on NGT GPUs
• Uses the builds in /cvmfs/alice.cern.ch/ of O2 and dependencies (pipeline takes about 7 minutes)
• Different GPUs are tested in parallel on different VMs
• O2 standalone benchmark works on the all ngt-resources:
  
  • Nvidia H100 188GB NVL
  • AMD Instinct MI300X
  • AMD Radeon Pro W7900
  • Nvidia L40S
    
• We are now using custom .par files
• Next steps:
  • Generate optimized .par files with Gabriele
  • Store results in csv format and plot them?
  • Display the plot in the web gui?
  • Add new architectures to O2 hardcoded ones?

Implement NGT SoA Code in O2 standalone benchmark

• Working on this fork of the AliceO2 repo
• Simplified and optimized the SoA code in the last few weeks
• Everything is running and we have identified the classes apply our SoA code to
• Next steps:
  • Discuss these classes with David
  • Implement the SoA code on those classes

10:35 → 10:40 Following up GPU to-dos

Speaker: Dr Vikas Singhal (Department of Atomic Energy (IN))

10:40 → 10:45 TPC Clusterization / OpenCL / Highly Ionizing Particles

Speaker: Felix Weiglhofer (Goethe University Frankfurt (DE))

10:45 → 10:50 ITS Tracking

Speakers: Felix Schlepper (CERN, Heidelberg University (DE)), Gabriele Cimador (CERN, Università and INFN Torino), Matteo Concas (CERN)

10:50 → 10:55 System Run Coordination Topics

Speaker: Ernst Hellbar (CERN)