Task Overview

Goals:

• 35kHz
• 1TB/s area detectors
• 50GB/s per GPU, avoid CPU where possible
  • "pipeline" as much as possible

Approach:

• use nvidia/cuda for now
  • in future consider kokkos/hip etc.
• python will be less useful

Algorithms:

• SZ/LC compression
• ML algorithms (peak-finding, triggering on teb?)
• detector calibration
• validation of reduction (stefano)

Infrastructure:

• datadev driver (TID)
• GPU-based file-writing with gpu-direct
  • generating correct xtc headers
• drp-gpu executable
• move to spack

Driver

Meeting with Larry, Ryan, Matt, Ric, cpo on Feb. 9, 2024

...

Diagram of system from conversation with Quincey Koziol and Rebanta Mitra on March 20th, 2024

Task Overview

Goals:

• 35kHz
• 1TB/s area detectors
• 50GB/s per GPU, avoid CPU where possible

Algorithms:

• SZ/LC compression
• ML algorithms (peak-finding, triggering on teb?)
• detector calibration
• validation of reduction (stefano)

Infrastructure:

...

• generating correct xtc headers

...