Benchmark execution time against the ~10x native-Geant4 budget #5

New Issue

2026-07-08T13:52:20+02:00

lbogner commented

2026-07-08 13:52:20 +02:00

(Migrated from gitlab.etp.kit.edu)

Benchmark the surrogate's execution time and track it against the target the final design must hit: roughly ~10× native Geant4.

Large nets are acceptable for now — don't over-optimize prematurely — but every architecture choice (WGAN, MoE, rollout parallelization) is ultimately measured against this number. Needs a reproducible per-step / per-shower timing benchmark for the current model and a native-Geant4 reference time on the same events to define "1×", then re-run for each candidate architecture.

Source: Jan & Tobias meeting 2026-07-07 §4. Cross-cutting target, not a one-off task.

Benchmark the surrogate's **execution time** and track it against the target the final design must hit: roughly **~10× native Geant4**. Large nets are acceptable *for now* — don't over-optimize prematurely — but every architecture choice (WGAN, MoE, rollout parallelization) is ultimately measured against this number. Needs a reproducible per-step / per-shower timing benchmark for the current model and a native-Geant4 reference time on the same events to define "1×", then re-run for each candidate architecture. Source: Jan & Tobias meeting 2026-07-07 §4. Cross-cutting target, not a one-off task.

lbogner commented

2026-07-08 16:15:14 +02:00

(Migrated from gitlab.etp.kit.edu)

mentioned in commit bac541240f

mentioned in commit bac541240f580b3cf92c5f49b3f924d7d0596b8d

lbogner commented

2026-07-17 12:32:21 +02:00

(Migrated from gitlab.etp.kit.edu)

mentioned in merge request !3

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: lars/giant#5