The CMS experiment at CERN uses a two-stage trigger system to filter and store events of physics
importance: a hardware-based Level 1 (L1) trigger that uses fast electronics (based on FPGAs and
ASICs) to process data in a pipeline fashion at 40 MHz with an output rate of around 110 kHz
and a software-based High-Level Trigger (HLT) run on computer farms with an average output
rate of around 5 kHz. Many novel trigger algorithms, coupled with technological developments
such as heterogeneous computing in GPUs were developed to cope with the increased centre of
mass energy, instantaneous luminosity and the physics needs of Run 3. This talk summarises the
performance of the CMS HLT during the first year(s) of Run 3.