Speaker
Description
The HEPiX Benchmarking Working group has been investigating fast benchmark applications with the objective of identifying candidates that can run quickly enough to avoid wasting compute resources, but still capable of accurately representing HEP workloads. Understanding how the underlying processor microarchitecture affects the results of these benchmarks is important to foresee scenarios where the scaling of HEP workloads and benchmark applications might diverge. In order to characterise the behaviour of these benchmarks, in-depth profiling activities have been carried out. In this talk we present an example of performance analysis, covering several layers of the stack, starting from Python source code down to the number of cycles required to execute assembly instructions.
| Length of talk (minutes) | 20 |
|---|
