Synthetic Benchmark on Curie
From SNIC Documentation
|Name||Synthetic Benchmark on Curie|
|Description||Synthetic Benchmark for PRACE Tier-0 Curie System|
Synthetic Benchmark for PRACE Tier-0 Curie System is a task at PRACE 1IP project (Task 7.4). It aims to benchmark the performance of a cluster-type Curie system which is one of largest HPC systems serviced by PRACE.
From SNIC, Soon-Heum "Jeff" Ko joined this project from July 1st, 2011. He takes the role of performing some benchmark analyses by using HPCC, P-SNAP and mixedMode tools.
- HPCC is a famous benchmark tool which consists of 7 sub-benchmarks, including a famous HPL tool which measures Linpack performance and STREAM benchmark which measures the sustainable memory bandwidth.
- P-SNAP is a tool to measure the system noise.
- mixedMode is a tool to instrument communication bandwidths of different communication modes (collective and point-to-point communications in various configurations).
We took benchmark runs of HPCC, P-SNAP and mixedMode on Tier-0 CURIE Bullx cluster. HPCC runs were performed to figure out Linpack benchmark performance up to 1024 cores, memory bandwidth, and cache miss ratio. P-snap measured the system noise, and mixedMode evaluates different MPI kernel performances on single-, tunneled- and threaded-mode. Those results are reported in PRACE 1-IP final report.
Full details will be updated here.
|Soon-Heum Ko (NSC)||NSC||Application expert||Computational fluid dynamics|