Hemepure benchmark #371
Hemepure benchmark #371
Conversation
connoraird
commented
Mar 21, 2025
connoraird
commented
- This relies on some changes from Gromacs benchmark #370 in order to work on cosma8, kathleen or myriad
- Adding benchmarks for hememure on CPU and GPU
connoraird
force-pushed
the
hemepure-benchmark
branch
from
224b2b3 to
6760333
Compare
benchmarks/apps/hemepure/hemepure.py
Outdated
| self.perf_patterns = { | ||
| 'Runtime': sn.extractsingle(r'\[Rank \d+, (?P<runtime>\S+) s, \d+ kB] :: SIMULATION FINISHED', | ||
| self.stdout, 'runtime', float, item=-1) | ||
| } |
There was a problem hiding this comment.
When I asked Jon McCulloug about performance metrics, his response was
Regarding performance metrics, time steps / wall clock does make sense, though for LBM codes this is usually expressed in terms of millions of lattice updates per second (MLUPS)
Calculated as (number sites * number steps)/(1000000 * wall clock time for simulation). We typically separated initialisation time and simulation time for this. MLUPS can also be expressed cumulatively or on a per core or per node basis depending on the information you want to convey.
I haven't looked at the output so I don't know what's there. If we can extract either time steps/unit of wall time or MLUPS, those would be a better performance metric
There was a problem hiding this comment.
Okay so simulated time steps divided by reported the wall time?
|
|
||
| HemePure is a optimized verion of HemeLB with improved memory, compilation and scaling | ||
|
|
||
| This directory includes a data directory, `input_data/pipe`, which can be used to benchmark both GPU and CPU machines. This data file was taken from https://github.com/UCL-CCS/HemePure/tree/master/cases/pipe. |
There was a problem hiding this comment.
Is this the same as the TestPipe data in https://zenodo.org/records/14859634
There was a problem hiding this comment.
It does not appear to be the same
connoraird
force-pushed
the
hemepure-benchmark
branch
4 times, most recently
from
122fd01 to
3feb602
Compare
connoraird
force-pushed
the
hemepure-benchmark
branch
from
3feb602 to
c551a2b
Compare
connoraird
force-pushed
the
hemepure-benchmark
branch
from
1020592 to
359546c
Compare
