Choose timezone
Your profile timezone:
Madgraph5 GPU development
→
Europe/Zurich
513/R-068 (CERN)
Madgraph5 GPU development
# Madgraph dev meeting Wed 23 Aug 2023
Present: OM, Fili Optolowicz, AV, Taylor, Nathan (later), Carl (later)
Excused: SH, ZW, SR
# Filip
(Brief recap for Taylor: extensive numerical stability studies with the CADNA tool).
Sent some material to OM.
Now writing the report and will circulate it.
Leaving this coming Sunday! So this is the last week at CERN.
OM: main result is that it is the final nail in the coffin that float is not precise enough!
AV: maybe a tiny bit more nuanced, one result is also that thetool is great and someone
could one day maybe use it to try and do something on specific places where precision is lost (though unlikely!).
# Taylor
Nathan and I submitted a paper to Supercomputing last Friday, we'll see later on if it is accepted.
We got a signoff to include some results on Aurora.
The paper can be amended a bit also after it is accepted.
This includes some madevent results with scaling.
On Polaris/A100, kernel execution is only some 10% while I/O causes some scaling issues.
AV: is this generate events, full workflow?
TC: yes, this is based on some MPI I wrote myself for managing different nodes.
The random seed depends on the MPI rank.
OM: note that the unweighting events is now improved with some work from SH,
which should reduce the pressure on tmp filesystems.
OM: can you please send us the paper about this?
TC: yes will send you the pdf around.
# Andrea
Shows some slides, discussion.
# Olivier
During the hackathon worked with SH on unweighting improvements related to /tmp I/O.
Shows some slides previously presetened at the hackathon.
# Carl
Have been doing some performance tests, have no results to show yet
There are minutes attached to this event.
Show them.
