Madgraph5 GPU development

Tuesday 21 Feb 2023, 15:00 → 16:30 Europe/Zurich

513/R-070 - Openlab Space (CERN)

# Madgraph meeting

Present: SH, JT, AV, OM, CV, NN, TC, WH
Excused: SR, ZW

## Round table

OM: ntr

CV: waiting to do performance tests when there is code ready for that if you have something specific for our T4s
SH: T4s are a bit old/weak, why those?
CV: this is what we bought for CMS at our T2, running both ML and reco
SH: do you know if they are single precision or double precision workflows?
CV: not sure, will ask colleagues

NN1: doing some tests on tt+4jet ggttgggg! But did not even manage to generate cudacpp code...
AV to OM: any work on recursion relations in madgraph actually?
OM: yes wrote a paper last year, this might evolve, but the integrator is not good for that
NN2: also noticed something strange, the Fortran MEs built within the SYCL installation give different results
It may be that the source code is different
AV: check the source code, maybe you are using a different upstream commit?

TC: ntr

WH: ntr

JT: working on containerization of scripts
AV: be aware we did a lot of work on this in the benchmarking project (mainly docker and singularity, not podman)
JT: one problem is that nvidia have good support for el7, but not much for el8...

AV1: gave the CAF talk last week or the week before
AV2: upgrading my infrastructure to alma9 from centos7

SH1: not done much, will try to arrange a chat with OM
Doing some unweighting on the GPU, debugging some issues
(this could be interesting to move extra cycles from the CPU to the GPU)
SH2: upgraded the CI to alma9
SH3: proposed an openlab summer student project for using the new fptype that tracks its own precision

## AOB

OM: upcoming meeting in Italy in September organized by Marco Zaro and Fabio Maltoni
https://indico.cern.ch/event/1240244/

Next meeting: Tue March 7th

 

 

 

 

15:00 → 15:10 News

15:10 → 15:30 Topical discussion

15:30 → 15:50 Round table

15:50 → 16:00 AoB