Madgraph5 GPU development

513/1-024 (CERN)



Show room on map
Madgraph5 GPU development
Zoom Meeting ID
Stefan Roiser
Useful links
Join via phone
Zoom URL

# Madgraph meeting Tue 28.05.2024
Present: SR, AV (minutes), CV, ZW, Ameya Thete (+ Alex)

## Intro

Ameya 2nd year PhD student on CMS at Madison.

Alex is a high-school student on job shadowing with AV just for the week.

## Round table

### AV

Andrea presents some slides
SR: do we still have an auto c++? - AV probably yes BACKEND=cppauto, will check
CV: there is a gcheck.exe in - AV thanks will do a grep
SR: about warps I fixed the CI today

### SR

SR: as of next week Daniele Massaro will join as a new fellow from Monday next week

SR: in Hamburg workshop learnt about the mcnet tooling workdhop in Durham,
co-organised by Chris Gutschow, SR was considering whether to send Daniele about HDF5
AV: may be a good idea
OM: probably HDF5 is not specific to Sherpa, not sure if Madgraph can use it
OM: also Daniele cannot yet give Madgraph position
SR: Daniele might join the MCNet school at CERN too
AV to OM: would binary HDF5 help for file input/output in madevent orchestration?
OM: since November we already improved another factor ~10 in some cases for file management,
so maybe hdf5 not useful, on the other hand for the grid usage and gridpacks hdf5 may be very interesting 
OM: hdf5 may also become interesting if pythia supports it, i.e. if hdf5 starts becoming
a bit like the LHE format, but remember the format must keep some flexibility
AV: two separate issues however txt vs binary format, and which schema (lhe),
so maybe we could do the same lhe schema with its flexibility also on hdf5 binary

SR: in DESY bumped into CINECA people (Daniele Cesini) and they would be a good contact 
to get some GPUs, also connected to CMS
AV: also in contact with Tommaso Boccali, agree it's good to have resources at CINECA

SR: was in contact with Daniel Elvira who is preparing a HL-LHC document and wanted some numbers,
we also had a meeting with Sapta
AV: thanks maybe can keep me in the loop?
SR: yes sure thanks

SR: reminder, nvidia CAF tomorrow and then AMD CAF mid-June, will circulate the links

### ZW

ZW: will present madgraph at PASC in Zurich next week, had a rehersal yesterday

### OM

OM: did some work for the PRs of AV to approve them

OM: also doing some work to remove stuff, AV did you have a look?
AV: not yet, but very happ if you can look at removing and,
maybe you can also document that?
OM: in general stuff that is inside internal shoudl not be touched because at runtime you are not
sure if the version from madgraph or the local version will be used, so they should not be different

SR: eventually can you OM syncronise gpucpp and gpucpp_wrap?
In particular you need to merge gpucpp into gpucpp_wrap because the latter is very much behind

OM: should we apply for the LUMI AMD call circulated by AV?
Some discussion, interesting but probably no, also because the deadline is soon

### NN

ntr, was busy on other things

### CV

testing the new makefiles from AV, now got it compiling
will start some profiling of a100 gpus

## AOB

SR next meeting would be 11 June
AV not there but go ahead without me


There are minutes attached to this event. Show them.