ATLAS GPU Training

31 Oct 2023, 15:00 → 2 Nov 2023, 19:00 Europe/Zurich

Alaettin Serhan Mete (Argonne National Laboratory (US)), Attila Krasznahorkay (CERN), Charles Leggett (Lawrence Berkeley National Lab (US)), Jim Pivarski (Princeton University), Matthew Stack, Robert Searles, Scott Snyder (Brookhaven National Laboratory (US)), Tom Gibbs (NVIDIA), Vakho Tsulaia (Lawrence Berkeley National Lab. (US))

This introductory GPU training session will focus on understanding the basics of GPU programming. It will cover GPU architectures, and how they differ from CPUs, how to write CUDA programs, and how to program and run on GPUs from python.

It will feature interactive exercises. The first 2 days will be run by Matt Stack and Robert Searles from NVIDIA and cover the basics of CUDA in C++.

The last day, which will be devoted to using python to run on GPUs, will be run by Jim Pivarski.

Participants will be given access to the NVIDIA training cluster to run their code.

The tutorial is remote only, and broadcast via zoom. A recording will be made of the sessions.

Each session will run from 15:00 to 18:00 CET

The registration is currently full. However the registration is only to get an account on the NVIDIA training cluster to be able to run the interactive tutorials. Even if you haven't registered, you are more than welcome to connect to the sessions. 

Please email cgleggett@lbl.gov if you would still like to try to register for a place, and I will add you to the waitlist.

NB: on Oct 31, Europe has ended daylight savings time, while the US has not, so please pay close attention to the time difference if you're not at CERN!

 

Google doc for questions

Introductory Material

JupyterNotebook.tgz

Parallelism.pptx.pdf

Preparation Instructions

Python Tutorial

Python Tutorial github

Questions and Notes

Recording, Day 1

Recording, Day 2

Recording Day 3, part 1

Recording Day 3, part 2

Tuesday 31 October

1 CUDA Programming Day 1

Speaker: Jonathan Bentz (NVIDIA)

Wednesday 1 November

2 CUDA Programming Day 2

Speaker: Jonathan Bentz (NVIDIA)

Thursday 2 November

3 Python Programming for GPUs

Schedule

• 0:00 (30 min) Introduce the Python tools; see lecture-slides.ipynb
• 0:30 (20 min) Compute the Z mass in CUDA with Numba
• 0:50 (20 min) Introduce tree-reduction
• 1:10 (20 min) Students work on Project 1: parallel histogram filling
• 1:30 (5 min) break
• 1:35 (20 min) Review Project 1 solutions (don't peek until you've tried it!)
• 1:55 (20 min) Introduce random seeding of parallel algorithms
• 2:15 (20 min) Students work on Project 2: compute area by random sampling
• 2:35 (5 min) break
• 2:40 (20 min) Review Project 2 solutions (don't peek until you've tried it!)
• 3:00 end

Mandelbrot on all accelerators

Python setup instructions