After the aiXcelerate 2023 was replaced by the VI-HPS Porting & Tuning Workshop, the traditional aiXcelerate took place again last year.
In the following blog post, we look back at aiXcelerate 2024 and take a look at the upcoming PPCES 2025.
Source: Own illustration
aiXcelerate 2024
aiXcelerate is an annual tuning workshop for HPC users. It includes lectures, which are open to everyone, and hands-on parts where participants apply the learned concepts to their own codes. Each year we focus on a different hot topic. Last year, from December 9-11, 2024, aiXcelerate focused on machine learning (ML) on NVIDIA GPUs (graphics processing units). This time, the aiXcelerate had a hybrid format; the lectures were held at the IT Center of RWTH Aachen University and participants could attend both in person and via Zoom. In addition to speakers from the IT Center staff, there were also contributions from NVIDIA. The practical part of the aiXcelerate (aka “BYO Code”) was carried out in person.
The aiXcelerate 2024 was organized in collaboration with NHR4CES, HPC.NRW, EDIH Rheinland and WestAI. We would also like to thank NEC and NVIDIA for sponsoring the catering during the event.
Presentations
A single lecture is given each morning that aiXcelerate takes place. The first topic presented covered performance analyses of machine learning (ML) codes and includes the use of the automatic RWTH performance monitoring system, as well as the use of NVIDIA’s Nsight tool and how to find bottlenecks with it. The second day focused on “Scaling ML codes across multiple GPUs/nodes”. Here, approaches with PyTorch (Distributed) and Tensorflow + Horovod are presented to accelerate ML codes by using more hardware in parallel. The third day dealt with “Handling data sets of ML codes” and the various options for storing and using ML data at runtime (on CLAIX). In addition, the use of check pointing in ML codes was presented in order to save the state of the application at certain intervals and thus ensure error tolerance. An average of almost 40 participants learned new content for analyzing and optimizing their ML codes
Bring-Your-Own (BYO) Code
The traditional “BYO code” was once again an integral part of aiXcelerate. This means that HPC users can bring their own software codes and optimize them in the tuning sessions. In addition to the presentations in the morning, the tuning sessions take place in the afternoon. In these sessions, what has been learned can be applied directly with the support and guidance of the HPC experts present on site. The IT Center staff and an expert from NVIDIA were also involved. A total of seven code teams were supervised and made good progress during the workshop. One voice from the event evaluation said, “This was extremely helpful. It was an amazing opportunity to get direct feedback and suggestions from experts, that are very familiar with the cluster.”
Our HPC events will be announced in good time via our event pages. Come and see us next time!
The Next IT Center Workshop
Are you new to our cluster or would you like to refresh your knowledge from scratch?
No problem!
In March, we will once again be hosting the “Parallel Programming in Computational Engineering and Science” (PPCES) event.
This course is aimed at users who have little to no experience in HPC and would like to familiarize themselves with parallel programming in general.
The IT Center invites you to attend from March 10 – 14. We will keep you up to date on our social media channels!
The agenda for this year’s PPCES comprises the following three main topics:
Monday and Tuesday: MPI
Wednesday and Thursday: OpenMP
Friday: Machine & Deep Learning
Source: PPCES
Responsible for the content of this article are Marc Weerts and Masimba Koschke.
Leave a Reply