Adaptyst performance analysis and full-stack system design workshop

Europe/Zurich
31/3-004 - IT Amphitheatre (CERN)

31/3-004 - IT Amphitheatre

CERN

105
Show room on map
Maksymilian Graczyk (CERN), Stefan Roiser (CERN)
Description

The workshop has already taken place, but you can still:

  • browse the Adaptyst introductory and demo/discussion slides and watch the corresponding recordings without any registration (see "Workshop recordings on CERNBox" in the timetable for the public videos)
  • browse all slides and watch all recordings with the free postponed registration option

If you want to stay up to date with Adaptyst and its events, consider joining our Mattermost team and signing up to the adaptyst-tool mailing list on GMS/e-groups.

Adaptyst is an open-source R&D project developed at CERN about scalable, architecture-agnostic, and comprehensive performance analysis with optional flexible full-stack (software + hardware) system design and elements of compilation for various workflows across the entire computing spectrum, from embedded to distributed/high-performance computing.

Performance is broadly understood here: it can be latency, throughput, energy efficiency etc. with any contextual information, e.g. budgetary constraints.

In the current stage, the tool concentrates on profiling. However, the ultimate goal of the project is suggesting the most optimal compute solution automatically to an end user given their workflow, constraints, and requirements while taking into account all sides of computation: software, hardware, and systems.

Therefore, in the later stages of R&D, we are aiming for Adaptyst to do automatically various activities related to designing software and hardware systems, e.g. analyse performance of a workflow on a given setup, pick the most optimal software and hardware parameters where possible, choose compute units amongst CPUs, GPUs, FPGAs, and custom accelerators, design a memory hierarchy, or plan out storage and networking.

The planned modular design of the tool on both the software and hardware sides (with the first version on the hardware side becoming available soon) allows Adaptyst to use existing work such as profilers easily and adapt to all types of workflows and computing systems regardless of the pace of market development.

This workshop is dedicated to Adaptyst and is a unique chance for you to shape the project:

  • You will be introduced to the tool.
  • You will hear several talks about Adaptyst-related work and current/potential applications of the tool.
  • You will be able to discuss your use case(s) and share your opinions, helping guide R&D directions of the project.
  • You will participate in the Adaptyst live demo session.

Sounds interesting? Register now for free! Tea and coffee will be provided. 

NEW: If attending all day is not feasible for you, the partial/postponed participation option is now available for the morning introductory session at 9:30-10:20 CERN time, the afternoon demo + discussion session at 15:30-17:00 CERN time, and/or the recordings (along with an optional one-to-one meeting to discuss your use case for Adaptyst on a date most convenient for you)! Live webcast is also available for everybody with a CERN account without having to register (no possibility of asking questions and contributing to discussions).

The workshop is open to people from all CERN departments and experiments (regardless of the computing / electronics for data processing or controls work they do) + external people by invitation, with a remote Zoom option for everybody either not based at CERN at the time of the event or taking part in the event only partially.

If you have any questions, please message adaptyst-contact [at] cern.ch (the previous e-mail had a dot instead of a dash and was thus incorrect: apologies!) or ask us in our Mattermost team.

Registration
Workshop registration (partial/postponed participation)
Participants
Webcast
There is a live webcast for this event
Zoom Meeting ID
62624082714
Host
Maksymilian Graczyk
Useful links
Join via phone
Zoom URL
    • 1
      Welcome and introduction to Adaptyst: a scalable and architecture-agnostic performance analysis tool 31/3-004 - IT Amphitheatre

      31/3-004 - IT Amphitheatre

      CERN

      105
      Show room on map
      Speaker: Maksymilian Graczyk (CERN)
    • 2
      Workshop recordings on CERNBox 31/3-004 - IT Amphitheatre

      31/3-004 - IT Amphitheatre

      CERN

      105
      Show room on map
    • 10:20
      Coffee break 31/3-009 - IT Amphitheatre Coffee Area

      31/3-009 - IT Amphitheatre Coffee Area

      CERN

      30
      Show room on map
    • 3
      Adaptive Heterogeneous Computing in Space: Opportunities and Challenges 31/3-004 - IT Amphitheatre

      31/3-004 - IT Amphitheatre

      CERN

      105
      Show room on map

      The talk will be in person and is fully confirmed to take place at the specified time.

      Abstract:
      As space missions grow increasingly complex, traditional homogeneous computing architectures struggle to meet the diverse and evolving demands of on-board data processing. Adaptive heterogeneous computing offers a promising solution by combining different processing elements—such as CPUs, GPUs, FPGAs, and AI accelerators—into flexible, reconfigurable systems. This talk explores how adaptive heterogeneous platforms can enhance autonomy, resilience, and performance in space applications, from real-time scientific data analysis to fault-tolerant spacecraft control. In this scenario, efficient and reliable profiling will become a critical aspects of the system. We will discuss the key opportunities this paradigm presents, including improved energy efficiency and mission adaptability, alongside the challenges limited power budgets, software complexity, and radiation tolerance. Finally, we will highlight emerging technologies and research directions shaping the future of space computing.

      Speaker: Prof. Andrea Guerrieri

      Andrea is a professor and researcher at the University of Applied Sciences Western Switzerland and EPFL. He is the head of the Adaptive Heterogeneous Systems Lab, specialized on adaptive and reconfigurable computing, electronic design automation for both terrestrial and space applications.

      He authored several dozens of papers presented at prestigious international venues such as the International Symposium of Field-Programmable Gate Arrays (ISFPGA), the Design Automation Conference (DAC), Space Computing Conference (SCC), and Quality Electronics Design (ISQED). He is also the author of a book on System-on-Chip Design with Arm, and editor of a book on Applications enabled by FPGA-based technology.

      Andrea’s work has earned him several prestigious awards, including Best Paper Awards at FPL 2024, HPEC 2024, and ISFPGA 2020, held in Seaside, California. He has also received multiple Best Paper nominations at FCCM 2022 in New York, FPL 2022 in Edinburgh, and the European HiPEAC 2022 award. In 2021, he was recognized as a Senior Member of IEEE. Andrea actively serves as a Technical Program Committee (TPC) member, artifact evaluator, and session chair for key conferences such as DAC, DATE, ISFPGA, ICCD, FCCM, FPL, and ISQED. He was honored with the Outstanding TPC Member Award at DAC 2024 and DAC 2025 in San Francisco, California, USA.

      He participates in and leads multiple international projects, and he is currently collaborating with partners across Switzerland and Silicon Valley (USA) to shape the next generation of EDA tools and reconfigurable computing platforms for both terrestrial and space applications. His key collaborators include industry giants like AMD-Xilinx, NVIDIA, Arm, NASA, and CERN, but also academic institutions such as ETH Zurich, University of Geneva, and California State University. He is chair of the Onboard Computing topic for the Swiss consortium CHEESE affiliated with NASA SSERVI (Solar System Exploration Research Virtual Institute), located at AMES Research Center, Mountain View, California, USA.

      Speaker: Prof. Andrea Guerrieri (HES-SO / EPFL)
    • 4
      Cache-aware roofline modelling for CPUs and GPUs 31/3-004 - IT Amphitheatre

      31/3-004 - IT Amphitheatre

      CERN

      105
      Show room on map

      The talk will be given remotely and is fully confirmed to take place at the specified time.

      Abstract:
      In recent years, HPC systems have become increasingly complex and heterogeneous, including CPUs and GPUs, making application development and optimization challenging. Performance models like the Cache-aware Roofline Model (CARM) offer effective guidance by providing insights into bottlenecks that limit the application’s ability to reach the system’s maximum performance. The focus of this talk is the CARM Tool, which was developed to provide CARM support, not only to all major CPU architectures and ISAs, i.e., x86 (Intel, AMD), ARM, and RISC-V, but also to GPUs (NVIDIA). The CARM tool also covers specialized accelerators for artificial intelligence, such as the NVIDIA Tensor Cores, for the performance and efficiency characterization of deep learning workloads. The proposed tool includes automatically generated microbenchmarks, specifically tailored to cover both computational units and all memory hierarchy levels. Additionally, in this talk, I will showcase the deployment of CARM in the EuroHPC centers and present a short tutorial on the tool usage.

      Speaker: Prof. Aleksandar Ilic

      Aleksandar Ilic (PhD’14) is an Associate Professor at the Electrical and Computer Engineering Department (DEEC) of Instituto Superior Técnico (IST), Universidade de Lisboa, and a Senior Researcher of the INESC-ID, Lisbon, Portugal. His research interests include high-performance and energy-efficient computing and modeling of parallel heterogeneous systems. He has contributed to more than 90 international journal and conference publications.

      Speaker: Prof. Aleksandar Ilic (INESC-ID)
    • 5
      Time margin for Q&A 31/3-004 - IT Amphitheatre

      31/3-004 - IT Amphitheatre

      CERN

      105
      Show room on map
    • 12:00
      Lunch
    • 6
      Performance challenges and analysis at LHC experiments (online and offline computing) 31/3-004 - IT Amphitheatre

      31/3-004 - IT Amphitheatre

      CERN

      105
      Show room on map

      Speakers:

      • Anton Alkin (ALICE): Computing in ALICE
      • Sebastien Ponce (LHCb): Computing in LHCb
      • Sioni Summers (CMS): Online computing in CMS
      Speakers: Anton Alkin (University of Sofia - St. Kliment Ohridski (BG)), Sebastien Ponce (CERN), Sioni Paris Summers (CERN)
    • 7
      Performance challenges and analysis at CERN outside of LHC experiments 31/3-004 - IT Amphitheatre

      31/3-004 - IT Amphitheatre

      CERN

      105
      Show room on map

      Speakers:

      • Giovanni Iadarola (BE-ABP-CAP): Beam dynamics simulations and performance challenge
      • Daniele Massaro (IT-FTI-PSE): Physics event generators
      • Lorenzo Rossi (SY-EPC-CCS): Embedded/Real-time/Bare-metal computing
      Speakers: Giovanni Iadarola (CERN), Daniele Massaro (CERN), Lorenzo Rossi (Politecnico di Milano (IT))
    • 8
      Time margin for Q&A 31/3-004 - IT Amphitheatre

      31/3-004 - IT Amphitheatre

      CERN

      105
      Show room on map
    • 15:15
      Coffee break 31/3-009 - IT Amphitheatre Coffee Area

      31/3-009 - IT Amphitheatre Coffee Area

      CERN

      30
      Show room on map
    • 9
      Adaptyst live demo, discussion about Adaptyst applications, gathering community inputs on Adaptyst R&D 31/3-004 - IT Amphitheatre

      31/3-004 - IT Amphitheatre

      CERN

      105
      Show room on map
      Speaker: Maksymilian Graczyk (CERN)