Search

The TREX project, in collaboration with the Austrian, Czech, and Slovak National Competence Centres for HPC, is hosting the "Code Tuning for the Exascale" workshop in Bratislava, Slovakia, from June 5th to 7th. This workshop is specifically designed for code developers and focuses on code optimisation. Participants will have the opportunity to engage in tutorials and training activities that emphasize performance, power consumption, and energy efficiency in HPC systems. The workshop also featured hands-on sessions showcasing the TREX CHAMP code and other valuable tools. The main objectives of the workshop are to enhance the practical experience in optimising HPC systems, help participants align their code implementation with HPC objectives, and provide valuable hands-on experience with practical QMC simulations using the CHAMP code. See the official event page for more details at TREX Events. &nbsp

We are excited to share that TREX participated in the prestigious 18th edition of the Teratec Forum, held on 31 May and 1 June 2023, at the Parc Floral in Paris. At the event, our team showcased our latest cutting-edge research and advancements in the Europa Village, an exclusive area dedicated to research projects. Europa Village serves as a vibrant hub of innovation, bringing together esteemed organisations such as Centres of Excellence, Competence Centres, research projects, infrastructure and service platforms, associations, and institutional bodies. Together, we are driving advancements and contributing to the success of economic and industrial players. We sincerely thank all those who visited the TREX booth at Stand #12 at the Teratec Forum. If you missed the chance to visit us, we invite you to explore our work and innovations by visiting the TREX website at https://trex-coe.eu/.

TREX is taking part in the EuroHPC Summit 2023 in Göteborg, Sweden,during the poster session on day 1 (20 March). We present here the software development strategy and the current achievements of the European Centre of Excellence “Targeting Real chemical accuracy at the EXascale” (TREX). The main objective of TREX is the development of a user-friendly and open-source software suite in the domain of stochastic electronic structure simulations, which integrates a set of flagship quantum Monte Carlo codes within an interoperable, high-performance platform. The core of our software efforts is the creation of two libraries, the TREXIO and the quantum Monte Carlo kernel library (QMCkl).

The poster was presented at the Psi-k 2022 conference, August 22-25. The poster presetns the software development strategy and the current achievements of the European Centre of Excellence “Targeting Real chemical accuracy at the EXascale” (TREX). The main objective of TREX is the development of a user-friendly and open-source software suite in the domain of stochastic electronic structure simulations, which integrates a set of flagship quantum Monte Carlo codes within an interoperable, high-performance platform. Core of our software efforts is the creation of two libraries, the TREXIO and the quantum Monte Carlo kernel library (QMCkl).

In this first document, after a brief Introduction (Section 1), we are describing the rationale for QMCkl design guidelines (Section 2). In particular, objectives and general methodology will be highlighted. Then, more precise details on the methodology will be given by walking through a detailed DGEMM example (Section 3). Finally, some preliminary performance results on a key QMCkl routine (JASTROW computation will be given). This first document essentially covers work in progress. As such, first, many implementation choices presented in this first document are still preliminary and might evolve. Second, we first focused on multicore CPU: although we kept in mind GPU constraints and issues, most of the implementation efforts targeted multicore CPU (X86 and ARM) and the first performance results have been obtained on X86. More performance tests on X86 are underway, and tests on ARM will be carried out in the coming months as well as the first QMCkl extensions to cover GPU. All of these topics will be covered in detail in the second deliverable. &nbsp

Dynamic voltage and frequency scaling proves to be an efficient way of reducing energy consumption of servers. Energy savings are typically achieved by setting a well-chosen frequency during some program phases. However, determining suitable program phases and their associated optimal frequencies is a complex problem. Moreover, hardware is constrained by non negligible frequency transition latencies. Thus, various heuristics were proposed to determine and apply frequencies, but evaluating their efficiency remains an issue. In this paper, we translate the energy minimization problem into a mixed integer program that specifically models most current hardware limitations. The problem solution then estimates the minimal energy consumption and the associated frequency schedule. The paper provides two different formulations and a discussion on the feasibility of each of them on realistic applications.

International audience; On modern multi-core processors, the growing gap between memory size, bandwidth and latency compared to computing capability makes the memory hierarchy predominant for performance. The Micro- kernel-Description-Language based Performance Evaluation Framework, MDL-PEF, accurately predicts optimized inner-loops performance de- pending on the loop's data access. The MDL-PEF approach revolves around a data flow description language, MDL. A static analysis step extracts the data flow structures of the assembly code. Then the pre- dictor uses pattern matching against an MDL-Microkernel database for predicting performance. Finally, MDL-PEF provides an automatic tool to initialize a pattern matching database for the target architecture. The overall system can predict the kernel performance on different plat- forms and optimizations, helping the user choose the best architecture for a given kernel. Preliminary experiments, with a 56 elements database, predict the innermost loop throughput of 636 binary loops of the NAS benchmarks with an average 10% of relative error. The performance pre- dictor is part of the Modular Assembly Quality Analyzer and Optimizer (MAQAO) performance tool framework. Future works will extend MDL- PEF to other architecture paradigms and more complex control flows such as outer loops.

This deliverable documents the initial performance analysis results obtained for all 6 TREX flagship applications. The focus was, in particular, the scaling of the application and the ability to exploit parallelism at all the different levels of modern HPC architectures. This ranges from the efficient use of SIMD instructions to the use of highly parallel compute accelerators like Graphics Processing Units (GPUs). For assessing the applications in terms of scalability, it needs to be taken into account that they differ in terms of their principle ability to be highly parallelized. Some of the applications, e.g. QMC=Chem, implement scalable methods with a strong focus on scalability, which could be demonstrated using up to 32,768 CPU cores. Furthermore, very encouraging results have been obtained for TurboRVB from GPU acceleration using Europe’s currently fastest supercomputer, i.e. JUWELS Booster. The performance results collected for this deliverable will help to guide further work and optimisations during the second half of the project.

The main goal of the TREX Centre of Excellence (CoE) is to enable the community codes for stochastic quantum chemical simulations to successfully exploit the most advanced European computing facilities, i.e., the EuroHPC pre-exascale and exascale machines, and foster the use of supercomputers with energy-efficient accelerators (e.g., GPU) by means of state-of-the-art quantum Monte Carlo (QMC) codes, developed in Europe and internationally recognised as key algorithmic assets. This will be achieved via a co-design approach to the development of the QMC community codes in order to prepare them for the next generations of HPC hardware. In line with this goal, this document sets out an effective communication, dissemination and stakeholder engagement strategy to be implemented by the consortium during the 36 months of the project. Consistent and content-rich communication is strategic to showcase the results and impact of the TREX project. This plan has been built on a continuous set of actions of communication and dissemination campaigns for the main TREX assets, primarily the TREX community codes for advanced QMC simulations, complex scalable workflow, and training and educational activities, ensuring coverage of all stakeholders and adequate visibility worldwide. The custom-made TREX web platform is at the heart of the engagement, outreach and exploration strategy of the project, it is the technological engine behind the technical support to end-users of TREX software, and hands-on workshops to forge a new generation of code developers. Various communication and promotional campaigns are presented in this document, with detailed timelines presenting the activities that will be put in place. For each of these campaigns, specific communication, dissemination, and stakeholder engagement activities and synergies within the EU HPC ecosystem are detailed to ensure the successful achievement of the project objectives &nbsp