OPTIMA, a EuroHPC-JU funded project launched in 2021, has developed an FPGA-based chip which allows the optimisation and transfer of industrial applications on HPC systems. FPGAs are programmable computer chips that boost the performance of industrial applications, minimise energy consumption, and contribute to propelling Europe towards global supercomputing leadership.
The OPTIMA project is an initiative driven by a consortium of small and medium-sized enterprises (SMEs). It is the outcome of teamwork among 10 partners, with six of them being SMEs, and representing six different countries: Greece, Germany, Spain, Italy, Switzerland and the Netherlands. It had a budget of €4 100 000 from Horizon Europe, the EU’s funding programme for research and innovation, and completed its work in November 2023.
EuroHPC JU interviewed Iakovos Mavroidis, researcher at the Technical University of Crete (TUC) and OPTIMA’s project coordinator. Below, he explains the key features of the project and explains what it has accomplished thus far.
Can you please describe the OPTIMA project in your own words?
Sure! The primary objectives of the OPTIMA project was to optimise and test industrial applications and open-source libraries onto HPC systems using FPGA chips.
These special chip technologies known as Field-Programmable Gate Arrays (FPGAs) act as high-performance engines and accelerators for specific applications needed to run a supercomputer.
FPGA chip technologies are known for their lower power consumption compared to server-class CPUs and GPUs. They enable reduced energy consumption across various applications that are needed for supercomputing and could foster new, more economic and environmentally friendly approaches to HPC.
To accomplish this task, the OPTIMA project used JUMAX and Alveo-based high-performance servers, which were equipped with FPGA chip technology and were used to test different types of simulations in robotics, geosciences and computational fluid dynamics (CFD).
What was achieved?
A major achievement is the development of the OPTIMA Open Source (OOPS) library, which includes 31 hardware components to support fundamental linear algebraic operations and Computer-Aided Engineering (CAE) problem-solving methods also known as ‘solvers’.
The OOPS library not only enhances raw performance for scientific algorithms, but also promotes high energy efficiency. This is achieved through the optimisation of basic mathematical operations, known as BLAS Kernels, which are fundamental components in computational mathematics. These result in potential energy savings of up to 50 times per BLAS kernels.
In addition, in hardware prototype testing, OPTIMA has demonstrated remarkable results: it has doubled the speed of the Preconditioned Conjugate Gradient (PCG) algorithm, which is a widely-used method for solving complex mathematical equations. It is also 3.4 times faster than an average robotic simulation, 4.7 times faster than computational fluid dynamics (CFD) and seven times faster than a basic mathematical operation on regular HPC software.
Can you give some examples of how OPTIMA supports European HPC users and how it promotes greener and more sustainable supercomputing?
Certainly: OPTIMA directly supports European HPC users by making the OPTIMA Open Source (OOPS) library openly available to all users. This allows developers to transfer without restrictions applications and legacy code onto FPGA-supported HPC systems.
In this way, we encourage users to download the template project of the OOPS library and try any of the available kernels/components. The library is fully open and accessible to anyone on GitHub.
On sustainable supercomputing, I am proud to say that the OOPS library enables significant performance enhancements and reduced energy consumption which aligns perfectly with EuroHPC’s goals for energy-efficient supercomputing!
What were the main challenges you encountered during the project's development, if any?
During the development of use cases on the OPTIMA FPGA-based infrastructures, we encountered several challenges, the most notable of which are the following:
- FPGA-based MPSoCs offer flexibility, but managing resources efficiently poses a significant challenge. There are several hardware constraints related to the number of DDR memory controllers and distribution (Double Data Rate) and FPGA chip technologies. We had to carefully balance functionality between these two technologies to ensure optimal performance without exceeding hardware constraints.
- While there have been significant advancements in FPGA chip-related tools, there are still difficulties in developing efficient FPGA-based application. Providing an easy interface for the programmer is still a challenge and it is still easier to manually implement this design on an FPGA chip.
- In today’s FPGA-based infrastructures, communication between an FPGA-accelerated application and the host processor necessitates high-latency memory transfers via PCIe.
- As FPGA chip designs become increasingly complex, ensuring that all signals meet timing requirements becomes more challenging.
Overall, we were able to overcome most of these challenges through careful planning, rigorous testing, and collaboration among team members with expertise in FPGA design, hardware, and software development.
How is the development of such a platform supporting the ambition of the EuroHPC JU to make Europe a world-leader in supercomputing?
By advancing HPC capabilities through FPGA optimisation tools and chip technologies for industrial applications, OPTIMA contributes to Europe’s competitiveness in the global HPC landscape. In this way, the OPTIMA project aligns with EuroHPC JU’s ambition to position Europe as a global leader in HPC. Moreover, as OPTIMA is a good example of an innovative SME-driven EuroHPC project.
What’s next for OPTIMA?
The experience gained from porting applications to the OPTIMA heterogeneous platforms is being leveraged by the involved SMEs already today. New applications are being developed for advanced cloud systems, AI accelerators, GPUs, and other chip designs.
Additionally, the OPTIMA Open Source (OOPS) library will be continuously updated with additional support for FPGAs chips and tools, the implementation of new kernels, and further improvements to existing ones. OPTIMA anticipates that the OOPS library will become a valuable tool for software developers who want to map their applications onto FPGA-supported HPC platforms.
Furthermore, the advanced FPGA-based infrastructure developed by OPTIMA is currently being used for both reproduce complex hardware systems under design (SuDs) and the execution of high-end AI-powered applications.
Additional details:
For more information, please visit OPTIMA Project Website.
Contact: Iakovos Mavroidis / iakovosmavrogmail [dot] com (iakovosmavro[at]gmail[dot]com)
Details
- Publication date
- 24 April 2024
- Author
- European High-Performance Computing Joint Undertaking