
News
Content for the HPC community and innovation enthusiasts: tutorials, news and press releases for users, engineers and administrators
- All News
- Aerospace & Defence
- Artificial Intelligence
- Blog
- Cloud Platform
- Cloud Platform
- E4 Various
- European Projects
- HPC
- Kubernetes Cluster
- Latest news
- Press
Towards an autonomously exascale Europe with EPI

In the future, it will be possible to choose low consumption processors made in Europe, without having to accept compromises on performance and use. Indeed, they will be suitable to support a large number of applications for science and engineering, in addition to AI ones.
By Marta Abba’, published March the 9th 2023, on the online magazine ZeroUno
Europe is preparing to “be exascale” and independent, thanks to the EuroHPC project. But not only. While we are witnessing its development, celebrating “our” LEONARDO supercomputer and counting down for JUPITER, Europe is also committing to the “surrounding” technology. Mediatically more hostile, but pragmatically fundamental. These efforts flow into the European Processor Initiative (EPI), now in its full second phase of implementation.
The project, within EuroHPC, aims to make the EU computing-independent, working on the future European processors, so that they will be high-performance and low-power. They would be a strategic advantage, thinking of European technological sovereignty, but also an environmental advantage, considering the dramatically better performance and energy efficiency that would be achieved.
With its 28 partners from 10 member countries, EPI actively works to support European supercomputing, focusing on processor technologies based on ISA and ARM architecture, to cover short-term needs, and on HPC acceleration technologies based on RISC-V, for long-term objectives. The funds used, originated from Horizon2020, through EuroHPC JU, result in an investment in areas in which many EU countries would like to have their say. In addition to the improvement of IT technologies and for the use of Big Data, competitive advantages are also expected for emerging applications such as autonomous-driving vehicles and industrial video analysis.
Long-term goals give concreteness, but they must be cultivated starting from the basic elements, making sure of the solidity and definitive nature of the intervention entrusted to EPI. In its first three-year phase (2018-2021), in fact, the planned activities concerned the creation of a general-purpose processor (GPP), and a POC of the EPI accelerator (EPAC).
To these two cross-sectoral elements, a specific application activity in the automotive field was then added. The first generation of GPP, codenamed Rhea, is the result of the integration of EPI technologies, achieved by maximizing performance, energy memory efficiency and bandwidth. It can already be configurable and customizable, and enables high-bandwidth data transfers between cores, accelerators, I/O, and shared memory resources. In this area there is the paw of the Italian E4 Computer Engineering, which created the Intermediate Bus Connector in support.
The same company also supplied the Daughter Board for the Test Chip of the EPI accelerator, an infrastructure an infrastructure with the task of ensuring energy-efficient acceleration. “This step was planned to make EPAC better serve a large number of applications. It allowed us to have it tested by the members of the consortium on their systems, in order to understand later which technologies to validate and which to improve” explains Cosimo Gianfreda, CEO of E4.
Perfection doesn’t always come right away, but the Test Chip and POC have shown that it is possible to both create a uniquely European design and use an open-source ISA free from proprietary licenses and export restrictions. An important certainty, more precious than ever because it was already obtained from the first phase. A good reason to look at the current application reference point such as the automotive one. The first phase of EPI, in fact, also saw the creation of high-performance embedded microcontrollers, capable of satisfying the growing demand for computing power in this sector in a safe and sustainable way. Tested in a homologated BMW X5, this platform includes integrated intelligent cameras and radar image analysis software but, above all, it is scalable and open to further technologies. A concrete element on which to fantasize one’s dreams on the cars of the future, autonomous at least up to level 4, thanks to EPI technologies.
Work in progress, industrialization projects in sight
The satisfactions and confirmations obtained in the first phase have not slowed down the second, which began in January 2022. It is the time to go beyond the initial developments, improving existing technologies, to give life to the new generation of structures for European HPC. It is also the time to set up industrialization and commercialization paths.
Ambitious goals, which make us imagine a robust and valid EPI for many European vertical segments and also capable of attracting non-EU interests. A necessary step to make this dream come true is the development of the second generation of GPP. The “aggregate” goal at this evolutionary step is an open common platform standard, aimed at efficiently interfacing processors and accelerators in the package, implementing cache coherence and validating the “toolchains” and execution times between processors and accelerators.
We also need a second generation of test chips for low-power accelerators, still to be tested, to ensure application concreteness. In its second phase, EPI also focuses on the edge computing area, with the processing of specific POCs related to autonomous shuttles or video surveillance, to optimize the HPC scalability path, adapting it to stringent and real-time constraints, typical of this market.
EPI members are working hard on those goals, with a focus on safety, energy efficiency and ease of implementation. These will be the months in which the best “scale-to-fit” of HPC processors and RISC-V accelerators, up to embedded solutions, will begin to emerge.
Towards the market, including Italian SMEs
There will still be challenges along this path: they must be transformed into an incentive to do better and better. The hardest, the human one, weighed heavily on the first phase: “it was difficult to find people able to carry on this project from the beginning. We have seen first-hand the need to create a new generation with adequate skills” explains Gianfreda, alluding to the technological challenge that led to EPI’s co-design approach.
“In the processor, we wanted to be sure there were adequate technological components, to allow it to better serve the different needs of a large number of codes. We have analyzed at least twenty of them, with a magnifying glass, to study their flow of instructions. Then we designed EPAC and GPP maximizing the performances of each. A demanding job that adds value to this European technology, even outside Europe” adds Gianfreda.
In addition to this “ability to speak many languages,” EPI has several competitive advantages over other processors. Speaking on versatility, one concern is the possibility of accelerating applications of various types, even of dense linear algebra. Most of its US or Chinese competitors, on the other hand, pay close attention to the AI world, also driven by the trends that regulate the global market.
By continuing to optimize performance and energy consumption, where room for improvement has been identified, between 2023 and 2024 we will arrive at EPAC 2.0 version, the production version. “It will be suitable for both engineering and more science and research related applications. System integrators will be able to include it into their products, contributing to its diffusion” explains Gianfreda, seeing the arrival of new processors and an EPI accelerator “that can also be deployed on the edge” as increasingly imminent. And that is what is needed for a more European future of applications related to the analysis of images collected by cameras. It is a fundamental technology in many contexts: from security checks in sensitive places, to those in industrial environments, up to the countless use cases that can be created in the field of smart cities and smart mobility. In this second phase, among other things, two Italian companies such as Leonardo and STMicroelectronics are taking care of the development of smart city intelligent edge applications. An important task, as important as their participation in the project, for those who believe in the stimulating effect. Italy counts on that, for EPI to also reach SMEs, over time. Gianfreda recalls that “they would have the advantage of having a technology with lower consumption and better performance, even compared to those offered to them by American and Chinese big tech companies. It is completely made in Europe”.