Valentin Plugaru, CTO at Luxprovide. Ilustration : Maison Moderne & Crédit photo : LuxpPovide

Valentin Plugaru, CTO at Luxprovide. Ilustration : Maison Moderne & Crédit photo : LuxpPovide

As one of the architects of the MeluXina supercomputer, LuxProvide Chief Technology Officer Valentin Plugaru is ideally placed to explain the platform’s evolution and positioning to answer new requirements for AI-driven solutions.

Valentin Plugaru has spent his career in implementing, growing and managing High Performance Computing platforms. Prior to joining LuxProvide in April 2020, he helped build the HPC facility of the University of Luxembourg as an instrument for scientific excellence over the better part of a decade. As Luxembourg embarked on the journey to empower the digital ecosystem with a new supercomputing platform to boost innovation and competitivity, he accompanied the launch of the MeluXina initiative. “Being selected to become LuxProvide’s Chief Technology Officer and lead the greenfield implementation of a new supercomputing platform was a tremendous opportunity,” Plugaru says.

As one of LuxProvide’s first hires, Plugaru gathered the team that set up MeluXina, attracting top talent from Luxembourg and abroad. He says it’s easy to be enthusiastic when working with supercomputing experts. “It’s a field that is always evolving, sometimes in unexpected ways. And the way supercomputers are used changes fast too. This keeps the ecosystem very fresh and is always challenging us.”

In a supercomputer all aspects of performance are maximized, from the type of processors, the accelerators, to the data systems and the specialized networks that link all components.

Valentin PlugaruCTOLuxProvide

Heterogeneous systems

Creating LuxProvide and bringing MeluXina to life derived from the need to help the national ecosystem in its digital transition. “The focus, when LuxProvide was created, was not only to implement supercomputing platforms but to provide expertise, solutions and guide users coming from a wide spectrum of activity sectors.”

Plugaru explains that supercomputers have started to be much more data oriented over the last years. The injection of high-end data analytics capabilities supporting machine learning, deep learning and artificial intelligence was key. “Supercomputers in the past were created as more monolithic systems. These days we're seeing many more heterogeneous systems being built, which have capabilities for a larger variety of workloads.” This also comes from the convergence of intensive simulation with requirements for data analytics and AI, for which the global appetite is growing extremely fast. “MeluXina was designed with vast capabilities for digital twins and AI, and it can accelerate many of the new kinds of use cases that we are seeing,” the CTO explains.

Plugaru compares supercomputers to Formula 1 cars. “All aspects of performance are maximised, from the type of processors, the accelerators, to the data systems and the specialised networks that link all components.” These specialised networks are a key differentiator in the computing world, says Plugaru. They make it possible, for example, to run virtual testing of prototypes at large scale and with unparalleled precision. In addition, they use specialised data platforms that are extremely well suited for large scale analytics that can support decision making. “On MeluXina we’ve implemented a multi-layer software ecosystem that includes development tools needed to build and optimise applications, frameworks for AI allowing training and execution of models – for example large language models and generative AI which are very much in the spotlight these days – and many other software packages that can be rapid enablers for industry to take advantage of the system.”

End-to-end user solutions

Users have access to a variety of specialised tools that can help different segments, from computer aided engineering to applications for life sciences and precision medicine, that can be enhanced with AI frameworks. LuxProvide also implemented on MeluXina different quantum simulators to prepare for the future. “You can define quantum circuits and see how they will behave on current ‘classical’ platforms. This is for now an advanced prototyping that requires significant computing power even for systems with limited amount of qbits [quantum bits], but this will help to better understand the world and to drive future developments.”

The key element over the next years will really be user feedback, which will drive how our platforms evolve.

Valentin PlugaruCTOLuxprovide

LuxProvide’s strategy is to offer more comprehensive solutions and integrated services. “This starts from understanding end user requirements, analysing how supercomputing plays into the panoply of these needs and matching them with end-to-end solutions. Such that it’s not only one part of a workload that is being accelerated, but the complete pipeline, and this requires more flexibility from our platforms than ever before.”

The beauty of MeluXina, according to Plugaru, is that it can provide a launching pad for many new projects and use cases. “We are preparing and adapting for the needs of the future,” Plugaru explains. Requirements come from strategic sectors - finance, life sciences, space as well as focus areas such as sustainability.

“Boosting the ecosystem also means providing expertise to first-time supercomputer users. Such platforms are specialised and require initial guidance, but with the help of our experts new projects are onboarded very fast. The key element over the next few years will really be user feedback, which will drive how our platforms evolve.”

Looking into the future, Plugaru says that new software solutions, as well as new chips designed to meet the intensive computing needs in the market will be transformative. “The rise of generative AI will drive new requirements and capabilities. Today we see AI models becoming efficient at some tasks such as answering questions. The future will see systems of models working together.”