EXTRA · Project

Self-Reconfiguring Supercomputer Hardware That Adapts to Your Workload in Real Time

digitalTestedTRL 4Thin data (2/5)

Imagine your computer could physically rewire itself depending on what task it's running — like a Swiss Army knife that automatically picks the right blade. That's what EXTRA built for supercomputers. Instead of fixed hardware that wastes energy on tasks it wasn't designed for, their chips reconfigure on the fly to become the ideal processor for whatever calculation is needed right now. The result: supercomputers that are both faster and use less power, because the hardware always matches the job.

By the numbers

consortium partners

countries represented

industry partners including SMEs

demonstration deliverables (initial + final integration)

The business problem

What needed solving

Supercomputers consume enormous amounts of energy running fixed hardware that can't adapt to changing workloads. When a calculation shifts from one phase to another, traditional processors keep running the same way — wasting power on circuits that aren't needed and lacking circuits that are. Companies running large-scale simulations, AI training, or real-time analytics pay the price in both electricity bills and processing delays.

The solution

What was built

The project delivered coarse-grain reconfigurable architectures with very low reconfiguration overhead, just-in-time synthesis tools for fast hardware re-compilation, and optimized HPC applications exploiting runtime reconfiguration. Concrete outputs include initial demonstrators and integration/final demonstrators proving the full chain from architecture to application.

Audience

Who needs this

HPC data center operators looking to cut power costs per computationFPGA and reconfigurable computing vendors seeking next-gen architecture IPCloud providers offering configurable HPC-as-a-serviceEDA tool companies building synthesis and compilation softwareResearch labs running large-scale simulations needing faster turnaround

Business applications

Who can put this to work

Cloud & HPC Data Centers

enterprise

Target: Large data center operators or cloud HPC providers

If you are a cloud HPC provider dealing with diverse customer workloads that each demand different hardware configurations — this project developed reconfigurable compute node architectures with very low reconfiguration overhead that can re-optimize themselves at runtime. Instead of buying separate specialized hardware for each workload type, a single reconfigurable node adapts, cutting idle silicon and energy waste. The consortium of 9 partners across 6 countries validated this through initial and final integration demonstrators.

Financial Services

mid-size

Target: Quantitative trading firms or risk analytics companies

If you are a financial analytics firm struggling with high-performance computing costs for real-time risk modeling and Monte Carlo simulations — this project built just-in-time synthesis tools that re-compile application phases into specialized hardware configurations on the fly. Your compute-heavy financial models could run on hardware that reshapes itself per calculation phase, reducing both processing time and power consumption across your HPC cluster.

Semiconductor & FPGA Design

any

Target: FPGA vendors or reconfigurable computing startups

If you are an FPGA or reconfigurable computing company looking for next-generation coarse-grain reconfigurable architectures — this project developed architectures that reconfigure at higher functionality levels, providing much faster reconfiguration than traditional bit-level approaches. The project also produced suggestions for improvements to reconfigurable technologies, offering a research-validated roadmap for your product development.

Frequently asked

Quick answers

What would it cost to license or adopt this technology?

The project was a publicly funded Research and Innovation Action (RIA) coordinated by Universiteit Gent. Licensing terms would need to be negotiated directly with the consortium partners. As a university-led project with 7 academic partners, technology transfer offices would typically handle licensing.

Can this work at industrial scale today?

The project produced initial demonstrators and integration/final demonstrators, proving the concept works in controlled settings. However, as the project focused on 'fundamental building blocks' for future exascale systems, scaling to production HPC environments would require further engineering and integration work.

Who owns the intellectual property?

IP is shared among the 9 consortium partners across 6 countries (BE, DE, EL, IT, NL, UK), including 2 industry partners. Specific licensing arrangements would depend on which partner developed which component. The 2 SMEs in the consortium may hold commercially relevant IP portions.

How does this compare to buying standard FPGA solutions?

Based on available project data, EXTRA's coarse-grain reconfigurable architectures provide much faster reconfiguration than bit-level approaches used in standard FPGAs. The just-in-time synthesis tools were specifically optimized for fast re-synthesis, addressing the key bottleneck of traditional reconfiguration speed.

Is this technology still relevant given advances since 2018?

The project ended in August 2018, but the core challenge it addressed — power-efficient heterogeneous computing for exascale — remains critical. The architectures and tools developed provide a research platform that subsequent European HPC initiatives have built upon. Reconfigurable computing is now more commercially relevant than when the project started.

What kind of technical support is available?

As a closed academic-led project, ongoing commercial support is not guaranteed. However, the coordinator Universiteit Gent and the 2 industry partners may offer consultancy or collaboration. The project website extrahpc.eu may have publications and technical documentation available.

Consortium

Who built it

The EXTRA consortium brings together 9 partners from 6 European countries (Belgium, Germany, Greece, Italy, Netherlands, UK), led by Universiteit Gent. With 7 universities and only 2 industry partners (22% industry ratio), this is a heavily research-oriented team — strong on academic depth but lighter on commercial pull-through. The 2 SMEs in the group could serve as commercialization bridges, but a business buyer should expect to invest in additional engineering to move from research demonstrators to production-grade systems.

UNIVERSITEIT GENTCoordinator · BE
THE CHANCELLOR MASTERS AND SCHOLARS OF THE UNIVERSITY OF CAMBRIDGEparticipant · UK
IMPERIAL COLLEGE OF SCIENCE TECHNOLOGY AND MEDICINEparticipant · UK
POLYTECHNEIO KRITISparticipant · EL
SYNELIXIS LYSEIS PLIROFORIKIS AUTOMATISMOU & TILEPIKOINONION ANONIMI ETAIRIAparticipant · EL
RUHR-UNIVERSITAET BOCHUMparticipant · DE
POLITECNICO DI MILANOparticipant · IT
MAXELER TECHNOLOGIES LIMITEDparticipant · UK
UNIVERSITEIT VAN AMSTERDAMparticipant · NL

How to reach the team

Universiteit Gent, Belgium — reach out through their technology transfer office or computer science department

Order a report on this consortium See UNIVERSITEIT GENT profile →

Next steps

Talk to the team behind this work.

Want an introduction to the EXTRA team to explore licensing their reconfigurable HPC technology? SciTransfer can arrange a direct connection with the right consortium partner for your use case.

Order a report on this topic More in Digital & ICT