MANGO · Project

Energy-Efficient Supercomputing Hardware That Guarantees Real-Time Performance for Demanding Applications

digitalTestedTRL 5Thin data (2/5)

Imagine you need a supercomputer that can process medical images or satellite data instantly — but today's systems either burn too much electricity or can't guarantee they'll finish on time. MANGO built a new kind of computing hardware that mixes different types of processors on one board, like having specialists and generalists working side by side. They also designed a passive cooling system (no fans, no pumps) to cut energy costs. The goal was to hit 100 PFLOPS of computing power while staying under 15 megawatts — roughly the difference between powering a small town and powering a neighbourhood.

By the numbers

100 PFLOPS

Target computing performance

<15 MWatt

Target power consumption ceiling

EUR 5,801,820

EU funding invested

Consortium partners

Countries represented

Total project deliverables

Physical platform phases built

The business problem

What needed solving

Today's high-performance computing systems waste enormous energy and cannot guarantee that time-critical tasks — like processing medical images or satellite data — finish on schedule. Businesses either overspend on hardware to build in safety margins, or accept unpredictable delays. There is no mainstream way to get both extreme computing power and real-time guarantees without burning through megawatts of electricity.

The solution

What was built

The project built three successive physical computing platforms: starting with emulation boards, progressing to custom motherboards housed in server racks with dedicated power distribution, and culminating in a final integrated manycore system with heterogeneous acceleration nodes and passive cooling. All target applications (diagnostic imaging, satellite processing, telecommunications) were validated on the final platform.

Audience

Who needs this

Data centre operators looking to reduce power costs while offering guaranteed SLAsMedical imaging companies needing real-time scan processing with predictable latencySatellite data processing firms handling large-scale earth observation workloadsTelecommunications infrastructure providers requiring QoS-guaranteed computingHPC hardware vendors seeking next-generation architecture designs for energy-efficient systems

Business applications

Who can put this to work

Data centre operations

enterprise

Target: Data centre operators and cloud infrastructure providers

If you are a data centre operator dealing with spiralling electricity bills and customers demanding guaranteed response times — this project developed heterogeneous computing nodes with native quality-of-service isolation and a two-phase passive cooling system. The architecture targets up to 100 PFLOPS at under 15 megawatts, which could significantly reduce your power-per-computation ratio while letting you promise real-time performance to tenants.

Medical imaging and diagnostics

mid-size

Target: Medical device and imaging companies

If you are a medical imaging company struggling with slow processing of CT or MRI scans — this project built and validated a manycore computing platform designed for QoS-sensitive applications like diagnostic imaging. The hardware ensures predictable bandwidth and latency, meaning your image reconstruction pipelines can meet strict clinical deadlines without over-provisioning expensive servers.

Satellite and geospatial analytics

mid-size

Target: Earth observation and satellite data processing firms

If you are a satellite data company that needs to process terabytes of imagery in near real-time — this project created reconfigurable acceleration nodes linked by a homogeneous interconnect that dynamically adapts to different workloads. The platform was validated for satellite technology applications across 10 partners in 7 countries, providing a blueprint for scaling up without proportional power increases.

Frequently asked

Quick answers

What would it cost to adopt this technology?

The project received EUR 5,801,820 in EU funding across 10 partners to develop the platform. Since this is research-stage hardware, commercial pricing does not exist yet. Any adoption would likely require co-development with the consortium partners or licensing of specific architectural designs.

Can this scale to production data centre environments?

The project explicitly targeted large-scale capacity computing scenarios and built three successive platform phases, culminating in a rack-level system with custom motherboards, power distribution, and cooling. The architecture was designed for 100 PFLOPS at under 15 megawatts, indicating data-centre-scale ambitions, though commercial manufacturing has not been demonstrated.

What about intellectual property and licensing?

IP was generated across 10 partners including 5 industry players from 7 countries. Licensing terms would need to be negotiated with the coordinator (Universitat Politecnica de Valencia) and relevant consortium members. The project produced 24 deliverables covering hardware designs, software tools, and thermal management systems.

Is the hardware actually built or just simulated?

Real hardware was built. The project delivered three physical platform phases: emulation boards for architecture exploration, custom motherboards with heterogeneous acceleration nodes in server racks, and a final integrated platform. A final validation deliverable confirmed all applications ran on the targeted real-time HPC system.

How does the cooling system work and what does it save?

MANGO developed a two-phase passive cooling system covering chip, board, and rack levels — meaning no fans or pumps are needed. Based on available project data, this was designed to help stay within the 15 megawatt target for 100 PFLOPS performance, but specific energy savings percentages were not published in the available materials.

What applications were actually tested on this platform?

The project validated applications in diagnostic imaging, satellite technology, and telecommunications — all requiring guaranteed quality-of-service and high performance. The final validation deliverable confirms all target applications ran successfully on the completed platform.

Consortium

Who built it

MANGO brought together 10 partners from 7 countries (Switzerland, Germany, Spain, France, Croatia, Italy, Netherlands), with a balanced mix of 5 industry players and 4 universities plus 1 research organisation. The 50% industry ratio is strong for a research project, meaning the technology was shaped by commercial reality from the start. The consortium includes 1 SME. The coordinator is Universitat Politecnica de Valencia in Spain, a major technical university with deep expertise in computer architecture. For a business looking to adopt or co-develop this technology, the presence of 5 industry partners means there are established commercial entities already familiar with the IP and capable of technology transfer.

UNIVERSITAT POLITECNICA DE VALENCIACoordinator · ES
THALES SIX GTS FRANCE SASparticipant · FR
CENTRO REGIONALE INFORMATION E COMMUNICATION TECHNOLOGY SCARLparticipant · IT
SVEUCILISTE U ZAGREBU FAKULTET ELEKTROTEHNIKE I RACUNARSTVAparticipant · HR
ECOLE POLYTECHNIQUE FEDERALE DE LAUSANNEparticipant · CH
PRO DESIGN Electronic GmbHparticipant · DE
POLITECNICO DI MILANOparticipant · IT
EATON INDUSTRIES (FRANCE) SASparticipant · FR
PHILIPS MEDICAL SYSTEMS NEDERLAND BVparticipant · NL

How to reach the team

Universitat Politecnica de Valencia (Spain) — reach out to the computer architecture department for licensing and collaboration enquiries.

Order a report on this consortium See UNIVERSITAT POLITECNICA DE VALENCIA profile →

Next steps

Talk to the team behind this work.

Want an introduction to the MANGO team? SciTransfer can connect you with the right technical contact and help you evaluate whether this architecture fits your computing infrastructure needs.

Order a report on this topic More in Digital & ICT