Most real-world data is like a mostly-empty spreadsheet — millions of cells but only a few actually have values in them. Crunching through all those empty cells wastes enormous amounts of computing power and electricity. SparCity built specialised software tools that teach supercomputers to skip the blanks and focus only on what matters, cutting both runtime and energy bills. They proved it works on four tough real-world problems: heart simulation, social network analysis, DNA sequencing, and self-driving cars.
Companies running data-heavy computations — from medical simulations to autonomous driving algorithms — waste massive amounts of processing power and energy on sparse data where most values are zero. Current software treats every data point equally, meaning supercomputers burn through electricity crunching empty cells. This drives up cloud computing bills and slows down time-to-result for critical applications.
The project built 3 key prototypes: a source-to-source compiler that auto-tunes computation kernels for specific hardware (GPUs, IPUs), performance and energy prediction models, and a dynamic topology information system for complex HPC setups. They also created a digital SuperTwin for evaluating hardware configurations. All tools were demonstrated in 2 public workshops with tutorials, totalling 19 deliverables.
If you are an autonomous driving company dealing with massive sensor data that needs real-time processing — this project developed optimised sparse computation tools that improve performance and energy efficiency on GPU and IPU hardware accelerators. The project demonstrated effectiveness on autonomous driving as one of 4 real-life application domains across 19 deliverables.
If you are a medical technology company running computational cardiology simulations that take too long or cost too much in cloud compute — this project built a kernel template generator and performance models specifically tuned for sparse computations. They validated their tools on computational cardiology as one of 4 challenging application domains.
If you are an HPC service provider struggling with energy costs and underperforming sparse workloads — this project created a digital SuperTwin that evaluates hardware configurations and tests what-if scenarios before you buy. The prototype includes a dynamic topology information system covering complex multi-device setups with GPUs and IPUs.
The project was publicly funded as a Research and Innovation Action under EuroHPC. The software tools and prototypes were developed as open research outputs. Licensing terms would need to be discussed directly with the consortium led by Koç University.
The tools were designed for exascale computing systems and were tested on 4 demanding real-life application domains: computational cardiology, social networks, bioinformatics, and autonomous driving. However, as research prototypes, they may require integration work before production deployment.
The project was funded under EuroHPC RIA, meaning IP typically stays with the consortium of 6 partners across 4 countries. Licensing arrangements would need to be negotiated with the coordinator, Koç University in Turkey, and potentially the 1 industry partner in the consortium.
The tools target modern HPC hardware including GPUs and IPUs (Intelligence Processing Units). The dynamic topology information system integrates with hwloc and vendor-specific tools to handle complex multi-device computing setups.
The project delivered working prototypes including a kernel template generator, performance and energy models, and a dynamic topology information system. Two workshops were held to demonstrate the tools. However, these remain research prototypes from a 3-year project that closed in March 2024.
The consortium held 2 workshops demonstrating the tools and providing tutorials on how to use them. Ongoing support would depend on individual arrangements with the 6 consortium partners, which include 3 universities and 2 research organisations.
The consortium of 6 partners across 4 countries (Germany, Norway, Portugal, Turkey) is research-heavy: 3 universities, 2 research organisations, and just 1 industry partner (17% industry ratio). There are zero SMEs. This signals a strong academic foundation but limited commercial pull — a business adopter would likely need to invest in bridging the gap between these research prototypes and production-grade tools. The coordinator is Koç University in Turkey, a respected research institution. The involvement of partners from Germany and Norway suggests alignment with major European HPC infrastructure investments.
Koç University, Turkey — reach out to the Computer Science or HPC department
Want to explore how SparCity's sparse computing tools could reduce your HPC costs? Contact SciTransfer for an introduction to the research team.
