Didem Unat is the project coordinator of a newly awarded project called SparCity, an optimization and co-design framework for sparse computation. Didem Unat works as a faculty member at the Koç University located in Istanbul, Turkey. Prior to joining Koç University, Didem Unat was a researcher at Lawrence Berkeley National Laboratory and highly involved in exascale co-design projects. Her research area includes programming environments, runtime systems, compilers, libraries for scientific and industrial applications. Her academic university group is involved in developing system software, programme modelling and prediction tools. In addition, they perform application parallelisation and optimisation. Koç University is a small but very well known research university in Turkey, specialized in logic programming. Koç manages the highest number of projects of all universities in Turkey for Horizon 2020. We invited Didem Unat in our virtual 4m below sealevel studio in downtown Almere to talk about SparCity.
In which part of Turkey is Koç University located?
Didem Unat: It is located in the European part of Istanbul in the north by the Black Sea, a region that is worth visiting.
Sounds nice. The project we are talking about is called SparCity. It has been awarded as part of a EuroHPC JU call. Is that correct?
Didem Unat: Yes. The topic of the project perfectly aligns with the vision of EuroHPC. The SparCity project aims at creating a supercomputing environment. It will provide efficient algorithms, coherent performance and programming tools, specifically designed for sparse computations. When talking about sparse computations, people think of scientific applications, finite element methods, mesh computation, etc. However, this project will also open up new usage areas for sparse computation, including data analytics and deep learning. That is why it is a very impactful project.
The goal is to perform node-level and system-level software optimizations for sparse operations. An interesting part of the project is that the team is looking at developing a digital SuperTwin in which will be elaborated what-if scenarios of emerging architectures in a co-design manner. The project will have a large societal impact because the team will demonstrate the usability of the framework for important real life applications coming from different realistic domains. These domains include computational cardiology, social networks, bio-informatics and autonomous driving.
These are the four application areas. Basically, the project is about solving sparse matrices in an innovative way. How does this relate to exascale machines or new architectures? You mentioned co-design?
Didem Unat: Sparse computation is not only about sparse matrices. The project also addresses graphs for irregular accesses and irregular computations. These are a lot more challenging than dense matrices in terms of optimisation. All the partners have a large experience in optimizing sparse computations, so this is a perfect fit for them and for the consortium.
How big is the consortium and who are the partners?
Didem Unat: The partners include Koç University which is the coordinator and another research university in Turkey, namely Sabanci University. The third partner is Simula, a research laboratory from Norway. The other partners are INESC-ID from Portugal, LMU Munich from Germany; and Graphcore, an AI chip company from Norway. All these partners already collaborated prior to this project proposal. They have been working together in pilot projects and have been publishing papers together. So, it was a natural collaboration and it was not that difficult to compose the SparCity proposal which received a topscore of 15 out of 15 from the HPC reviewers.
That is very good, 15 out of 15. Only very few have managed that. Congratulations with this result. What is the budget for the project and what is the duration?
Didem Unat: The budget is 2,6 million euro which will be split between the six partners. The project will have a duration of 3 years and will kick off January 1, 2021.
Will the results only be ready after three years or will there be intermediate results?
Didem Unat: Of course, we will have intermediate deliverables, every year. In the first year, we will have some deliverables and six months after that, we will have project results available online. We plan to make all our project results, data, source code, and publications open to the public. They will be available on the project website.
Do you collaborate with other EuroHPC projects, other projects or computing centres in that area?
Didem Unat: We have plans. When they will be officially announced, we will know the other projects. We will of course keep in touch with those. This year, there is another EuroHPC call for pilots for computing infrastructure. We have started conversations with those pilots. They are still under preparation so it is very early to say something concrete.
Thank you very much for this interview.