Discoverer is Bulgaria’s First Petascale Supercomputer that is currently ranked 113th in the TOP 500 world ranking of most high-performing machines. It is managed by Consortium “Petascale Supercomputer - Bulgaria’’ with leading partner Sofia Tech Park.
SeeNews recently spoke with Petar Statev, chairman of the Supervisory Board of Sofia Tech Park and executive director of the Consortium “Petascale Supercomputer - Bulgaria”.
Mr. Statev, in what areas can the machine perform tasks and is there interest from business and scientific teams to use the machine’s capacity?
Discoverer has strongest capabilities, highest scientific and business interest and real demand in the areas of High Resolution Weather Forecasting and Environmental modelling, Computational Fluid Dynamics for flow and mass transfer simulations in complex geometries, Quantum Chemistry and Molecular Dynamic simulations in the areas of drug and new materials discovery, 3D reconstruction of satellites and drone images. These areas require a significant amount of high performance CPU calculations coupled with very fast network interconnect between the servers and very fast access to large high performance storage. We also invest time and resources in educating our current and future users, by providing free courses and materials on how to use our supercomputer. Petascale Supercomputer - Bulgaria participates in the EU Master Program in High Performance Computing (HPC) aimed to educate future HPC talent. This is critical for long term, sustainable utilisation and usage of Discoverer, because without a growing community of people who know how to operate such infrastructure it will be underutilised and its impact will be limited.
What projects is the machine currently working on and what tasks does it solve?
In all areas mentioned above we have running projects. Some of Discover’s current work are projects that create simulations of the climate changes of extreme meteorological phenomena and study the propagation and emerging properties of an incipient short gamma-ray burst jet piercing through the environment surrounding an accreting black hole formed after a binary neutron star merger. Most of the projects come from the Bulgarian Scientific Community and from access calls from EuroHPC JU/PRACE. These are consequences of the way Discoverer has been funded – 65% by the Bulgarian government and 35% from EuroHPC JU, the organisation that manages and co-funds large HPC Infrastructure in the EU. For purely scientific and non-commercial projects Discoverer’s resources are provided for free given that the applications pass the evaluation of international HPC experts. Projects are evaluated on their scientific merits, novelty and societal impact and all results need to be reported and published for public access. For a limited number of projects, which are of commercial interest, we are allowed to work with companies on a business to business basis, which means that access to Discoverer’s resource is paid, but in return companies are allowed to commercialise their results and there is no publication requirement. We are allowed to spend up to 20% of our computational resources per year on such commercial projects. Discoverer is part of the EU wide EuroHPC Supercomputer Network, which currently consists of 5 other petascale supercomputers and 2 exascale ones. In the future these supercomputers can run in federation where users can get access to all of the resources depending on availability and suitability of the projects.
Technologies are constantly evolving, how will Discoverer, with the parameters it currently has, maintain its competitiveness?
Currently, Discoverer has only a CPU partition, which though very powerful, is not suited for all types of applications, especially those in the areas of Machine Learning (ML) and Artificial Intelligence (AI). We are currently in the process of upgrading and in 2022 we plan to offer our customers access to accelerated GPU partition and to significant extension of our high performance storage. This will expand Discoverer’s capabilities in the AI/ML space where we see significantly growing interest from both local and international academic and business customers. Another area where we plan to expand our capabilities is Quantum Computing (QC), where we will start offering access to simulators/emulators of quantum computers and to real QC afterwards. The reason we do not jump immediately to the QC bandwagon is that we want to build our local community and train future users how to work with QC and then we will invest in the hardware. This emphasises our belief that there are direct benefits of offering Discoverer’s resources to local and international scientific and business communities and indirect benefits using the infrastructure to educate and train future talents. We believe that in the long run indirect benefits will outperform the immediate local ones.