One of the most powerful AI platforms in the Central and Eastern European region, based on the NVIDIA DGX B200 server, has been launched in Lithuania. The platform, which cost 1 million euros and was installed at the VILNIUS TECH Digital Defense Competence Center, is designed to develop scientific research and innovations in the fields of AI, digital defense, and data science. University scientists say that this infrastructure will help them catch up and in some areas surpass international research centers, and it will also be available for use by private companies.
The new artificial intelligence (AI) computing platform opens up the opportunity for scientists to work tens or even hundreds of times faster than before and more quickly move their ideas from theory to real-world solutions.
"Due to the shorter time for experiments, modeling, and analysis of large data sets, it will be possible to develop prototypes and publish results faster. With this system in-house, we do not need to rely on external services, so we can ensure compliance with EU regulations on data sovereignty and the use of AI, as well as data privacy," says Vitalijus Gurčinas, director of the Digital Defense Competence Center established this year.
The platform will be available not only to university scientists, researchers, doctoral candidates, employees and students, but also to external partners – public institutions and businesses. “The computing and AI platform is intended to be used as a sandbox for cooperation between university and private sector entities, especially those involved in cybersecurity and defense projects. Therefore, we invite businesses to jointly develop solutions, use the testing environment and conduct joint research. Companies with valuable data and domain expertise, but without the infrastructure for AI experiments, are welcome,” says V. Gurčinas.
The latest DGX server model has arrived
The platform for the university was designed, delivered and implemented by the technology company “3RTechnology”, the only one in Lithuania with NVIDIA priority partner status. According to the company’s director Rėdas Šimelis, the project included not only the latest DGX server model currently available, air-cooled, but also other components: CPU and GPU servers, data storage, network switches, cooling and power supply systems. The entire equipment cost more than 1 million euros.
“We designed and prepared the infrastructure for work according to the needs of the university, so that the platform would be universal and could provide different types of resources according to the researcher’s tasks during the use of the platform. The NVIDIA DGX B200 server itself weighs more than 130 kg, and with packaging materials – over 180 kg, so one of the challenges was its transportation and safe delivery to the data center room,” says R. Šimelis. According to him, the times when the latest technologies reached our country only a couple of years after their launch on the market are already in the past. The implementation of this project shows that Lithuania is moving along with global technology trends.
Among the ambitious plans is a solution that has no analogues in the world
VILNIUS TECH Head of the Department of Electronic Systems, Professor Dr. Artūras Serackis, says that the platform provides opportunities that could only be dreamed of before. If previously it took a month to train one AI model, now scientists will be able to do the same work in half a day. In addition, it will be possible to work with models that were previously impossible to process.
"We can use this platform to research large language models and launch language models with many more parameters, which by specializing in one or another area of knowledge, we can achieve that the specialized language model will be almost infallible, answering questions in one area and providing a sufficiently detailed answer. In this way, we can provide businesses with the opportunity to practically verify whether their own large language model can be adapted to their needs, plan the costs of acquiring the necessary infrastructure, or refuse unmeasured investments in infrastructure in advance," shares A. Serackis.
The platform is already ready for use, and, according to the project manager, Prof. Dr. Dalius Matuzevičius, it is not easy to decide which research to start with, because the spectrum of ideas and projects is extremely wide. “For the first time, we have the opportunity not only to launch, but also to effectively train specialized AI models designed to generate 3D objects from individual photos. This is extremely important for our ongoing project – the development of a unique solution for three-dimensional modeling of eyeglass frames. We hope that the new infrastructure will soon allow us to achieve a significant technological breakthrough,” says the VILNIUS TECH professor.
