This August, University of Tartu’s Rocket cluster will see the arrival of two new machines, each equipped with four high-VRAM H200 NVL 141GB PCIe 5.0 x16 GPUs. These powerful units are specifically designed for training large AI models, offering an impressive combined video memory of up to 564 GB within a single machine, seamlessly interconnected via a rapid internal NVLINK connection. For even more demanding computations, a 200gbps Infiniband connection between the machines allows for calculations across multiple units, providing enhanced scalability.

In addition to these machines for model training, the university has also expanded its cloud resources with eight NVIDIA L40S GPUs, each boasting 48GB of VRAM. These GPUs are tailored for the application of machine learning models (inference) and can be easily accessed by users, with the option to order up to two per virtual machine from the self-service portal minu.etais.ee.

“To be well prepared for the second half of 2025, we purchased these top-level GPUs at the University of Tartu HPC Center,” confirmed Ivar Koppel, ETAIS CEO and University of Tartu HPC Center`s Director

Developing our own language models for secure and sensitive data processing and teaching the Estonian language are essential to remain in international competition and not lose our digital capabilities.

The Estonian Scientific Computing Infrastructure (ETAIS) is working hand-in-hand with the Institute of the Estonian Language, which is spearheading the development of the Estonian language model with the active involvement of major universities. This collaborative ecosystem is exemplified by projects like baromeeter.ai, which invites public participation in identifying the most effective Estonian AI solutions.

At the heart of language model development lies the crucial need for high-performance computing resources, specifically Graphics Processing Units (GPUs) and Video Random Access Memory (VRAM). To meet the escalating demands of the latter half of 2025, the University of Tartu HPC Center has made a strategic acquisition of top-tier GPUs.