The Department of Computer-Aided Design Systems at Lviv Polytechnic held negotiations with Dell Technologies, Supermicro, Hewlett Packard Enterprise, ASUS, Lenovo, GIGABYTE and Cisco, consulted directly with representatives of NVIDIA — and selected a platform that did not even exist at the beginning of the selection process. The GIGABYTE XN24-VC0 server based on the GB200 Grace Blackwell architecture was presented only in January 2026: the appearance of this node changed the solution landscape and outperformed H100 and H200, which were considered favorites.
What's Inside
The foundation will be two NVIDIA GB200 Grace Blackwell superchips in a compact 2U chassis with direct liquid cooling. Together they provide 4 Blackwell GPUs and 2 NVIDIA Grace CPUs, 744 GB of ultra-fast HBM3E memory for graphics accelerators and another 960 GB of LPDDR5X for processors. NVIDIA ConnectX-8 network interfaces provide bandwidth up to 800 Gbps — this is critical for tasks where data doesn't fit in a single machine and needs to be distributed between nodes.
The Same Architecture — Different Scale
Comparison with what is being done on GB200 in other countries shows both the scale of the platform's capabilities and the scale of the difference in resources. Japan's RIKEN institute launched the ROQUO supercomputer with 135 such nodes — 540 Blackwell GPUs, NVIDIA Quantum-X800 network; in June 2026 it demonstrated 19.8 petaflops in the HPL test and ranked 106th in TOP500. Italian consortium CINECA is deploying the IT4LIA AI factory: 8192 Blackwell GPUs, over 1.5 PB of HBM3E memory, approximately 15 MW of power consumption.
Lviv Polytechnic will have one node. But the same type of node — which means compatible software stack, the same frameworks, the ability in the future to connect to federated computing networks.
What This Still Doesn't Solve
One GB200 node means the university is at the first-step level, not a player in the global HPC market. The real value of such infrastructure emerges when there are two things: research groups with specific tasks (molecular modeling, climate simulations, training large language models) and regular access for students and external partners — not just for the department's own projects.
The practical history of this computing center will begin with the first research tasks on NVIDIA GB200.
Lviv Polytechnic, official statement
For now, the university is awaiting delivery and preparing for installation and load testing. After launch, it will rank among the top Ukrainian universities in terms of power of its own local AI infrastructure — which in the current dependence on cloud computing and unstable internet in wartime conditions has quite practical, not just rating value.
The main question is not about hardware: will a public access policy for resources emerge by September 2026 — for which groups, by what criteria, will other universities be able to connect? Without this, the most modern node risks remaining a closed infrastructure of a single department.
