bg
Energy and housing and communal services
14:01, 19 November 2026
views
1

Powering Megascience: How a Supercomputer Data Center Is Protected Against Power Disruptions

Engineers have adapted a data center with a total computing capacity of 390 teraflops to operate alongside the SKIF (Siberian Circular Photon Source) shared-use center. They designed a dedicated power supply architecture for the facility.

The computing cluster at the SKIF shared-use center was designed by the RSC Group, one of Russia’s leading supercomputer developers and manufacturers. The data center houses 24 RSC Tornado servers with a combined performance of 182 teraflops, as well as two servers equipped with high-performance graphics accelerators, delivering 208 teraflops. The equipment uses a proprietary liquid cooling system. Compared with air cooling, it not only removes heat more effectively but also improves the facility’s energy efficiency, reduces noise levels and makes the equipment easier to operate.

Reliable power supply was a particular priority. Guaranteed power is essential to protecting the results generated by the entire synchrotron facility. Any disruption could result in the loss of valuable data, which megascience facilities generate continuously. That is why the data center is located inside the SKIF shared-use center’s electrical services building and is equipped with a 300 kW uninterruptible power supply system. The facility also has an environmental control system that maintains the temperature and humidity levels required for server and computing equipment. The synchrotron itself generates around 4 MW of heat during operation, while its total power consumption is 19,500 kVA.

Scaling Up Without Power Gaps: From Seven Beamlines to 30

The data center’s network infrastructure is already designed to receive experimental data from all 30 SKIF beamlines included in the project. Its existing equipment has sufficient capacity to process data from the first phase of the cluster, which is scheduled to come online in August 2027.

All 30 beamlines are expected to be operational by 2035. As new phases come online, the data center’s capacity will need to increase, raising its power demand and requiring additional cooling and uninterruptible power supply systems. Russian developers offer modular power management systems for data centers that allow equipment to be installed quickly and energized without shutting down the main bank of server racks. This is particularly important for a synchrotron that operates around the clock, seven days a week.

Lessons from Govorun: Energy Efficiency as an Industry Standard

For the RSC Group, SKIF is not its first experience designing data centers for megascience facilities. The company previously developed the Govorun supercomputer at the Joint Institute for Nuclear Research in Dubna. There, it faced similar challenges involving power supply reliability and energy efficiency. Govorun’s data center uses high-performance GPUs with direct-to-chip liquid cooling based on hot water. This approach eliminates the need for energy-intensive compressor chillers and simplifies the infrastructure by removing the need for refrigerant circuits.

In February 2026, SKIF commissioned its central engineering systems control center. It collects real-time signals from power supply, heating and cooling system sensors. A total of 47 control and server cabinets were installed across the complex. Engineers also developed dedicated control algorithms that automatically adjust equipment operation and prevent potential failures. As a result, the complex’s unified digital environment helps ensure reliable power supply.

Power: A Bottleneck for Scientific Data Centers

In the coming years, the growth of megascience computing infrastructure is likely to be constrained less by server performance than by the availability of electrical capacity for new connections and the cost of securing that capacity. This makes the SKIF project particularly relevant: its design incorporates energy-efficient liquid cooling, uninterruptible power supply systems and intelligent control of the engineering infrastructure from the outset.

For data centers running compute-intensive workloads, power supply reliability is not the only concern. They also need threefold redundancy to protect data. In other words, power must remain uninterrupted, which requires building out redundant infrastructure.
quote

like
heart
fun
wow
sad
angry
Latest news
Important
Recommended
previous
next